WO2019153759A1 - 确定运输方案的方法、训练快速装载模型的方法及设备 - Google Patents

确定运输方案的方法、训练快速装载模型的方法及设备 Download PDF

Info

Publication number
WO2019153759A1
WO2019153759A1 PCT/CN2018/108534 CN2018108534W WO2019153759A1 WO 2019153759 A1 WO2019153759 A1 WO 2019153759A1 CN 2018108534 W CN2018108534 W CN 2018108534W WO 2019153759 A1 WO2019153759 A1 WO 2019153759A1
Authority
WO
WIPO (PCT)
Prior art keywords
plan
goods
path
cargo
distribution
Prior art date
Application number
PCT/CN2018/108534
Other languages
English (en)
French (fr)
Inventor
陈浩洋
李启仁
袁明轩
曾嘉
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP18905663.3A priority Critical patent/EP3739530A4/en
Publication of WO2019153759A1 publication Critical patent/WO2019153759A1/zh
Priority to US16/986,508 priority patent/US20200364664A1/en

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/08Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/083Shipping
    • G06Q10/0835Relationships between shipper or supplier and carriers
    • G06Q10/08355Routing methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/18Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • G06F30/27Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/04Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/04Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G06Q10/047Optimisation of routes or paths, e.g. travelling salesman problem
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/08Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/08Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/083Shipping
    • G06Q10/0838Historical data
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40Business processes related to the transportation industry

Definitions

  • the present application relates to the field of logistics, and in particular to a method for determining a transportation scheme, a method and a device for training a quick loading model.
  • the path scheme is obtained by the ant colony algorithm, and then the loading scheme of the cargo in the route scheme is obtained through the three-dimensional loading simulation, that is, the loading manner of the cargo in the container.
  • the three-dimensional loading simulation cannot process the goods in parallel, and a large amount of calculation is required, which takes a lot of time, especially in a scene with a large amount of goods, which consumes a larger amount of time, thereby reducing the efficiency of obtaining the loading scheme and outputting the mounting rate. In turn, the efficiency of the target transportation plan is affected.
  • the embodiment of the present application provides a method for determining a transportation scheme, a method and a device for training a quick loading model, and is used for carrying out cargo transportation, especially in a large and complicated transportation scenario, which can quickly obtain a target transportation scheme and reduce transportation costs. Improve transportation efficiency.
  • the first aspect of the present application provides a method for determining a transportation solution, which may include:
  • one path plan may include at least one transport path
  • the first set of goods allocation plans corresponding to each of the at least one path plan includes at least one cargo allocation plan
  • each of the at least one path plan Each of the corresponding goods distribution plan sets is a plan for allocating the goods to be transported for the corresponding path plan; determining, corresponding to each path plan of the at least one path plan, according to the fast load model a mounting rate of each cargo allocation scheme in a set of goods distribution, the fast loading model being obtained by offline training through offline simulation data, the offline simulation data including a historical loading scheme calculated by a three-dimensional loading algorithm, the mounting ratio Allocation plan for a certain goods
  • the proportion of the cargo loaded into the container occupies the container; the sourcing rate is used in each of the at least one path plan and the first set of goods allocation plans corresponding to each of the container
  • each of the fast load models that perform offline training through offline simulation data may determine each The installation rate of each cargo allocation scheme in the first cargo allocation scheme set corresponding to the path scheme can quickly obtain the installation ratio of all the cargo allocation schemes corresponding to each route scheme, and can reduce the acquisition of all the cargo corresponding to each route scheme.
  • the length of the placement rate of the distribution plan increases the efficiency of determining the target transportation plan.
  • the fast loading model is obtained by offline simulation training through offline simulation data, and the offline simulation data includes data obtained by three-dimensional loading operation, which can improve the accuracy of the obtained mounting rate.
  • the obtaining the at least one path plan and the first set of the first item of the goods plan corresponding to each of the at least one path plan may include:
  • a target shipping order is obtained, where the target shipping order includes transportation node information and a cargo information to be transported, the transportation node information includes a shipping starting point, a shipping destination, and M picking points, and the information to be transported includes distribution at the M picking points.
  • the information of the goods to be transported, the M is a positive integer; and then determining the at least one path plan according to the transport node in the transport node information, wherein one path plan may include at least one transport path, each of the at least one transport path
  • the transport path includes a freight origin, a freight destination, and N pick-up points in the M pick-up points, where N is a positive integer, and N ⁇ M, to complete the goods to be transported distributed at the M pick-up points, the at least one route plan
  • Each of the path plans covers the M pick-up points; the allocation of the goods to be transported is performed for each of the at least one path plan to obtain each of the at least one path plan
  • the path planning and the goods distribution are performed according to the information provided in the target shipping list, and when the path plan is determined, the path planning may be directly performed according to the transportation node, and the path search may be reduced.
  • the time is long, the efficiency of the path planning is improved, and after the path planning is completed, the goods are allocated according to the planned path plan to obtain the goods distribution set of each path plan, and then each path plan and each path plan corresponding to each path plan.
  • the integrated evaluation of each cargo allocation plan in the collection of goods distribution plan to obtain the target transportation plan can improve the overall efficiency of the target transportation plan.
  • the determining the at least one path scheme according to the transit node information may include:
  • the transfer hyperparameters of the M picking points are initialized based on the historical path data to obtain a hyperparameter matrix; the transition probability of the M picking points is determined by the superparametric matrix a distribution, the transfer probability distribution comprising a transfer probability of a container in the transport path between the freight origin and the M pick-up points, between the freight destination and the M pick-up points, or between the M pick-up points; The probability distribution determines each of the at least one path plan to obtain the at least one path plan.
  • the path planning may be performed by using historical path data, specifically including initializing the transfer hyperparameters of the M picking points by using the historical path data, and then determining the transfer point transition probability distribution according to the transfer hyperparameter, the probability distribution is The probability of the transfer of the container of each transport path in the route plan between the pick-up point and the port. It should be understood that the more times a jump occurs in the historical path data, the higher the probability that the jump corresponds, and each transport path of each path plan in the at least one path plan may be determined according to the obtained pick-up point transition probability distribution. The efficiency of obtaining the at least one path solution can be further improved, and the path parameter transfer hyperparameter is calculated through the historical path data, so that the obtained path solution can be made more accurate.
  • the method may further include:
  • the transfer hyperparameters of the M pick-up points are initialized based on a heuristic algorithm to obtain the hyper-parametric matrix.
  • the at least one Each of the path plans in each of the path plans performs the allocation of the goods to be transported to obtain each of the first item allocation plan sets corresponding to each of the at least one path plan Can include:
  • the goods of each of the M pick-up points obtained from the target shipping list are clustered according to the clustering conditions to obtain a clustering result, which may include the length, width, height and weight of the goods.
  • the clustering condition may further include a material, a bearing coefficient, or a minimum area, and the clustering result is sampled and calculated by the first cargo allocation hyperparameter of each of the M picking points to obtain the M
  • the first cargo distribution mode of each picking point of each picking point, the first cargo allocation hyperparameter of each picking point of the picking point is the super distribution of goods for each picking point of the M picking points a parameter
  • each of the first goods distribution modes in each of the M delivery points is allocated in a manner of allocating goods distributed at the delivery point for the corresponding route plan, the first cargo distribution is super
  • the parameter may be a uniformly distributed hyperparameter, or may be obtained by repeating the last cargo allocation plan when the goods are repeatedly distributed; the first cargo from each of the M picking points
  • the characteristics of the goods distributed by the delivery point may be referred to, including length, width, height or weight, and the precise clustering may be used.
  • fuzzy clustering which can be adjusted according to actual needs, the goods of the pick-up point can be quickly classified, so that the goods can be quickly distributed to obtain each of the goods distribution plan sets corresponding to each path plan in the at least one path plan. Goods distribution plan.
  • the first feature vector is used to indicate the goods to be transported in a certain goods distribution plan a feature value, for example, a vector consisting of length, width, height, or weight of the goods in each cargo distribution plan; each cargo in the first cargo distribution plan corresponding to each of the at least one route plan a first feature vector of the allocation scheme is input into the fast load model to obtain a mounting rate of each cargo allocation scheme in the first cargo allocation scheme set corresponding to each of the at least one path scheme, the mounting ratio Included in a volume mounting ratio and a load mounting ratio, the volume mounting ratio includes a ratio of a volume occupied by a cargo allocated to each of the at least one of the at least one routing scheme to a container load volume, the load carrying capacity a set of weights of goods allocated for each of the at least one of the at least one path plan The proportion of the load that is boxed.
  • acquiring a first feature vector of each cargo allocation scheme in the first cargo allocation scheme set corresponding to each of the obtained at least one path scheme where the first feature vector is indicative of at least a feature value of one of the first item allocation plan corresponding to each path plan in each of the path plans, each of the first item of the item allocation plan corresponding to each of the at least one path plan
  • the first feature vector of the cargo allocation scheme is input to the fast loading model, and the mounting rate of each cargo allocation scheme in the first cargo allocation scheme set corresponding to each of the at least one routing scheme may be obtained, the mounting The rate may include a volume mounting rate and a load mounting rate.
  • the first feature vector of each cargo allocation scheme may be input into the fast loading model, and the first corresponding to each path scheme in at least one path scheme may be quickly obtained.
  • the rate of installation of each cargo allocation plan in the collection of goods distribution plan improve each cargo Mounting efficiency of feature programs.
  • the acquiring the first cargo allocation plan set corresponding to each path plan in the at least one path solution may include:
  • the second feature vector of each of the goods to be transported includes a length, a width, a height and a weight of the corresponding goods; according to each of the goods to be transported Calculating, by the second feature vector of the goods, the goods distributed by each of the M pick-up points for each of the first goods allocation plans corresponding to each of the at least one path plan a third feature vector, the third feature vector of each of the first goods distribution plans corresponding to each of the at least one path plan includes a second feature vector of each of the goods to be transported Mean and covariance; weighting the third feature vector of each cargo allocation scheme in the first cargo allocation scheme corresponding to each of the at least one path scheme to obtain each of the corresponding at least one path scheme The first feature vector of each cargo allocation scheme in the first set of goods allocation schemes corresponding to the path schemes.
  • the specific step of acquiring the first feature vector of each of the first item allocation plan sets corresponding to each path plan of the at least one path plan may be: first obtaining the goods to be transported a second feature vector of each of the goods, and calculating, according to the second feature vector of each cargo, each of the M cargo pick-up points corresponding to each of the at least one route plan a third feature vector of the goods distribution plan, and weighting the third feature vector, and finally obtaining the first of each cargo allocation plan in the first cargo allocation plan set corresponding to each path plan in the at least one path plan A feature vector.
  • the first embodiment of the first aspect of the present application, and the sixth embodiment of the first aspect of the present application in the seventh embodiment of the first aspect of the present application, And integrating, by the mounting ratio, each of the at least one path plan and each of the first item allocation plan corresponding to each of the at least one path plan, To determine the target transportation plan, you can include:
  • the scores of each of the goods allocation schemes may be calculated by using a preset evaluation function and a mounting rate for all the goods allocation schemes obtained, and if there is no high score in all the goods allocation schemes And determining, by the second threshold value, the target goods distribution plan from the goods distribution whose score is higher than the second threshold, and determining the one goods distribution plan if the number of the goods distribution plan is higher than the second threshold For the target goods distribution plan, if the number of the goods distribution plan higher than the second threshold is at least two, the goods distribution plan with the highest score may be randomly determined from the at least two goods distribution plans higher than the second threshold.
  • the solution is a target goods distribution plan, and the path plan corresponding to the target goods distribution plan is determined as a target path plan to obtain a target transportation plan.
  • each goods distribution plan is scored to determine a target goods distribution plan, The optimal target transportation plan can be obtained.
  • the evaluation function includes:
  • the For the path plan vector m is the number of containers, The volumetric loading rate vector for m containers, The loading rate vector of the m containers; the ⁇ , the ⁇ and the ⁇ are weighting parameters, the r Vi is the volume mounting rate of the i-th container, and the r Wi is the load of the i-th container Rate, the rate For the average volume mounting rate of the m containers, The average load mounting rate for the m containers.
  • the embodiment of the present application adds an evaluation function for evaluating a goods distribution plan and a path plan, and an optimal target transportation plan can be obtained according to the evaluation function.
  • the method further includes:
  • the clustering result is sampled and calculated by the second cargo allocation hyperparameter of each of the M picking points, to Obtaining a second set of goods distribution manners of each of the M pick-up points, and each of the items of the second pick-up mode of each of the M pick-up points is allocated for the corresponding path plan pair
  • the manner in which the goods distributed at the pick-up point are distributed, and the second item allocation hyper-parameter of each of the M pick-up points is in the first set of goods allocation schemes corresponding to each of the at least one path plan
  • Each cargo allocation plan updates the first cargo allocation hyperparameter of each of the M picking points; and selects the cargo from the second cargo allocation manner of each of the M picking points
  • the allocation manner is combined to obtain each of the goods distribution plans in the second set of goods distribution plans corresponding to each of the at least one path plan,
  • Each of the second item allocation plan sets corresponding to each of the one path plan is a plan for allocating the goods to be transported for
  • the first of each pick-up point may be adopted by each of the goods distribution plans in the first set of goods distribution plans.
  • the cargo allocation hyperparameter is updated, and the second cargo allocation hyperparameter of each picking point is obtained, and then the cargo is re-allocated for each cargo picking point according to the second cargo allocation hyperparameter to obtain the second of each route scheme.
  • Each of the goods distribution plans in the set of goods distribution plans and then continuing to perform a further integrated evaluation of each of the goods distribution plans in the second set of goods distribution plans until a stop condition is reached, for example, obtaining a score greater than the second threshold
  • the path plan may be re-planned, or the goods may be re-allocated directly through the at least one path plan.
  • the method further includes :
  • the loading plan is the loading mode of the goods to be transported in the container in each transport path in the target path plan.
  • the model of the container may be further determined, and the container type may be adjusted according to the installation rate to determine the container model that is more closely matched with the installation rate to save transportation costs.
  • the loading plan can be further generated by the three-dimensional loading algorithm to determine the loading mode of the goods in the container, which can improve the efficiency of loading the goods.
  • the eleventh embodiment of the first aspect of the present application Integrating each of the goods distribution plans in the first set of goods distribution plans corresponding to each of the at least one path plan and each of the at least one path plan by the mounting rate Before evaluating to determine the target transportation plan, the method also includes:
  • the mounting rate determines that the L pick-up points of the M pick-up points further include the remaining goods not allocated to the container, determining a remaining cargo path plan and a remaining goods allocation plan for the remaining goods, the L ⁇ the M,
  • the L is a positive integer
  • Integrating each of the at least one path plan and each of the at least one path plan corresponding to each of the at least one path plan by using the mounting rate to Identify the target transportation plan can include:
  • the route planning and the distribution of the goods may be performed according to the remaining goods.
  • the target path plan and the target goods distribution plan as the target transportation plan, to obtain a complete transportation plan for the goods to be transported.
  • the second aspect of the present application provides a method for training a fast loading model, which may include:
  • offline simulation data is acquired, the offline simulation data includes a historical loading scheme and a historical mounting ratio calculated by three-dimensional loading during off-line simulation; and then a feature vector is obtained from the offline simulation data, the feature vector including the historical loading scheme Corresponding historical transport cargo eigenvalues; the eigenvectors are replaced with training data in a preset format; the predictive model is trained by the training data to obtain a fast loading model for outputting goods for each transport path
  • the placement rate of each of the goods distribution plans in the distribution plan set which is the ratio of the goods loaded into the container in each of the goods distribution plans.
  • the fast loading model can be trained by offline simulation data, which is used to quickly obtain the mounting rate of the cargo allocation scheme, and can improve the efficiency of determining the target transportation scheme.
  • the preset format is: (feature vector, historical mounting ratio).
  • the prediction model may include, but is not limited to, a linear regression model, a ridge regression model, LASSO model, support vector machine model, random forest model, XgBoost model or artificial neural network model.
  • the acquiring module may include:
  • Each of the first historical goods allocation plan sets corresponding to each path plan in the solution is a plan for allocating the historical transport goods for the corresponding path plan; determining the at least one historical path plan according to the three-dimensional loading algorithm a mounting rate of each of the goods distribution plans in the first set of goods distribution corresponding to each of the path plans, and a loading plan, the ratio of the proportion of the goods loaded into the container in the cargo allocation plan occupying the container; Through the mounting rate
  • Each of the historical path plans and each of the first historical goods allocation plan sets corresponding to each of the at least one historical path plan are integrated and evaluated to determine a target transportation
  • the three-dimensional loading algorithm model for off-line training through off-line simulation data determines the mounting rate of the cargo allocation scheme corresponding to each path scheme, and the cargo allocation scheme corresponding to each path scheme can be accurately obtained.
  • the installation rate improves the accuracy of determining offline simulation data.
  • the at least one historical path solution and the first historical cargo allocation plan set corresponding to each of the at least one historical path solution are obtained, Can include:
  • the historical freight bill includes transport node information and a historical transport cargo information.
  • the transport node information includes a freight origin, a freight destination, and M pick-up points, and the historical transport cargo information is distributed at the M pick-up points.
  • the M being a positive integer; then determining the at least one historical route plan according to the transport node in the transport node information, wherein the one route plan may include at least one transport path, the at least one transport path Each transport path includes a freight origin, a cargo destination, and N pick-up points in the M pick-up points, where N is a positive integer and N ⁇ M, to complete the historical transport cargo distributed at the M pick-up points, the at least one history
  • Each of the path plans covers the M pick-up points; the historical transport goods are allocated for each of the at least one historical path plan to obtain the at least one historical path
  • Each of the first historical goods distribution plan sets corresponding to each path plan in the plan Distribution plan.
  • the route planning and the cargo distribution are performed according to the information provided in the historical freight bill.
  • the route planning may be directly performed according to the transport node, and the route search may be reduced.
  • the time is long to improve the efficiency of path planning.
  • the goods are allocated according to the planned path plan to obtain the historical cargo allocation plan set of each path plan, followed by each path plan and each path plan.
  • the integrated evaluation of each cargo allocation plan in the corresponding historical cargo distribution plan set to obtain the target transportation plan can improve the overall efficiency of the target transportation plan.
  • the determining the at least one historical path solution according to the transit node information may include:
  • the transfer hyperparameters of the M pick-up points based on the historical path data to obtain a hyperparameter matrix, the historical path data including when transporting the goods to be transported historically a historical path plan; determining, by the hyperparametric matrix, a transition probability distribution of the M pick-up points, the transfer probability distribution including a container in the transport path between the freight start point and the M pick-up points, the freight destination and the M a transition probability between the pick-up points or between the M pick-up points; determining each of the at least one historical path plan based on the transition probability distribution to obtain the at least one historical path plan.
  • the path planning may be performed by using historical path data, specifically including initializing the transfer hyperparameters of the M picking points by using the historical path data, and then determining the transfer point transition probability distribution according to the transfer hyperparameter, the probability distribution is The probability of the transfer of the container of each transport path in the route plan between the pick-up point and the port. It should be understood that the more times a jump occurs in the historical path data, the higher the probability that the jump corresponds, and each transport of each path scheme in at least one historical path scheme may be determined according to the obtained pick-up point transition probability distribution.
  • the path can further improve the efficiency of obtaining the at least one historical path solution, and calculating the delivery point transfer hyperparameter through the historical path data, so that the obtained path solution can be more accurate.
  • the method may further include:
  • the transfer hyperparameters of the M pick-up points are initialized based on a heuristic algorithm to obtain the hyper-parametric matrix.
  • the transfer parameter hyperparameter of the pick-up point cannot be initialized by the historical path data at this time, and the heuristic algorithm can be selected to initialize the transfer hyper-parameter of the pick-up point, and a parameter for determining the delivery point hyper-parameter is added. the way.
  • the at least one Each of the path plans in the historical path plan performs the allocation of the historical transport goods to obtain each of the first historical goods distribution plan sets corresponding to each of the at least one historical path plan
  • the cargo distribution plan can include:
  • the goods of each of the M pick-up points obtained from the historical freight bill are clustered according to the clustering conditions to obtain a clustering result, which may include the length, width, height and weight of the goods.
  • the clustering condition may further include a material, a bearing coefficient, or a minimum area, and the clustering result is sampled and calculated by the first cargo allocation hyperparameter of each of the M picking points to obtain the M
  • the first cargo distribution mode of each picking point of each picking point, the first cargo allocation hyperparameter of each picking point of the picking point is the super distribution of goods for each picking point of the M picking points a parameter
  • each of the first goods distribution modes in each of the M delivery points is allocated in a manner of allocating goods distributed at the delivery point for the corresponding route plan, the first cargo distribution is super
  • the parameter may be a uniformly distributed hyperparameter, or may be obtained by repeating the last cargo allocation plan when the goods are repeatedly distributed; the first cargo from each of the M picking points
  • the characteristics of the goods distributed by the delivery point may be referred to, including length, width, height or weight, and the precise clustering may be used.
  • the fuzzy clustering is used, which can be adjusted according to actual needs, and the goods of the pick-up point can be quickly classified, so that the goods can be quickly distributed to obtain the historical goods distribution plan set corresponding to each path plan in the at least one historical path plan.
  • Each cargo distribution plan is used, which can be adjusted according to actual needs, and the goods of the pick-up point can be quickly classified, so that the goods can be quickly distributed to obtain the historical goods distribution plan set corresponding to each path plan in the at least one historical path plan.
  • the seventh embodiment of the second aspect of the present application Determining, by the mounting rate, each of the at least one historical path plan and each of the first historical goods distribution plan sets corresponding to each of the at least one historical path plan Integrate the evaluation to determine the target transportation plan, which can include:
  • the scores of each of the goods allocation schemes may be calculated by using a preset evaluation function and a mounting rate for all the goods allocation schemes obtained, and if there is no high score in all the goods allocation schemes And determining, by the second threshold value, the target goods distribution plan from the goods distribution whose score is higher than the second threshold, and determining the one goods distribution plan if the number of the goods distribution plan is higher than the second threshold For the target goods distribution plan, if the number of the goods distribution plan higher than the second threshold is at least two, the goods distribution plan with the highest score may be randomly determined from the at least two goods distribution plans higher than the second threshold.
  • the solution is a target goods distribution plan, and the path plan corresponding to the target goods distribution plan is determined as a target path plan to obtain a target transportation plan.
  • each goods distribution plan is scored to determine a target goods distribution plan, The optimal target transportation plan can be obtained.
  • the evaluation function includes:
  • the For the path plan vector m is the number of containers, The volumetric loading rate vector for m containers, The loading rate vector of the m containers; the ⁇ , the ⁇ and the ⁇ are weighting parameters, the r Vi is the volume mounting rate of the i-th container, and the r Wi is the load of the i-th container Rate, the rate For the average volume mounting rate of the m containers, The average load mounting rate for the m containers.
  • the embodiment of the present application adds an evaluation function for evaluating a goods distribution plan and a path plan, and an optimal target transportation plan can be obtained according to the evaluation function.
  • the method further includes:
  • the clustering result is sampled and calculated by the second cargo allocation hyperparameter of each of the M picking points, to Obtaining a second set of goods distribution manners of each of the M pick-up points, and each of the items of the second pick-up mode of each of the M pick-up points is allocated for the corresponding path plan pair
  • the manner in which the goods distributed at the pick-up point are distributed, and the second item allocation hyper-parameter of each of the M pick-up points is a first historical item allocation plan corresponding to each path plan in the at least one historical path plan
  • Each cargo allocation plan in the set updates the first cargo allocation hyperparameter of each of the M picking points; and the second cargo allocation method set of each picking point of the M picking points respectively Selecting a cargo allocation manner to combine to obtain each of the second historical cargo allocation schemes corresponding to each of the at least one historical path scheme a distribution plan, each of the second historical goods allocation plan sets corresponding to each of the at least one historical path plan is
  • each cargo delivery plan in the first historical cargo allocation scheme set may be used for each cargo delivery point.
  • a cargo distribution hyperparameter is updated to obtain a second cargo allocation hyperparameter for each pickup point, and then the cargo is re-allocated for each cargo pick-up point according to the second cargo distribution hyperparameter to obtain the first route scheme
  • each cargo distribution plan in the set of historical goods distribution plans and then continuing to perform further integrated evaluation on each of the goods distribution plans in the second historical goods distribution plan set until a stop condition is reached, for example, obtaining a score greater than the The second threshold of the goods allocation scheme, or the number of iterations, reaches a preset number of times. Therefore, in the embodiment of the present application, by repeatedly distributing the goods distribution plan and integrating the evaluation, a better target goods distribution plan and a target path plan can be obtained.
  • the path plan may be re-planned, or the goods may be re-allocated directly through the at least one historical path plan.
  • each of the first historical goods allocation plan sets corresponding to each of the at least one historical path plan and each of the at least one historical path plan by the mounting rate
  • the method also includes:
  • the mounting rate determines that the L pick-up points of the M pick-up points further include the remaining goods not allocated to the container, determining a remaining cargo path plan and a remaining goods allocation plan for the remaining goods, the L ⁇ the M,
  • the L is a positive integer
  • Integrating each of the at least one historical path plan and each of the first historical item allocation plan corresponding to each of the at least one historical path plan by the mounting rate Evaluation to determine the target transportation plan can include:
  • the route planning and the cargo distribution may be performed according to the remaining cargo.
  • the target path plan and the target goods distribution plan as the target transportation plan, to obtain a complete transportation plan for the historical transportation goods.
  • the third aspect of the present application provides a determining apparatus, including:
  • An obtaining module configured to acquire at least one path plan and a first set of goods allocation plans corresponding to each of the at least one path plan, each path plan of the at least one path plan is for transporting goods to be transported a planned transport path, the first set of goods allocation plans corresponding to each of the at least one path plan includes at least one goods distribution plan, and the first set of goods allocation plans corresponding to each of the at least one path plan
  • Each of the goods distribution schemes is a scheme for allocating the goods to be transported for the corresponding route plan;
  • a fast loading module configured to determine, according to the fast loading model, a mounting rate of each of the first item allocation sets corresponding to each of the at least one path plan, the fast loading model is to simulate data through offline Obtaining offline training, the offline simulation data includes a historical loading plan calculated by a three-dimensional loading algorithm, and the mounting ratio is a ratio of a container loaded with a container occupying the container in a certain cargo allocation scheme;
  • An evaluation module by means of the mounting rate, an integrated evaluation of each of the at least one path plan and each of the first item allocation plan corresponding to each of the at least one path plan To determine a target transportation plan, wherein the target transportation plan includes a target route allocation plan corresponding to the target path plan and the target path plan.
  • the acquiring module includes:
  • Obtaining a sub-module configured to obtain a target shipping slip
  • the target shipping bill includes transportation node information and cargo information to be transported
  • the transportation node information includes a shipping starting point, a shipping destination, and M picking points
  • the to-be-transported cargo information is included in the M information of the goods to be transported at the pick-up point, the M is a positive integer;
  • each of the at least one path plan includes at least one transport path
  • each of the at least one transport path includes a shipping start point And a cargo picking point and N picking points in the M picking points
  • each of the at least one path scheme covers the M picking points, the N is a positive integer, and N ⁇ M;
  • a goods distribution sub-module configured to allocate the goods to be transported for each of the at least one path plan to obtain a first corresponding to each path plan of the at least one path plan Each cargo allocation plan in the collection of goods distribution plans.
  • the path planning sub-module is specifically configured to:
  • the determining apparatus further includes:
  • an initialization module configured to initialize the transfer hyper-parameters of the M pick-up points based on a heuristic algorithm to obtain the super-parameter matrix if the number of the historical path data is not greater than the first threshold.
  • the cargo distribution sub-module Specifically used for:
  • the goods of each of the M pick-up points are clustered according to the clustering conditions to obtain a clustering result, which includes the length, width, height and weight of the goods;
  • the clustering result is sampled and calculated by using the first cargo allocation hyperparameter of each of the M picking points to obtain a first set of goods allocation manners of each of the M picking points, the M
  • the first item allocation hyperparameter of each picking point in the picking point is a hyper parameter for allocating goods to each picking point of the M picking points, and the first item allocation mode set of each picking point of the M picking points
  • Each of the goods is allocated in such a way as to distribute the goods distributed at the pick-up point for the corresponding route plan;
  • the fast loading module is specific Used for:
  • the first feature vector is used to indicate the goods to be transported in a certain goods distribution plan Characteristic value
  • the volume of the cargo allocated by each transport path in the path plan occupies a proportion of the load volume of the container
  • the load mounting rate is the weight of the cargo allocated by each of the at least one of the at least one route plan The proportion of the load on the container.
  • the fast loading module is specifically configured to:
  • the second feature vector of each of the goods to be transported includes a length, a width, a height, and a weight of the corresponding goods;
  • a third feature vector of each of the goods distribution plans includes each of the goods to be transported The mean and covariance of the second eigenvector of the goods;
  • the evaluation module is specifically used to:
  • the all-carriage allocation plan includes a goods distribution plan with a score higher than a second threshold, determining the target goods distribution plan from the goods distribution plan whose score is higher than the second threshold, and corresponding to the target goods distribution plan a path plan as the target path plan;
  • the target transportation plan is determined by the target goods distribution plan and the target path plan.
  • the evaluation function includes:
  • the For the path plan vector m is the number of containers, The volumetric loading rate vector for m containers, The loading rate vector of the m containers; the ⁇ , the ⁇ and the ⁇ are weighting parameters, the r Vi is the volume mounting rate of the i-th container, and the r Wi is the load of the i-th container Rate, the rate For the average volume mounting rate of the m containers, The average load mounting rate for the m containers.
  • the evaluation module is further configured to:
  • the clustering result is sampled and calculated by the second cargo allocation hyperparameter of each of the M picking points, to Obtaining a second set of goods distribution manners of each of the M pick-up points, and each of the items of the second pick-up mode of each of the M pick-up points is allocated for the corresponding path plan pair
  • the manner in which the goods distributed at the pick-up point are distributed, and the second item allocation hyper-parameter of each of the M pick-up points is in the first set of goods allocation schemes corresponding to each of the at least one path plan
  • Each cargo allocation plan updates the first cargo allocation hyperparameter of each of the M picking points;
  • each of the second goods distribution plan sets corresponding to each of the at least one path plan is a plan for allocating the goods to be transported for the corresponding path plan;
  • the determining device further includes:
  • a post-processing module configured to perform an integrated evaluation of each of the at least one path plan and the goods allocation plan of each of the at least one path plan by the mounting rate to determine the target transportation plan, Determining a model of a container of each transport path in the target path scheme according to the target cargo allocation scheme and the target route scheme;
  • a three-dimensional loading module configured to generate a loading plan according to a model of a container of each transportation path in the target path solution determined by the post-processing module, and a loading plan for each item of the to-be-transported goods in the target path plan The loading method inside the container in the transport path.
  • the determining device may further include:
  • each of the at least one path plan and each of the first item allocation plan sets corresponding to each of the at least one path plan are performed by the mounting rate Integrating the evaluation to determine the target transportation plan, and if the sizing rate determines that the L picking points in the M picking points further include the remaining goods not allocated to the container, determining a remaining cargo path plan for the remaining goods And the remaining goods allocation scheme, the L ⁇ the M, the L is a positive integer;
  • the evaluation module is further configured to, by the mounting rate, each of the at least one path plan and each of the corresponding first item allocation plan sets, and the remaining goods path plan and the remaining The cargo distribution plan is integrated and evaluated to determine the target transportation plan.
  • a fourth aspect of the present invention provides a training apparatus, including:
  • An acquisition module configured to acquire offline simulation data, where the offline simulation data includes a historical loading scheme and a historical mounting ratio calculated by three-dimensional loading;
  • the acquiring module is further configured to acquire a feature vector from the offline simulation data, where the feature vector includes a feature value of a historical transport cargo corresponding to the historical loading solution;
  • a conversion module configured to convert the feature vector into training data in a preset format
  • a training module for training the prediction model by the training data to obtain a fast loading model for outputting a mounting rate of each cargo allocation scheme in the collection of goods distribution schemes of each transportation route, the mounting ratio The proportion of the container in which the container is loaded for each cargo allocation plan occupies the container.
  • the preset format is: (feature vector, historical mounting ratio).
  • the prediction model includes: a linear regression model, a ridge regression model, an LASSO model, and a support Vector machine model, random forest model, XgBoost model or artificial neural network model.
  • the acquiring module may include:
  • Obtaining a sub-module configured to acquire at least one historical path solution and a first historical cargo allocation plan set corresponding to each of the at least one historical path solution, each of the at least one historical path solution being for history a transportation route planned for transporting the goods for transportation, the first historical goods distribution plan set corresponding to each of the at least one historical path plan includes at least one goods distribution plan, each of the at least one historical path plan Each of the corresponding first historical goods allocation plan sets is a plan for allocating the historical transport goods for the corresponding path plan;
  • a three-dimensional loading sub-module configured to determine, according to the three-dimensional loading algorithm, a mounting rate of each of the first goods distribution sets corresponding to each of the at least one historical path solution, the mounting rate is a certain The proportion of goods loaded into the container in the cargo distribution plan that occupy the container;
  • An evaluation submodule by each of the at least one historical path plan and each of the first historical goods allocation plan sets corresponding to each of the at least one historical path plan by the mounting rate
  • the solution performs an integrated evaluation to determine a target transportation plan, wherein the target transportation plan includes a target route allocation plan corresponding to the target path plan and the target path plan.
  • the three-dimensional loading algorithm can be used for calculation when training the fast loading model, and the historical loading scheme corresponding to the historical path scheme is obtained, and the mounting rate of the cargo allocation scheme corresponding to the historical path scheme can be accurately output.
  • the acquiring submodule includes:
  • An obtaining unit configured to obtain a historical shipping slip, the historical shipping bill includes transportation node information and historical transportation cargo information, where the transportation node information includes a shipping origin, a shipping destination, and M picking points, and the historical shipping cargo information is distributed in the M
  • the information of the historical transport cargo at the pick-up point, the M is a positive integer
  • each of the at least one historical path plan includes at least one transport path, each of the at least one transport path includes a freight a start point, a freight destination, and N pick-up points in the M pick-up points, each path plan of the at least one historical path plan covers the M pick-up points, the N is a positive integer, and N ⁇ M;
  • a goods distribution unit configured to allocate the historical transportation goods for each of the at least one historical path solution to obtain a corresponding one of each of the at least one historical path solution A distribution plan for each item in a collection of historical goods distribution plans.
  • the path planning unit is specifically configured to:
  • the training apparatus further includes:
  • an initialization module configured to initialize the transfer hyper-parameters of the M pick-up points based on a heuristic algorithm to obtain the super-parameter matrix if the number of the historical path data is not greater than the first threshold.
  • the cargo distribution unit Specifically used for:
  • the goods of each of the M pick-up points are clustered according to the clustering conditions to obtain a clustering result, which includes the length, width, height and weight of the goods;
  • the clustering result is sampled and calculated by using the first cargo allocation hyperparameter of each of the M picking points to obtain a first set of goods allocation manners of each of the M picking points, the M
  • the first item allocation hyperparameter of each picking point in the picking point is a hyper parameter for allocating goods to each picking point of the M picking points, and the first item allocation mode set of each picking point of the M picking points
  • Each of the goods is allocated in such a way as to distribute the goods distributed at the pick-up point for the corresponding route plan;
  • the evaluation sub-module Specifically for:
  • the all-carriage allocation plan includes a goods distribution plan with a score higher than a second threshold, determining the target goods distribution plan from the goods distribution plan whose score is higher than the second threshold, and corresponding to the target goods distribution plan a path plan as the target path plan;
  • the target transportation plan is determined by the target goods distribution plan and the target path plan.
  • the evaluation function includes:
  • the For the path plan vector m is the number of containers, The volumetric loading rate vector for m containers, The loading rate vector of the m containers; the ⁇ , the ⁇ and the ⁇ are weighting parameters, the r Vi is the volume mounting rate of the i-th container, and the r Wi is the load of the i-th container Rate, the rate For the average volume mounting rate of the m containers, The average load mounting rate for the m containers.
  • the evaluation sub-module is further configured to:
  • the clustering result is sampled and calculated by the second cargo allocation hyperparameter of each of the M picking points, to Obtaining a second set of goods distribution manners of each of the M pick-up points, and each of the items of the second pick-up mode of each of the M pick-up points is allocated for the corresponding path plan pair
  • the manner in which the goods distributed at the pick-up point are distributed, and the second item allocation hyper-parameter of each of the M pick-up points is a first historical item allocation plan corresponding to each path plan in the at least one historical path plan
  • Each cargo allocation plan in the set updates the first cargo allocation hyperparameter of each of the M picking points;
  • each of the second historical goods distribution plan sets corresponding to each of the at least one historical path plan is a distribution of the historical transportation goods for the corresponding path plan Program;
  • a path rate for each of the at least one historical path plan by the evaluation function and a mounting rate of each of the second historical goods allocation plan sets of each of the at least one historical path plan a score calculation for each of the cargo allocation schemes in the second historical cargo allocation plan set, and a mounting rate of each of the cargo distribution plans in the second historical cargo allocation plan set of each of the at least one historical route plans Obtained from the three-dimensional loading submodel.
  • the path plan can be re-planned, or the goods can be re-allocated directly through the at least one historical path plan.
  • the training device further includes:
  • the distribution plan performs an integrated evaluation to determine the remaining transportation plan, and if the L-lifting point in the M pick-up points further includes the remaining goods not allocated to the container, the remaining items are determined for the remaining goods.
  • the cargo path plan and the remaining goods allocation plan the L ⁇ the M, the L being a positive integer;
  • the evaluation sub-module is further configured to, by the mounting rate, each of the at least one historical path plan and each of the corresponding first historical goods distribution plan sets, and the remaining goods path plan Integrate evaluation with the remaining cargo allocation plan to determine the target transportation plan.
  • the fifth aspect of the present application provides a determining apparatus, which may include:
  • processor a processor, a memory, a bus, and an input/output interface, the processor, the memory and the input/output interface being connected through the bus;
  • the memory for storing program code
  • the steps of the method provided by the first aspect of the present application are performed when the processor calls the program code in the memory.
  • the sixth aspect of the present application provides a training apparatus, which may include:
  • processor a processor, a memory, a bus, and an input/output interface, the processor, the memory and the input/output interface being connected through the bus;
  • the memory for storing program code
  • a seventh aspect of the embodiments of the present application provides a storage medium having stored thereon, where the programmable instructions are executed on a computer, causing the computer to perform the first aspect or the first aspect described in any implementation manner. Methods.
  • the storage medium includes: a U disk, a mobile hard disk, a read only memory (English abbreviation ROM, English full name: Read-Only Memory), a random access memory (English abbreviation: RAM, English full name: Random Access Memory), a disk or a disk. And other media that can store program code.
  • An eighth aspect of the embodiments of the present application provides a storage medium having stored thereon, where the programmable instructions are executed on a computer, causing the computer to perform the second aspect or the second aspect described in any implementation manner.
  • the storage medium includes: a U disk, a mobile hard disk, a read only memory (English abbreviation ROM, English full name: Read-Only Memory), a random access memory (English abbreviation: RAM, English full name: Random Access Memory), a disk or a disk. And other media that can store program code.
  • a ninth aspect of the embodiments of the present application provides a computer program product, comprising: computer software instructions, wherein the computer software instructions are loadable by a processor to implement the flow in the method for determining a transportation solution of the first aspect.
  • a tenth aspect of the embodiments of the present application provides a computer program product, the computer program product comprising computer software instructions, wherein the computer software instructions are loaded by a processor to implement the process in the method for training the fast loading model of the second aspect. .
  • An eleventh aspect of the present application provides a simulation system, including a determining device and a training device, the determining device configured to perform the steps in any one of the first aspect to the first aspect of the present application; The steps of any of the second to second aspects of the present application.
  • the embodiments of the present application have the following advantages:
  • the fast loading model can be used to obtain the loading ratio of the cargo allocation scheme corresponding to each path scheme, and then the target transportation scheme is determined according to the mounting ratio, wherein the fast loading model is offline by offline simulation data.
  • the training is obtained.
  • the offline simulation data is a historical loading scheme obtained by three-dimensional calculation.
  • the fast loading model can quickly obtain the loading ratio of the cargo allocation scheme, and the loading ratio can be directly calculated without the need of three-dimensional calculation.
  • the installation rate of the cargo distribution plan can be quickly obtained, and the time required to obtain the installation rate can be reduced, thereby improving the efficiency of obtaining the target transportation plan.
  • FIG. 1 is a schematic diagram of an application scenario of a method for determining a transportation scheme according to an embodiment of the present application
  • FIG. 2 is a schematic diagram of an embodiment of a method for determining a transportation scheme in an embodiment of the present application
  • FIG. 3 is a schematic diagram of another embodiment of a method for determining a transportation scheme in an embodiment of the present application.
  • FIG. 4 is a schematic diagram of another embodiment of a method for determining a transportation scheme in an embodiment of the present application.
  • FIG. 5 is a schematic diagram of another embodiment of a method for determining a transportation scheme according to an embodiment of the present application.
  • FIG. 6 is a schematic diagram of another embodiment of a method for determining a transportation scheme in an embodiment of the present application.
  • FIG. 7 is a schematic diagram of another embodiment of a method for determining a transportation scheme according to an embodiment of the present application.
  • FIG. 8 is a schematic diagram of another embodiment of a method for determining a transportation scheme according to an embodiment of the present application.
  • FIG. 9 is a schematic diagram of another embodiment of a method for determining a transportation scheme in an embodiment of the present application.
  • FIG. 10 is a schematic diagram of another embodiment of a method for determining a transportation scheme according to an embodiment of the present application.
  • FIG. 11 is a schematic diagram of another embodiment of a method for determining a transportation scheme according to an embodiment of the present application.
  • FIG. 12 is a schematic diagram of another embodiment of a method for determining a transportation scheme in an embodiment of the present application.
  • FIG. 13 is a schematic diagram of another embodiment of a method for determining a transportation scheme according to an embodiment of the present application.
  • FIG. 14 is a schematic diagram of an embodiment of a training prediction model in an embodiment of the present application.
  • 15 is a schematic diagram of another embodiment of a method for determining a transportation scheme in an embodiment of the present application.
  • 16 is a schematic diagram of an embodiment of a three-dimensional loading simulation in an embodiment of the present application
  • FIG. 17 is a schematic diagram of an embodiment of a determining apparatus in an embodiment of the present application.
  • FIG. 18 is a schematic diagram of an embodiment of a training device according to an embodiment of the present application.
  • FIG. 19 is a schematic diagram of another embodiment of a determining apparatus in an embodiment of the present application.
  • FIG. 20 is a schematic diagram of another embodiment of a training device according to an embodiment of the present application.
  • the embodiment of the present application provides a method for determining a transportation scheme, a method and a device for training a quick loading model, and is used for carrying out cargo transportation, especially in a large and complicated transportation scenario, which can quickly obtain a target transportation scheme and reduce transportation costs. Improve transportation efficiency.
  • Container loading simulation is the core problem in the field of logistics. Therefore, the loading simulation of containers needs to achieve high efficiency and accuracy. Efficient means that it can respond quickly, and can output the loading result in a short time through the input data, so it can preferentially seize the logistics resources, shorten the delivery time, and ensure the timely transportation and delivery of the goods. Accuracy means that the output of the output is effective, which can improve the utilization of the container and reduce the transportation cost.
  • the scenario applied in the embodiment of the present application can be as shown in FIG. 1 .
  • the embodiment of the present application only takes two ports, four pick-up points, and two containers as examples. In practical applications, the number of ports, pick-up points, and containers can be Adjust according to actual needs, which is not limited here.
  • the target freight bill is determined.
  • the target freight bill includes the goods to be transported, the pick-up point and the port.
  • the goods to be transported are distributed at the pick-up point D1, the pick-up point D2, the pick-up point D3 and the pick-up point D4, and the goods to be transported need to be transported through the container. Transport to port 2, container at port 1.
  • the number of containers and the path of each container are then determined, as well as the cargo distribution plan and loading plan for each container at the pick-up point.
  • the method for determining a transportation plan provided by the embodiment of the present application can quickly generate an optimal transportation path and a loading plan of the container, and can improve the mounting rate of the container and the efficiency of transportation.
  • the container can be transported by a freight car and one container can be transported by a freight car.
  • the target shipper requires that the goods of D1D2, D3 and D4 be transported to port 2, and the route plan can be determined as follows: Container 1 departs from port 1, passes D1 and D3, transports goods of D1 and D3, and then arrives at port 2, After calculation, the container 1 has a volume mounting rate of 95% and a load loading rate of 96%. Container 2 starts from port 1, passes D2 and D4, transports D2 and D4 goods, and then arrives at port 2, after calculation, container The volumetric loading rate of 2 is 97%, and the load mounting rate is 98%. Therefore, the method for determining the transportation plan provided by the embodiment of the present application can obtain the optimal path for the transportation of the goods, and improve the volume mounting rate and the load mounting rate of the container.
  • a schematic diagram of an embodiment of a method for determining a transportation scheme in the embodiment of the present application includes:
  • the target shipper is a shipper to be transported, and the target shipper includes information of a transport node and a cargo to be transported, the transport node includes a port and M pick-up points, wherein the port may include freight for transporting the goods to be transported
  • the starting point and the ending point of the freight, the starting point of the freight and the ending point of the freight may be the same port or different ports, and the goods to be transported are distributed at each picking point of the M picking points, and the number of the port may be one,
  • the number of the M pick-up points may be one, or may be multiple, and is not limited herein.
  • the goods list to be transported may be obtained by the user input or may be generated by the actual transportation system.
  • each of the at least one path plan may include at least one transport path, that is, a complete path plan composed of a plurality of transport paths, the at least one path plan composing a path plan set, and the path plan set includes at least one Path plan.
  • one transport path can correspond to one container. If multiple containers are needed to complete the goods to be transported, the transport path of multiple containers can be determined. If the goods of one pick-up point cannot be transported through one container, multiple containers can be used.
  • a pickup point can be passed by multiple containers.
  • the demand in the target air waybill is to transport the goods at the pick-up points D1, D2, D3 and D4 to the port 2, the container starts from the port 1, and the super-parameters of the pick-up point transfer can be initialized, and then the hyperparameters are transferred through the pick-up point.
  • the probability of picking up the transfer point, determining the transfer probability between the pick-up points, or between the port and the pick-up point, can obtain multiple transport routes according to the transfer point transfer probability, including from port 1 to D1, D2, D3, D4, to the port 2, or at least one transport path from port 1 to D2, D1, D4, D3, to port 2, etc.
  • the at least one transport path may constitute at least one path plan, the at least one path plan composing a path plan set.
  • the target air waybill includes information of the transport node and the goods to be transported distributed at the M pick-up points
  • the transport node includes a port, M pick-up points, wherein M is a positive integer
  • the port includes a freight start point and a freight end point.
  • the freight origin and the freight destination may be the same port or different ports.
  • a path plan set is obtained, where the path plan set always includes at least one path plan, each path plan of the at least one path plan includes at least one transport path, and one path plan is taken as an example, the one
  • the path plan includes at least one transport path, and all transport paths in the one path plan cover the M pick-up points, and one of the one path plan may cover L pick-up points in the M pick-up points, L ⁇ M.
  • the planning path may be adjusted according to historical path data or preset weights.
  • the transition probability between the port to the pick-up point or the pick-up point may be obtained according to the historical path plan, or may be initialized by random distribution or uniform distribution.
  • the transition probability between the pick-up points is then generated for each transport path of each path plan in the set of path plans based on the transition probability between the pick-up points.
  • the method of the present invention can quickly generate each path plan in the path plan set by using the transition probability between the pick-up points, and improve the efficiency of obtaining the path plan.
  • Each path plan in the set of path plans also needs to be compared in detail to determine the most suitable path plan.
  • the conditions for determining the most suitable path plan may include the length of the path, the mounting rate of the container in the path plan or The difference between the volume mounting rate of the container and the load mounting rate in the path scheme, wherein the shorter the path length, the better the path scheme; the larger the container mounting rate in the path scheme, the better the path scheme; The smaller the difference between the volume mounting rate of the container and the load mounting rate, the better the path plan.
  • At least one path plan is included in the path plan set, so further cargo allocation is required for each path plan of the at least one path plan, and each transport path in each path plan in the path plan set is performed.
  • the goods are distributed, that is, the goods to be transported are allocated to the containers corresponding to each transport path in each path plan, and the set of goods allocation plans of each path plan of the at least one path plan, that is, the first set of goods allocation plans, is obtained.
  • the goods of each delivery point can be clustered before the distribution of goods, and then a probability distribution is used for the same type of goods to determine the distribution mode.
  • the reference conditions for clustering may be the length, width, height, weight, minimum contact area, material, and pressure coefficient of the cargo. To improve efficiency, some conditions may be used for clustering, for example, length, width, height, and Weight is clustered. Among them, the clustering may include precise clustering and fuzzy clustering.
  • clustering by length, width, height, and weight is taken as an example.
  • Accurate clustering may classify goods with four identical features into one class, and the clustering is a small-grained clustering mode. The speed of clustering is also fast. Clustering in this way can improve the accuracy of the distribution of goods.
  • the individual that is, the evaluation operation cost of a path scheme, including increasing the operation time and reducing the efficiency of the operation, is increased. Therefore, when the number of categories of the precise clustering is greater than a preset threshold, clustering operations may also be performed by a clustering algorithm such as K-means clustering, Gaussian mixture model or hierarchical clustering method. Therefore, the goods of each pick-up point can obtain the distribution of each kind by the clustering algorithm, and obtain the distribution of each item in the set of goods allocation scheme corresponding to each path plan in at least one path plan by the distribution of the respective types. Program.
  • the mounting ratio may include a volume mounting ratio With the load mounting rate.
  • the fast loading model is obtained by offline training of offline simulation data, which may be a historical loading scheme calculated by three-dimensional loading.
  • the volume mounting rate is the proportion of the cargo volume occupied by the container allocated to the container
  • the load mounting rate is the ratio of the load weight of the container occupied by the cargo allocated to the container.
  • the feature can be extracted from the historical loading scheme, the offline simulation data is converted into training data in a preset format, and then the training model is used to train the prediction model, and the loading scheme of the transportation product can be predicted by the prediction model.
  • Output mounting rate Compared with the prior art, the three-dimensional loading is used for online calculation.
  • the fast loading model in the embodiment of the present application can quickly output the mounting rate of each of the at least one path solution, and improve the efficiency of obtaining the target transportation solution.
  • each path plan of the at least one path plan needs to be compared and evaluated with the corresponding goods distribution plan, and the target path plan, that is, the at least one path is selected.
  • the specific evaluation method may be that the shorter the route of the route plan is, the better the container installation rate is, and the better the difference between the volume mounting rate of the container and the load mounting rate is, and the better, the different conditions, including the path
  • the difference between the length of the solution, the mounting rate of the container, the container mounting rate and the load mounting rate, and the target route plan and the corresponding target cargo allocation plan, that is, the target transportation plan may be included by the collaborative evaluation.
  • the optimal or sub-optimal transportation plan for the target shipper may be included by the collaborative evaluation.
  • iterative path planning, cargo allocation, fast loading, and integration evaluation can be performed on the path scheme in the path scheme set, after the stopping condition is reached, for example, when the number of iterations reaches a threshold, or an optimal path is obtained. After the program and the goods distribution plan, etc., the target transportation plan is output.
  • the path plan set may be obtained, where the path plan set includes at least one path plan, each of the at least one path plan
  • the path scheme is composed of at least one transport path, that is, at least one transport path constitutes a complete path plan, and then each of the path plans in the path plan set is allocated for goods, and at least one path plan is obtained.
  • the fast loading model is obtained by offline simulation data, and the offline simulation data includes a historical loading solution calculated by three-dimensional loading. Compared with the existing solution, the three-dimensional loading is used to perform an online operation to obtain a loading solution.
  • the foregoing describes the flow of the method for determining the transportation plan in the embodiment of the present application.
  • the method for determining the transportation plan in the embodiment of the present application is described in more detail below. Referring to FIG. 3, the method for determining the transportation plan in the embodiment of the present application is described. A schematic of another embodiment.
  • the process of determining the transportation plan may be: after receiving the target shipping order 301, performing hyperparameter initialization 302, initializing the delivery point transfer hyperparameter, and after the delivery point transfer hyperparameter initialization is completed, according to the delivery point Transferring the hyperparameter to obtain a pick-up point transition probability distribution, and performing path planning 303 according to the pick-up point transition probability distribution, obtaining a path plan set, and then performing cargo allocation 304 on each path plan in the path plan set to obtain a path plan set A collection of goods distribution schemes for each route plan, followed by rapid loading 306, rapid loading of the goods for the route plan and the cargo distribution plan, output of the actual installation rate, followed by individual evaluation 305, integration of the route plan and the cargo distribution plan, and The evaluation further integrates the path plan set and the goods allocation plan to obtain the target path plan and the corresponding target goods allocation plan, that is, the simulation result 309.
  • the fast loading model in the fast loading step is obtained by off-line training of the offline simulation data 308 obtained from the offline simulation 307.
  • steps 303-306 may be repeated to perform repeated path plan exploration and cargo allocation exploration until the stop condition is reached, and the target path plan and the corresponding target goods allocation plan are obtained, or the path plan may be directly
  • One of the path schemes in the set and the corresponding cargo allocation scheme are used as the target route scheme and the corresponding target cargo allocation scheme, which may be adjusted according to actual design requirements, which is not limited herein.
  • a target shipping order 301 is obtained, through which the transportation node and the goods to be transported are known, the transportation node includes a port and a delivery point, and the goods to be transported are distributed at each delivery point, wherein the port may include The starting point of the freight and the ending point of the freight, the starting point of the freight and the ending point of the freight can be a port or a different port.
  • the target shipping order can indicate that the port 1 is the starting point of the freight and will be distributed at the picking point 1 and the picking point 2 Transport cargo to port 2.
  • the pick-up point transfer hyper-parameter initialization is performed, and the pick-up point transition probability distribution is obtained according to the initial pick-up point transfer hyper-parameter, and the pick-up point transition probability distribution includes the probability of the container transferring from the port to the pick-up point or the pick-up point, for example, from The probability of picking point D1 to picking point D2.
  • the specific process of the Bayesian estimation algorithm may be: first, assign a prior distribution to the amount to be estimated, and then combine the experimental data, calculate the posterior distribution according to the Bayesian formula, and then obtain the estimate by the posterior distribution. Estimated value. Therefore, the method for determining the goods distribution plan in the embodiment of the present application can also calculate the delivery point transfer hyperparameter by the Bayesian estimation algorithm.
  • the prior distribution can be obtained from historical data or user experience. In the actual business system, a large amount of historical path data is accumulated, and a large number of samples to be estimated can be extracted from these historical path data, and these samples are used to super The parameters are estimated.
  • the historical path data can be used as a priori data.
  • the prior data can also be adjusted according to the experience of the actual dispatcher.
  • the hyperparameter cannot be estimated, and the heuristic algorithm can be used for estimation.
  • the following can be used in the embodiment of the present application. The estimation algorithm and the heuristic algorithm are described in detail.
  • the transfer mode is an effective transfer mode, that is, the transfer mode is determined to be a valid transfer mode according to the pick-up point in the target shipper.
  • the preset parameter ⁇ obeys the Dirichlet ( ⁇ ) distribution
  • is the hyperparameter
  • the posterior distribution also obeys the Dirichlet distribution. Between the prior distribution and the posterior distribution, only the hyperparameters are changed, so that the calculation of the posterior distribution can be simplified.
  • can be regarded as obeying the Dirichlet ( ⁇ ) distribution.
  • Each historical path data can represent a historical transportation path.
  • the existing historical path data is: Port ⁇ D 1 ⁇ D 3 ⁇ D 6 ⁇ Port, which means that one transportation path is, starting from the port and passing through D 1 , D 3 , D 6 returned to the port.
  • the specific initialization process may include: first selecting a pick-up point or port in the target ship list as the current starting point, and filtering out the historical path data. If there are k pick-up points in the shipper, the k+1 historical path data may be determined. .
  • any picking point in the shipper as the current jump point can form an effective transfer mode, and at least one corresponding effective transfer mode needs to exist in the filtered historical path data, for example,
  • the target freight bill from the port to D 1 , at least one of the selected historical route data includes from the port to D 1 , as shown in Figure 5, assuming that the target freight bill includes the port and two Picking points D 1 and D 4 , the historical path data selected from the port to D 1 or from the port to D 4 are used as effective transfer mode.
  • Historical data path in each window respectively count the number of effective transfer of the pattern occurs, and based on this calculate the corresponding parameter of the multinomial distribution, assuming t windows, t ⁇ N min, a sample may be acquired in the t ⁇ .
  • the specific statistical process can be as shown in FIG. 7 , and the path data of one window is selected, and the historical path data of the window is counted.
  • the current starting point is a port
  • the current jumping point is D 1
  • the statistics are from the port to the D.
  • the number of historical paths of 1 After counting the number of transfers between the port and the pick-up point or the pick-up point, a normalized calculation is performed to obtain a multi-distributed parameter ⁇ .
  • the hyperparametric calculation can be performed by a heuristic algorithm. If there are k pick-up points in the target shipper, taking one of the pick-up points D i as an example, the total volume of the goods is V i , the total weight is W i , the maximum volume that a container can load is V, and the maximum load is W.
  • the specific algorithm flow of the heuristic algorithm may include:
  • the superparameter of the polynomial distribution corresponding to the delivery point D i is:
  • the multi-parameter hyperparameters corresponding to the port are:
  • ⁇ Port ( ⁇ (1 ) ,..., ⁇ (i-1) ,1, ⁇ (i+1) ), the super parameter corresponding to the picking point D i is 1, and the super parameters of the other picking points can be set to ⁇ , and the remaining picking points
  • the corresponding ⁇ can be set to a very small number, for example, 0.00001, 0.0000001, and the like.
  • the method for determining the transportation scheme can use the historical path data to perform hyper-parameter initialization and pass the historical path.
  • the data can get a more accurate path scheme and can improve the efficiency of the subsequent path scheme.
  • the path planning is performed. It is necessary to plan the route of transportation, determine the route plan for transportation and the number of containers.
  • the pick-up point transfer hyper-parameter matrix can be obtained, and the pick-up point transfer hyper-parameter matrix can be used to generate the pick-up point transition probability matrix, including the transition probability distribution between the port and the pick-up point, or between the pick-up points.
  • the probability that the transfer probability is transferred from one port to one pick-up point or from one pick-up point to another pick-up point can be obtained based on historical path data, for example, as shown in Fig.
  • the embodiment of the present application can obtain the estimated value of the transition probability matrix by using the Dirichlet distribution to obtain the expected value, and obtain the transition probability matrix by taking the expected value.
  • the weight of the transition probability in the path may be increased, and the transition probability may be set by itself to increase the weight of the transition probability in the path.
  • a path plan set that is, a path plan population
  • the transition probability matrix can be used to obtain the probability of the transfer between the port and the pick-up point or the pick-up point, and then the transfer probability matrix is used to determine the path plan set.
  • the current starting point is the port.
  • step 1 is performed.
  • the next jump point is selected from the port.
  • the sampling probability is 0.23, which is determined as the delivery point A.
  • the subsequent delivery point determination method is similar to the determination of the delivery point A.
  • step 2 and step 3 the picking point C and the port are determined in turn, and the path plan is obtained: port ⁇ picking point A ⁇ picking point C ⁇ port.
  • the probability P i of each party path scheme being selected can be calculated, i represents the i-th individual, that is, a path scheme, and each path scheme in at least one path scheme can be evaluated by the evaluation function, and the individual probability is calculated.
  • the formula can be:
  • j represents the jth individual
  • n represents the total number of individuals
  • f(i) is the score of the individual.
  • the individual score function is an evaluation function in the individual evaluation step, which will be described in detail in the individual evaluation step of step 305. No explanation is given. Therefore, the calculation formula of the individual probability shows that the probability that the path scheme is selected is related to the evaluation of the path scheme. The higher the path scheme evaluation, the greater the probability of being selected, which can be understood as the higher the evaluation of the path scheme. The better the plan.
  • the path plan set is updated, and in addition, the hyperparameter matrix may be updated in addition to the updated path plan set, so as to be improved afterwards
  • the path planning can be continued through the updated hyperparameter matrix to obtain a more accurate path scheme.
  • the transfer mode of the picking point A ⁇ the picking point B of m individuals appears 1 Secondly, the value corresponding to the transfer from the pickup point A to the delivery point B in the corresponding hyperparametric matrix is increased by 1, the original value is 0.6, then the increased value is 1.6, and the other transfer calculations are analogous.
  • the picking point transition probability matrix can be obtained through the hyperparametric matrix, and then the path planning is performed again.
  • the pick-up point transition probability matrix is obtained according to the initialized pick-up point transfer hyper-parameter, and the first path plan set is determined according to the pick-up point transition probability matrix, where the first path plan set includes at least one a path plan, and then performing, according to each path plan in the first path plan set, each cargo allocation plan in the set of goods allocation plans of each path plan in the first path plan set, and then Each path plan in the first path plan set and each path plan in the first path plan set corresponding to each path plan in the first path plan set are integrated and evaluated, and the result of the integrated evaluation is not obtained properly.
  • the picking point transfer hyperparameter may be updated according to each path scheme in the first path scheme set, and the updated picking point transition probability matrix is obtained through the updated picking point transfer hyperparameter, Re-transport the transport path according to the updated pick-up point transfer probability matrix Stroke, to obtain a second set of program path, the second path includes at least a set of program paths embodiment, after the re-allocation scheme and an integrated evaluation scheme set goods second path each path, to obtain the target transport program.
  • each of the path plans in the path plan set may be allocated for goods, in each path plan in the path plan set.
  • Each transportation route carries out the distribution of goods to obtain the goods loaded in the container in each transportation.
  • a path plan set can be seen as a population. Each individual in the population represents a path plan. When the individual is generated, each transport path in the path plan is confirmed, so the required number of containers and each container The path has been determined. In practical applications, it can be seen that the container has a one-to-one correspondence with the transportation route, and the container and the freight vehicle correspond one-to-one. Then, at this time, it is necessary to distribute the goods loaded into the container to determine the goods loaded in each container.
  • the goods can be clustered, and then the goods are distributed by clustering.
  • the reference conditions for clustering may be the length, width, height, weight, minimum contact area, material, and pressure coefficient of the cargo.
  • some conditions may be used for clustering, for example, length, width, height, and Weight is clustered.
  • the clustering may include precise clustering and fuzzy clustering. In the embodiment of the present application, clustering by length, width, height, and weight is taken as an example.
  • Accurate clustering may classify goods with four identical features into one class, and the clustering is a small-grained clustering mode.
  • the speed of clustering is also fast. Clustering in this way can improve the accuracy of the distribution of goods.
  • the clustering algorithm may be performed by a clustering algorithm such as K-means clustering, Gaussian mixture model or hierarchical clustering method.
  • the goods are classified by the method of clustering, and the goods are distributed after the classification, which can improve the accuracy of the subsequent loading scheme of the goods.
  • a schematic diagram of the distribution of goods can be as shown in FIG. 11, wherein each pickup point has goods waiting to be transported, for example, the type of goods to be transported in A in FIG. 11 includes goods 1, goods 2 or goods m.
  • A corresponding to the first cargo distribution superparameters ⁇ A1 , ⁇ A2 and Then, through sampling, the goods to be transported are sorted for each cargo pick-up point, and a category population representing a set of goods distribution modes is generated for each pick-up point, and the individual in the category of the stock represents the manner of distribution of the goods in the pick-up point.
  • picking point A has two containers passing through, in Fig.
  • the individual in stock 1 indicates that the goods in picking point A are assigned to the two containers, that is, the goods of each category are on each container.
  • the number of allocations Since a stock can only represent the distribution method of goods at a pick-up point, it cannot represent a complete distribution plan. Therefore, it is necessary to combine the goods distribution methods of all pick-up points to form a complete plan before evaluating the plan, that is, collaborative evaluation.
  • the representative individuals may be randomly selected in the population, or may be the optimal individuals in the population, or may be the optimal individuals determined after random selection and multiple integrations.
  • the item allocation plan can be evaluated by an individual evaluation, that is, the evaluation function in step 305, and the score of each path plan in the set of goods allocation plan is calculated according to the evaluation function, and the evaluation function will be described in the detailed description of step 305 below. It is not explained here.
  • the cargo allocation scheme higher than the second threshold may update the first cargo allocation hyperparameter of the picking point through the obtained first cargo allocation scheme set to obtain the second cargo allocation hyperparameter of the picking point, and pass the second cargo Allocating hyper-parameters to re-allocate the goods to be transported, and obtaining a second set of goods allocation scheme corresponding to the path plan, and continuing to perform integrated evaluation on each of the goods distribution plans in the second set of goods distribution plan to determine the target goods distribution Scheme and corresponding target path scheme.
  • the goods distribution may be repeated until the stopping condition is reached, for example, the number of iterations reaches a preset number of times, and the result of the integrated evaluation is higher than the second threshold, and the number of the goods distribution plan reaches the preset number. Stop after determining the target cargo allocation plan.
  • the algorithm for calculating the individual selection probability is performed by using an algorithm similar to that in the path planning step 303, and then n individuals are selected according to the individual selection probability. And update the original pick-up point hyper-parameter matrix of each pick-up point to obtain the super-parameter matrix of the target pick-up point. That is, after determining the goods distribution plan of each pick-up point, a process of further learning may be included, which may be used to repeatedly distribute the goods.
  • the specific process of updating the cargo allocation hyper-parameters may be as shown in FIG.
  • the label 1 is the same category
  • the label 2 is the same category
  • the label 3 is the same category
  • the category 1 is the category 1 in the first container, that is, the vehicle 1 is assigned 2 , the corresponding car 1, class 1 hyperparameters plus 2, the original pick-up point hyper-parameter matrix in the car 1, class 1 super-parameter is 1.0, then add 2 after the target pick-up point hyper-parameter matrix in the car 1, Class 1 has a hyperparameter of 3.0, other categories, and container hyperparameters and so on.
  • the individual evaluation integrates the path plan and the goods distribution plan, and evaluates the path plan and the goods distribution plan.
  • the evaluation indicators include: the path length of the path plan, the container mounting rate, the difference between the container's volume mounting rate and the load mounting rate.
  • the transportation cost can be saved and the transportation efficiency can be improved; the larger the container mounting rate, that is, the more goods loaded in the container, that is, the same.
  • the container loading rate can be obtained by fast loading, step 306.
  • the container with unbalanced mounting rate can be obtained, and the volume mounting rate and the load mounting rate are distributed on the average volume mounting rate of all the goods on the freight list. Both sides that are suitable for the average load mounting rate, ie among them,
  • r Vi is the volume mounting rate of the i-th container
  • r Wi is the load mounting rate of the i-th container.
  • the embodiment of the present application provides an evaluation function that compares the length of the evaluation path with the mounting rate of the container, and also compares the difference between the volume mounting rate and the load mounting rate.
  • the evaluation function is: among them, For the path plan vector, including m containers, For the volumetric loading rate vector of the m containers, For the load-carrying rate vector of the m containers; ⁇ , ⁇ and ⁇ are weight parameters, r Vi is the volume mounting rate of the i-th container, and r Wi is the load-carrying rate of the i-th container, For the average volume mounting rate of the m containers, The average load mounting rate for the m containers.
  • the path scheme can be evaluated by the evaluation function, and some path schemes with invalid or low evaluation are selected, and the path scheme set is updated.
  • the fast loading simulation of each path plan in the path plan set may be performed by the fast load model to obtain the mounting rate of each path plan in each of the route allocation plan sets corresponding to each path plan.
  • the distribution plan of the goods obtained by the distribution of goods that is, the collection of goods loaded into the container
  • the installation rate of the container can be quickly obtained.
  • the fast loading model is obtained according to the offline simulation data 308, which includes a historical loading scheme obtained by three-dimensional loading
  • the offline simulation data can be obtained by offline simulation 307, and the step of offline simulation 307 is determined in the embodiment of the present application.
  • the steps for the target transportation plan are similar.
  • the mounting rate of each cargo allocation corresponding to each path scheme in at least one path scheme is obtained, and the mounting ratio may also be based on the mounting ratio. Estimating the cargo loaded into the container, and obtaining a feasible solution, that is, judging whether each cargo can be loaded into the container. If there is any remaining cargo that cannot be loaded into the container, further cargo distribution is required for the remaining cargo to be able to be transported. The target transportation plan for the complete transportation of the goods.
  • step 304, step 305 and step 306 together constitute a step of exploring the goods distribution plan
  • the goods distribution plan exploring step 303 is composed of the path plan exploration step, and determining the target transportation plan, that is, the target transportation path corresponding to the target transportation path
  • the target transportation plan is output, and the department transports the goods according to the target transportation plan.
  • the target path plan and the corresponding target goods distribution plan may be obtained by repeating steps 303-306, and the goods to be transported are transported through the target path plan and the corresponding target goods distribution plan.
  • the specific acquisition process of the fast loading model and the loading scheme may be divided into an offline training part and an online prediction part as shown in FIG. 14 .
  • the offline training part is explained. Specifically, as shown in FIG. 14, a large amount of high-quality offline simulation data is obtained through offline simulation, and the feature extraction in the offline simulation data, that is, the container cargo allocation scheme is converted into a set of feature vectors.
  • the prediction model is trained from the offline simulation data, and the prediction model is used to output the volume mounting rate and the load mounting rate of the input data.
  • the prediction model may include: a linear regression model, a ridge regression model, a LASSO model, a support vector machine model, a random forest model, an XgBoost model, or an artificial neural network model.
  • Offline training is similar to online feature extraction. The difference is that offline training extracts features from offline simulation data.
  • Online prediction is from the collection of goods assigned to the container, ie from each Features are extracted from the cargo allocation plan.
  • the specific process of extracting features includes first extracting features of a single cargo to obtain a feature vector of a single cargo, ie, a second feature vector, the characteristics of the single cargo including: length, width, height, and weight of the cargo, and may also include minimum contact Area, material, pressure coefficient, etc.
  • the material and the pressure coefficient are categorical variables, that is, related to the category of the goods, and the category dimension is not large.
  • the one-hot coding method can be used to indicate the material and the bearing coefficient.
  • the container passes through each pick-up point.
  • a final feature vector that is, a first feature vector of each cargo allocation scheme in the first cargo allocation scheme set corresponding to each path scheme in the at least one path scheme
  • n is the number of pick-up points
  • r i is the quantized value of the order of the pick-up point i in the path plan
  • r i and t i are data based on the historical path, which are analyzed by the corresponding analysis method.
  • each historical loading plan data in the offline simulation data is converted into training data in a preset format, which may be (feature vector, mounting rate), and then model training is performed to obtain fast Load the model.
  • Two models need to be trained, including a model that predicts the volumetric loading rate and a model that predicts the loading rate of the load.
  • the fast loading model includes a model that predicts the volumetric loading rate and a model that predicts the loading rate of the load.
  • the set of goods allocated to the container is converted into a feature vector of a preset format, and then input into the trained model to obtain a corresponding output value.
  • the goods allocated to the container may not be loaded into the container, and when the hyperparameter is updated, only the part loaded into the container can be considered, and the super parameter can be updated by the quantity of the loaded goods.
  • the goods distribution plan determined at the time of goods distribution includes the goods collection loaded into the container, and the mounting rate prediction model can only predict the mounting rate of the goods collection, and cannot determine whether the goods in the goods collection can be loaded into the container. Therefore, it is necessary to estimate the set of goods that can be loaded into the container based on the information of the mounting rate.
  • linear programming can be used to achieve the problem.
  • the volume of the i-th cargo is v i
  • the weight is w i
  • p i is the probability that the i-th cargo can be loaded into the container
  • the volume output by the mounting rate prediction model is The loading rate is r v
  • the load mounting rate is r w
  • the predicted volume and weight of the cargo loaded into the container are V and W, respectively.
  • the linear programming can be defined as:
  • Solving the above equation can be obtained a set of p i, set sorted in descending order of p i of goods in the goods, and then order goods taken sequentially, to know the total volume of goods removed more than V, or the total weight of more than W, then
  • the given set of goods is an estimate of a feasible solution, that is, the goods that can be loaded into the container.
  • the post-processing can also be performed, that is, after determining the target transportation scheme, the appropriate container model can be further determined and the final loading scheme can be generated.
  • the box type in which the container can be used may be 40HQ, and the volume and weight of the cargo loaded in each container can be obtained by using the cargo distribution exploration and path scheme exploration in FIG. 3, and the volume and load of the box type that can be selected cannot be smaller than the cargo. Volume and weight.
  • the three-dimensional loading can also be used to determine the final loading plan of the goods to be transported in the container, so that all the goods can be accurately loaded into the container, and the goods are loaded when the actual loading of the goods is carried out.
  • the container type of the container can be selected to be successfully loaded and the lowest cost model to reduce the cost of the container.
  • the appropriate box type and final loading plan are determined as a complete plan. If there is any remaining goods, there is goods that cannot be loaded into the container, then the remaining goods can be re-reloaded. Generate a virtual shipper, repeat the exploration of the cargo distribution plan, and obtain the transportation plan for the remaining goods to complete the transportation of the remaining goods.
  • the offline simulation data of the training fast loading model can be obtained by offline simulation, and the specific process of offline simulation is similar to the online simulation.
  • the difference includes offline simulation using historical freight bill for simulation, and online simulation using current shipping order.
  • FIG. 15 is a schematic diagram of another embodiment of a method for determining a transportation scheme in the embodiment of the present application.
  • the steps of path planning, goods distribution, and individual evaluation of the historical freight bill need to be re-executed.
  • a historical freight bill which includes historical pick-up point information and information of historical goods to be transported
  • path planning according to the historical freight bill to obtain a historical path corresponding to the historical freight bill.
  • the solution, and then according to the historical path scheme the goods are allocated to obtain a historical cargo allocation plan set corresponding to each historical path plan, and then obtained by the three-dimensional loading operation to each cargo allocation plan in the historical goods distribution plan set corresponding to the path plan.
  • the mounting rate and the loading mode, and then the integrated evaluation of each of the at least one historical path plan and each of the corresponding historical goods allocation plan sets according to the mounting rate of each goods distribution plan get a historical transportation plan.
  • the steps in the offline simulation include the steps of hyperparameter initialization, path planning, cargo allocation, and individual evaluation are similar to the steps of the hyperparameter initialization, path planning, cargo allocation, and individual evaluation of the online simulation in FIG. 3, specifically No longer.
  • the following describes the different steps of offline simulation and online simulation.
  • the three-dimensional loading algorithm can be used to calculate the mounting rate, that is, the specific loading mode is obtained.
  • the volumetric loading rate and the load mounting rate of the container can be obtained by using the three-dimensional loading operation, and the volume mounting rate and the load mounting rate can be used.
  • the scheme is evaluated, and the step of evaluating the path scheme is similar to the individual evaluation step 305 in FIG. 3 described above.
  • the embodiment of the present application can use the heuristic algorithm based on Corner Point and Extreme Point to complete the simulated loading of the cargo. As shown in FIG.
  • the spatial state of the container is first determined, and then a series of candidate placement points are obtained. Then try to place them one by one until you find the right placement point.
  • the Extreme Point algorithm scans the area where the goods are overhead, so it will generate more candidate points, and can obtain accurate mounting rate and loading scheme, which can improve the container usage rate.
  • the embodiment of the present application is offline.
  • Simulation and post-processing use 3D loading simulation, which can improve the efficiency of the cargo loading scheme and the output mounting rate.
  • the 3D loading algorithm is used to calculate the mounting rate, which can obtain a more accurate mounting rate.
  • the three-dimensional loading algorithm is used to obtain the loading plan during processing, and the loading mode of the goods can be known during transportation to improve the transportation efficiency.
  • the method for determining the transportation path provided by the embodiment of the present application is described in detail.
  • the device provided in the embodiment of the present application is described below.
  • the determining device is described. Referring to FIG. 17, the determining device may include:
  • the obtaining module 1701 is configured to acquire at least one path plan and a first set of goods allocation plans corresponding to each of the at least one path plan, each of the at least one path plan is for transporting goods to be transported And the planned transportation route, the first cargo allocation plan set corresponding to each of the at least one path plan includes at least one cargo allocation plan, and the first cargo allocation plan corresponding to each of the at least one path plan
  • Each of the goods distribution plans in the set is a plan for allocating the goods to be transported for the corresponding path plan;
  • a fast loading module 1702 configured to determine, according to the fast loading model, a mounting rate of each of the first item allocation sets corresponding to each of the at least one path plan, the fast loading model is offline simulation
  • the data is obtained by offline training, and the offline simulation data includes a historical loading plan calculated by a three-dimensional loading algorithm, and the mounting ratio is a ratio of a container loaded with a container occupying the container in a certain cargo allocation scheme;
  • the evaluation module 1703 integrates each of the at least one path plan and each of the first item allocation plan corresponding to each of the at least one path plan by the mounting rate
  • the evaluation determines a target transportation plan, wherein the target transportation plan includes a target goods allocation plan corresponding to the target path plan and the target path plan.
  • the obtaining module 1701 may include:
  • the obtaining sub-module 17011 is configured to obtain a target shipping slip, where the target shipping manifest includes transportation node information and cargo information to be transported, the transportation node information includes a shipping origin, a shipping destination, and M picking points, and the information of the to-be-sold goods is distributed Information of the goods to be transported at the M pick-up points, the M being a positive integer;
  • each of the at least one path plan includes at least one transport path, each of the at least one transport path includes a freight a start point, a freight destination, and N pick-up points in the M pick-up points, each path plan of the at least one path plan covers the M pick-up points, the N is a positive integer, and N ⁇ M;
  • a goods distribution sub-module 17013 configured to allocate the goods to be transported for each of the at least one path plan to obtain a corresponding one of each path plan of the at least one path plan A distribution plan for each item in a collection of goods distribution plans.
  • the path planning sub-module 17012 is specifically configured to:
  • the determining apparatus may further include:
  • the initialization module 1704 is configured to initialize the transfer hyper-parameters of the M pick-up points based on the heuristic algorithm to obtain the super-parameter matrix if the number of the historical path data is not greater than the first threshold.
  • the goods distribution sub-module 17013 is specifically configured to:
  • the goods of each of the M pick-up points are clustered according to the clustering conditions to obtain a clustering result, which includes the length, width, height and weight of the goods;
  • the clustering result is sampled and calculated by using the first cargo allocation hyperparameter of each of the M picking points to obtain a first set of goods allocation manners of each of the M picking points, the M
  • the first item allocation hyperparameter of each picking point in the picking point is a hyper parameter for allocating goods to each picking point of the M picking points, and the first item allocation mode set of each picking point of the M picking points
  • Each of the goods is allocated in such a way as to distribute the goods distributed at the pick-up point for the corresponding route plan;
  • the fast load module 1702 is specifically configured to:
  • the first feature vector is used to indicate the goods to be transported in a certain goods distribution plan Characteristic value
  • the volume of the cargo allocated by each transport path in the path plan occupies a proportion of the load volume of the container
  • the load mounting rate is the weight of the cargo allocated by each of the at least one of the at least one route plan The proportion of the load on the container.
  • the fast load module 1702 is specifically configured to:
  • the second feature vector of each of the goods to be transported includes a length, a width, a height, and a weight of the corresponding goods;
  • a third feature vector of each of the goods distribution plans includes each of the goods to be transported The mean and covariance of the second eigenvector of the goods;
  • the evaluation module 1703 is specifically configured to:
  • the all-carriage allocation plan includes a goods distribution plan with a score higher than a second threshold, determining the target goods distribution plan from the goods distribution plan whose score is higher than the second threshold, and corresponding to the target goods distribution plan a path plan as the target path plan;
  • the target transportation plan is determined by the target goods distribution plan and the target path plan.
  • the evaluation function includes:
  • the For the path plan vector m is the number of containers, The volumetric loading rate vector for m containers, The loading rate vector of the m containers; the ⁇ , the ⁇ and the ⁇ are weighting parameters, the r Vi is the volume mounting rate of the i-th container, and the r Wi is the load of the i-th container Rate, the rate For the average volume mounting rate of the m containers, The average load mounting rate for the m containers.
  • the evaluation module 1703 is further configured to:
  • the clustering result is sampled and calculated by the second cargo allocation hyperparameter of each of the M picking points, to Obtaining a second set of goods distribution manners of each of the M pick-up points, and each of the items of the second pick-up mode of each of the M pick-up points is allocated for the corresponding path plan pair
  • the manner in which the goods distributed at the pick-up point are distributed, and the second item allocation hyper-parameter of each of the M pick-up points is in the first set of goods allocation schemes corresponding to each of the at least one path plan
  • Each cargo allocation plan updates the first cargo allocation hyperparameter of each of the M picking points;
  • each of the second goods distribution plan sets corresponding to each of the at least one path plan is a plan for allocating the goods to be transported for the corresponding path plan;
  • the determining apparatus may further include:
  • a post-processing module 1705 configured to perform an integrated evaluation of each of the at least one path plan and the goods allocation plan of each of the at least one path plan by the mounting rate to determine the target transportation plan Determining, according to the target cargo allocation plan and the target route plan, a model of a container of each transport path in the target route plan;
  • the three-dimensional loading module 1706 is configured to generate a loading plan according to the model of the container of each transportation path in the target path solution determined by the post-processing module 1705, and the loading plan is that the goods to be transported are in the target path plan.
  • the determining apparatus may further include:
  • a determining module 1707 in each of the first item allocation plan sets corresponding to each of the at least one path plan and each of the at least one path plan by the mounting rate
  • the remaining goods path is determined for the remaining goods.
  • the evaluation module 1703 is further configured to use, by the mounting rate, each of the at least one path plan and each of the corresponding first item allocation plan sets, and the remaining item path plan and the The remaining goods distribution plan is integrated and evaluated to determine the target transportation plan.
  • a schematic diagram of an embodiment of the training device in the embodiment of the present application may include:
  • the obtaining module 1801 is configured to acquire offline simulation data, where the offline simulation data includes a historical loading plan and a historical mounting rate calculated by three-dimensional loading;
  • the obtaining module 1801 is further configured to acquire a feature vector from the offline simulation data, where the feature vector includes a feature value of the historical transport cargo corresponding to the historical loading solution;
  • a conversion module 1802 configured to convert the feature vector into training data in a preset format
  • a training module 1803 configured to train the prediction model by using the training data to obtain a fast loading model, which is used to output a mounting rate of each cargo allocation scheme in the collection of goods allocation schemes of each transportation path, the mounting The rate at which the goods loaded into the container in each of the goods distribution schemes occupy the container.
  • the preset format is: (feature vector, historical mounting rate).
  • the prediction model includes: a linear regression model, a ridge regression model, a LASSO model, a support vector machine model, a random forest model, an XgBoost model, or an artificial neural network model.
  • the obtaining module 1801 may include:
  • the obtaining sub-module 18011 is configured to acquire at least one historical path solution and a first historical cargo allocation plan set corresponding to each of the at least one historical path solution, where each path plan of the at least one historical path solution is a transportation route planned for transportation of historical transportation goods, the first historical cargo distribution plan set corresponding to each of the at least one historical path plan includes at least one cargo distribution plan, each path of the at least one historical path plan Each of the first historical goods allocation plan sets corresponding to the plan is a plan for allocating the historical transport goods for the corresponding path plan;
  • a three-dimensional loading sub-module 18012 configured to determine, according to the three-dimensional loading algorithm, a mounting rate of each of the first item allocation sets corresponding to each of the at least one historical path plan, the mounting rate is a certain The proportion of goods loaded into the container in a cargo distribution plan that occupies the container;
  • the evaluation sub-module 18013 by the mounting rate, each of the at least one historical path plan and each of the first historical goods distribution plan sets corresponding to each of the at least one historical path plan
  • the distribution plan performs an integrated evaluation to determine a target transportation plan, wherein the target transportation plan includes a target route allocation plan corresponding to the target path plan and the target path plan.
  • the three-dimensional loading algorithm can be used for calculation when training the fast loading model, and the historical loading scheme corresponding to the historical path scheme is obtained, and the mounting rate of the cargo allocation scheme corresponding to the historical path scheme can be accurately output.
  • the obtaining submodule 18011 includes:
  • the obtaining unit 180111 is configured to acquire a historical shipping slip, the historical shipping bill includes transportation node information and a historical transportation cargo information, where the transportation node information includes a shipping origin, a shipping destination, and M picking points, where the historical shipping information includes Information of the historical transport cargo of M pick-up points, the M being a positive integer;
  • the path planning unit 180112 is configured to determine, according to the transportation node information, the at least one historical path solution, each of the at least one historical path solution includes at least one transportation path, and each of the at least one transportation path includes a shipping origin, a shipping destination, and N picking points in the M picking points, each path plan of the at least one historical path plan covers the M picking points, the N is a positive integer, and N ⁇ M;
  • a goods distribution unit 180113 configured to perform allocation of the historical transportation goods for each of the at least one historical path solution to obtain a corresponding path plan for each of the at least one historical path solution Each cargo allocation plan in the first historical cargo allocation plan set.
  • the path planning unit 180112 is specifically configured to:
  • the training device further includes:
  • the initialization module 1804 is configured to initialize the transfer hyper-parameters of the M pick-up points based on the heuristic algorithm to obtain the super-parameter matrix if the number of the historical path data is not greater than the first threshold.
  • the goods distribution unit 180113 is specifically configured to:
  • the goods of each of the M pick-up points are clustered according to the clustering conditions to obtain a clustering result, which includes the length, width, height and weight of the goods;
  • the clustering result is sampled and calculated by using the first cargo allocation hyperparameter of each of the M picking points to obtain a first set of goods allocation manners of each of the M picking points, the M
  • the first item allocation hyperparameter of each picking point in the picking point is a hyper parameter for allocating goods to each picking point of the M picking points, and the first item allocation mode set of each picking point of the M picking points
  • Each of the goods is allocated in such a way as to distribute the goods distributed at the pick-up point for the corresponding route plan;
  • the evaluation sub-module 18013 is specifically configured to:
  • the all-carriage allocation plan includes a goods distribution plan with a score higher than a second threshold, determining the target goods distribution plan from the goods distribution plan whose score is higher than the second threshold, and corresponding to the target goods distribution plan a path plan as the target path plan;
  • the target transportation plan is determined by the target goods distribution plan and the target path plan.
  • the evaluation function includes:
  • the For the path plan vector m is the number of containers, The volumetric loading rate vector for m containers, The loading rate vector of the m containers; the ⁇ , the ⁇ and the ⁇ are weighting parameters, the r Vi is the volume mounting rate of the i-th container, and the r Wi is the load of the i-th container Rate, the rate For the average volume mounting rate of the m containers, The average load mounting rate for the m containers.
  • the evaluation sub-module 18013 is further configured to:
  • the clustering result is sampled and calculated by the second cargo allocation hyperparameter of each of the M picking points, to Obtaining a second set of goods distribution manners of each of the M pick-up points, and each of the items of the second pick-up mode of each of the M pick-up points is allocated for the corresponding path plan pair
  • the manner in which the goods distributed at the pick-up point are distributed, and the second item allocation hyper-parameter of each of the M pick-up points is a first historical item allocation plan corresponding to each path plan in the at least one historical path plan
  • Each cargo allocation plan in the set updates the first cargo allocation hyperparameter of each of the M picking points;
  • each of the second historical goods distribution plan sets corresponding to each of the at least one historical path plan is a distribution of the historical transportation goods for the corresponding path plan Program;
  • a path rate for each of the at least one historical path plan by the evaluation function and a mounting rate of each of the second historical goods allocation plan sets of each of the at least one historical path plan a score calculation for each of the cargo allocation schemes in the second historical cargo allocation plan set, and a mounting rate of each of the cargo distribution plans in the second historical cargo allocation plan set of each of the at least one historical route plans Obtained from the three-dimensional loading submodel.
  • the path plan may be re-planned, or the goods may be re-allocated directly through the at least one historical path plan.
  • the training device further includes:
  • each of the first historical goods allocation plan set corresponding to each of the at least one historical path plan and each of the at least one historical path plan by the mounting rate The cargo distribution plan performs an integrated evaluation to determine the target transportation plan, and if the actual installation rate determines that the L pick-up points in the M pick-up points further include the remaining goods not allocated to the container, the remaining goods are determined
  • the evaluation sub-module 18013 is further configured to use, by the mounting ratio, each of the at least one historical path plan and each of the corresponding first historical goods distribution plan sets, and the remaining goods path
  • the programme and the remaining cargo allocation plan are integrated and evaluated to determine the target transportation plan.
  • FIG. 19 is a schematic structural diagram of a determining apparatus according to an embodiment of the present application.
  • the determining apparatus 1900 may generate a large difference due to different configurations or performances, and may include one or more central processing units (CPUs) 1922. (eg, one or more processors) and memory 1932, one or more storage media 1930 that store application 1942 or data 1944 (eg, one or one storage device in Shanghai).
  • the memory 1932 and the storage medium 1930 may be short-term storage or persistent storage.
  • the program stored on storage medium 1930 may include one or more modules (not shown), each of which may include a series of instruction operations in the determining device.
  • the central processor 1922 can be configured to communicate with the storage medium 1930 on which a series of instruction operations in the storage medium 1930 are performed.
  • the determining device 1900 can also include one or more power sources 1926, one or more wired or wireless network interfaces 1950, one or more input and output interfaces 1958, and/or one or more operating systems 1941, such as Windows ServerTM, Mac. OS XTM, UnixTM, LinuxTM, FreeBSDTM and more.
  • operating systems 1941 such as Windows ServerTM, Mac. OS XTM, UnixTM, LinuxTM, FreeBSDTM and more.
  • FIG. 20 is a schematic structural diagram of a training apparatus according to an embodiment of the present application.
  • the training apparatus 2000 may generate a large difference due to different configurations or performances, and may include one or more central processing units (CPUs) 2022. (eg, one or more processors) and memory 2032, one or more storage media 2030 that store application 2042 or data 2044 (eg, one or one storage device in Shanghai).
  • the memory 2032 and the storage medium 2030 may be short-term storage or persistent storage.
  • the program stored on storage medium 2030 may include one or more modules (not shown), each of which may include a series of instruction operations in the training device.
  • central processor 2022 can be arranged to communicate with storage medium 2030 to perform a series of instruction operations in storage medium 2030 on training device 2000.
  • Training device 2000 may also include one or more power sources 2026, one or more wired or wireless network interfaces 2050, one or more input and output interfaces 2058, and/or one or more operating systems 2041, such as Windows ServerTM, Mac. OS XTM, UnixTM, LinuxTM, FreeBSDTM and more.
  • operating systems 2041 such as Windows ServerTM, Mac. OS XTM, UnixTM, LinuxTM, FreeBSDTM and more.
  • the disclosed system, apparatus, and method may be implemented in other manners.
  • the device embodiments described above are merely illustrative.
  • the division of the unit is only a logical function division.
  • there may be another division manner for example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be in an electrical, mechanical or other form.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium.
  • a computer readable storage medium A number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of Figures 2 through 16 of various embodiments of the present application.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Operations Research (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Quality & Reliability (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Game Theory and Decision Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Hardware Design (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Geometry (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)

Abstract

本申请实施例提供一种确定运输方案的方法、训练快速装载模型的方法及设备,用于得到货物运输方案,特别在运输量大、复杂的场景下,可以快速得到目标运输方案,减小运输成本,提高运输效率。确定运输方案的方法包括:获取至少一个路径方案以及每个路径方案对应的第一货物分配方案集合,该每个路径方案为针对待运输货物进行运输而规划的运输路径,一个第一货物分配方案集合包括至少一个货物分配方案;根据快速装载模型确定每个货物分配方案的实装率,该快速装载模型为通过包括三维装载算法计算得到的历史装载方案的离线仿真数据进行离线训练得到;通过实装率对每个路径方案以及对应的货物分配方案进行整合评价,以确定目标运输方案。

Description

确定运输方案的方法、训练快速装载模型的方法及设备
本申请要求于2018年02月06日提交中国专利局、申请号为201810118531.9、申请名称为“确定运输方案的方法、训练快速装载模型的方法及设备”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及物流领域,特别涉及一种确定运输方案的方法、训练快速装载模型的方法及设备。
背景技术
随着经济的发展,运输行业也越来越追求高效与准确,集装箱更是运输的主要容器。随着提货点以及需要运输的货物越来越多,为提高运输的效率,需要对运输的路径进行规划。
在现有方案中,通过蚁群算法得到路径方案,然后通过三维装载仿真得到路径方案中货物的装载方案,即货物在集装箱内的装载方式。但三维装载仿真无法并行处理货物,需要进行大量运算,耗费大量时间,特别是在货物量较大的场景下,耗费更大量的时间,因此降低了得到货物装载方案,输出实装率的效率,进而影响得到目标运输方案的效率。
发明内容
本申请实施例提供一种确定运输方案的方法、训练快速装载模型的方法及设备,用于进行货物运输,特别在运输量大、复杂的场景下,可以快速得到目标运输方案,减小运输成本,提高运输效率。
有鉴于此,本申请第一方面提供一种确定运输方案的方法,可以包括:
首先获取至少一个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合,该至少一个路径方案中的每个路径方案为针对待运输货物进行运输而规划的运输路径,其中,一个路径方案可以包括至少一条运输路径,该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合包括至少一个货物分配方案,该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该待运输货物进行分配的方案;根据快速装载模型确定该至少一个路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,该快速装载模型为通过离线仿真数据进行离线训练得到,该离线仿真数据包括通过三维装载算法计算得到的历史装载方案,该实装率为某一货物分配方案中装入集装箱的货物占用该集装箱的比例;通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,该目标运输方案包括目标路径方案与该目标路径方案对应的目标货物分配方案。
在本申请实施例中,在确定至少一个路径方案以及该至少一个路径方案中每个路径方案对应的第一货物分配方案集合后,可以根据通过离线仿真数据进行离线训练的快速装载 模型确定每个路径方案对应的第一货物分配方案集合中每个货物分配方案的实装率,可以快速得到每个路径方案对应的所有货物分配方案的实装率,可以降低获取每个路径方案对应的所有货物分配方案的实装率的时长,提高确定目标运输方案的效率。其中,快速装载模型由通过离线仿真数据进行离线仿真训练得到,该离线仿真数据包括由三维装载运算得到数据,可以提高得到的实装率的准确性。
结合本申请第一方面,在本申请第一方面的第一种实施方式中,获取至少一个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合,可以包括:
首先获取目标货运单,该目标货运单包括运输节点信息以及待运输货物信息,该运输节点信息包括货运起点、货运终点以及M个提货点,该待运输货物信息包括分布在该M个提货点的该待运输货物的信息,该M为正整数;然后根据该运输节点信息中的运输节点确定该至少一个路径方案,其中,一个路径方案可以包括至少一条运输路径,该至少一条运输路径中的每条运输路径包括货运起点、货运终点以及M个提货点中N个提货点,该N为正整数,且N≤M,为完成分布在该M个提货点的待运输货物,该至少一个路径方案中的每个路径方案均覆盖该M个提货点;为该至少一个路径方案中的每个路径方案中的每条运输路径进行该待运输货物的分配,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
在本申请实施方式中,在获取目标货运单后,根据该目标货运单中提供的信息进行路径规划以及货物分配,在确定路径方案时,可直接根据运输节点进行路径规划,可以减少进行路径搜索的时长,提高路径规划的效率,在路径规划完成之后再根据规划得到的路径方案进行货物分配,以得到每个路径方案的货物分配集合,后续再对每个路径方案以及每个路径方案对应的货物分配方案集合中的每个货物分配方案进行整合评价,以得到目标运输方案,可以提高得到目标运输方案的整体效率。
结合本申请第一方面的第一种实施方式,在本申请第一方面的第二种实施方式中,该根据该运输节点信息确定该至少一个路径方案,可以包括:
若历史路径数据的数量大于第一阈值,则基于该历史路径数据对该M个提货点的转移超参数进行初始化,以得到超参数矩阵;通过该超参数矩阵确定该M个提货点的转移概率分布,该转移概率分布包括运输路径中的集装箱在该货运起点与该M个提货点之间、该货运终点与该M个提货点之间或该M个提货点之间的转移概率;根据该转移概率分布确定该至少一个路径方案中的每个路径方案中的每条运输路径,以得到该至少一个路径方案。
在本申请实施方式中,可以通过历史路径数据进行路径规划,具体包括利用历史路径数据对M个提货点的转移超参数进行初始化,然后根据转移超参数确定提货点转移概率分布,该概率分布为路径方案中的每条运输路径的集装箱在提货点以及港口之间的转移概率。应理解,历史路径数据中一个跳转产生的次数越多,那么该跳转对应的概率越高,可以根据得到的提货点转移概率分布确定至少一个路径方案中每个路径方案的每条运输路径,可以进一步提高获取该至少一个路径方案的效率,且通过历史路径数据计算提货点转移超参数,可以使得到的路径方案更准确。
结合本申请第一方面的第二种实施方式,在本申请第一方面的第三种实施方式中,该 方法还可以包括:
若该历史路径数据的数量不大于该第一阈值,则基于启发式算法对该M个提货点的转移超参数进行初始化,以得到该超参数矩阵。
当历史路径数据数量不足时,此时存在无法通过历史路径数据对提货点的转移超参数进行初始化的情况,可以选择启发式算法对提货点的转移超参数进行初始化,增加了一种确定提货点超参数的方式。
结合本申请第一方面的第一种实施方式至本申请第一方面的第三种实施方式中的任一实施方式,在本申请第一方面的第四种实施方式中,该为该至少一个路径方案中的每个路径方案中的每条运输路径进行该待运输货物的分配,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案,可以包括:
对从目标货运单中获取到的M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,该聚类条件可以包括货物的长度、宽度、高度以及重量,此外,该聚类条件还可以包括材质、承压系数或最小面积等,通过该M个提货点中每个提货点的第一货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第一货物分配方式集合,该M个提货点中每个提货点的第一货物分配超参数为对该M个提货点中每个提货点进行货物分配的超参数,该M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该第一货物分配超参数可以是均匀分布的超参数,也可以是重复进行货物分配时通过上一次货物分配方案更新得到;从该M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
在本申请实施方式中,在对提货点的货物进行分配时,可以参考提货点所分布的货物的特征进行聚类,包括长度、宽度、高度或重量等特征,可以使用精确聚类,也可以使用模糊聚类,具体可根据实际需求调整,可以快速地将提货点的货物进行分类,从而快速地进行货物分配得到该至少一个路径方案中的每个路径方案对应的货物分配方案集合中的每个货物分配方案。
结合本申请第一方面的第一种实施方式至本申请第一方面的第四种实施方式中的任一实施方式,在本申请第一方面的第五种实施方式中,该根据快速装载模型确定该至少一个路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,可以包括:
获取该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,该第一特征向量用于指示某一货物分配方案中待运输货物的特征值,例如,每个货物分配方案中货物的长度、宽度、高度或重量等组成的向量;将该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第一特征向量输入该快速装载模型中,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中每个货物分配方案的实装率,该实装率包括体积实装率与载重实装率,该体积实装率包括该至少一个路径方案中的每个路径方案中的每条运输路径所分配 的货物的体积占用集装箱荷载体积的比例,该载重实装率该至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的重量占用集装箱的荷重的比例。
在本申请实施方式中,获取所得到的至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,该第一特征向量为指示至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中其中一个货物分配方案的特征值,将该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量输入该快速装载模型,可以得到至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的实装率,该实装率可以包括体积实装率与载重实装率,因此,可以将每个货物分配方案的第一特征向量输入快速装载模型中,可以快速得到至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的实装率,提高得到每个货物分配方案的实装率的效率。
结合本申请第一方面的第五种实施方式,在本申请第一方面的第六种实施方式中,该获取该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,可以包括:
获取该待运输货物中的每个货物的第二特征向量,该待运输货物中的每个货物的第二特征向量包括对应货物的长度、宽度、高度以及重量;根据该待运输货物中每个货物的第二特征向量计算出该M个提货点中的每个提货点分布的货物针对该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量,该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量包括该待运输货物中的每个货物的第二特征向量的均值与协方差;对该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量进行加权组合得到对应的该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量。
在本申请实施方式中,获取该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量的具体步骤可以是,首先获取待运输货物中每个货物的第二特征向量,并根据每个货物的第二特征向量计算出该M个提货点中针对至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量,并对该第三特征向量进行加权计算,最后得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量。
结合本申请第一方面、本申请第一方面的第一种实施方式至本申请第一方面的第六种实施方式中的任一实施方式,在本申请第一方面的第七种实施方式中,该通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,可以包括:
通过预置的评价函数以及该实装率,对所获取到的所有货物分配方案进行得分计算;若该所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从该得分高于该第二阈值的货物分配方案中确定该目标货物分配方案,以及将该目标货物分配方案对应的路径 方案作为该目标路径方案;通过该目标货物分配方案与该目标路径方案确定该目标运输方案。
在本申请实施方式中,可以通过预置的评价函数以及实装率对所获取到的所有货物分配方案进行得分计算得到每个货物分配方案的得分,若该所有货物分配方案中不存在得分高于第二阈值的货物分配方案,则从该得分高于第二阈值的货物分配中确定目标货物分配方案,若高于第二阈值的货物分配方案的数量为一个,则确定该一个货物分配方案为目标货物分配方案,若高于第二阈值的货物分配方案的数量为至少两个,则可以从该至少两个高于第二阈值的货物分配方案中随机确定一个或确定得分最高的货物分配方案为目标货物分配方案,以及确定该目标货物分配方案对应的路径方案为目标路径方案,以得到目标运输方案,本申请实施方式中通过对每个货物分配方案进行打分以确定目标货物分配方案,可以得到最优的目标运输方案。
结合本申请第一方面的第七种实施方式,在本申请第一方面的第八种实施方式中,该评价函数包括:
Figure PCTCN2018108534-appb-000001
其中,该
Figure PCTCN2018108534-appb-000002
为路径方案向量,m为集装箱的数量,
Figure PCTCN2018108534-appb-000003
为m个集装箱的体积实装率向量,
Figure PCTCN2018108534-appb-000004
为该m个集装箱的载重实装率向量;该α、该β与该γ为权重参数,该r Vi为第i个集装箱的体积实装率,该r Wi为第i个集装箱的载重实装率,该
Figure PCTCN2018108534-appb-000005
为该m个集装箱的平均体积实装率,该
Figure PCTCN2018108534-appb-000006
为该m个集装箱的平均载重实装率。
本申请实施方式增加了一种对货物分配方案以及路径方案进行评价的评价函数,可以根据该评价函数得到最优的目标运输方案。
结合本申请第一方面的第七种实施方式或本申请第一方面的第八种实施方式,在本申请第一方面的第九种实施方式中,该方法还包括:
若该所有货物分配方案中不包括得分高于该第二阈值的货物分配方案,则通过该M个提货点中每个提货点的第二货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第二货物分配方式集合,该M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该M个提货点中每个提货点的第二货物分配超参数为通过该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案对该M个提货点中每个提货点的第一货物分配超参数进行更新得到;从该M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案,该至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该待运输货物进行分配的方案;通过该评价函数以及该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率,对该至少一个路径方案中的每 个路径方案的第二货物分配方案集合中的每个货物分配方案进行得分计算,该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率由该快速装载模型得到。
本申请实施方式中,若该所有分配方案中不包括得分高于该第二阈值的货物分配方案,则可以通过第一货物分配方案集合中的每个货物分配方案对每个提货点的第一货物分配超参数进行更新,得到每个提货点的第二货物分配超参数,然后根据该第二货物分配超参数对每个提货点的货物重新进行货物分配,以得到每个路径方案的第二货物分配方案集合中的每个货物分配方案,之后继续对该第二货物分配方案集合中的每个货物分配方案进行进一步地整合评价,直到达到停止条件为止,例如,得到得分大于该第二阈值的货物分配方案,或迭代的次数达到预置的次数。因此,本申请实施方式中通过对货物分配方案进行重复分配以及整合评价,可以得到更优的目标货物分配方案以及目标路径方案。
应理解,对货物分配方案进行重复分配时,还可以重新对路径方案进行规划,也可以直接通过该至少一个路径方案重新进行货物分配。
结合本申请第一方面、本申请第一方面的第一种实施方式至本申请第一方面的第九种实施方式中的任一实施方式,在本申请第一方面的第十种实施方式中,该通过该实装率对该至少一个路径方案中的每个路径方案与该至少一个路径方案中的每个路径方案的货物分配方案进行整合评价,以确定目标运输方案之后,该方法还包括:
根据该目标货物分配方案和该目标路径方案,确定该目标路径方案中每条运输路径的集装箱的型号;根据该目标路径方案中每条运输路径的集装箱的型号以及三维装载算法生成装载方案,该装载方案为该待运输货物在该目标路径方案中每条运输路径中的集装箱内的装载方式。
在本申请实施方式中,确定目标运输方案后,还可以进一步确定集装箱的型号,可以根据实装率进行调整,确定与实装率更匹配的集装箱型号,以节约运输成本。且在确定集装箱型号后,可以进一步通过三维装载算法生成装载方案,确定货物在集装箱内的装载方式,可以提高货物装载的效率。
结合本申请第一方面、本申请第一方面的第一种实施方式至本申请第一方面的第十种实施方式中的任一实施方式,在本申请第一方面的第十一种实施方式中,在该通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案之前,该方法还包括:
若该实装率确定该M个提货点中的L个提货点还包括未分配到该集装箱的剩余货物,则为该剩余货物确定剩余货物路径方案和剩余货物分配方案,该L≤该M,该L为正整数;
该通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,可以包括:
通过该实装率对该至少一个路径方案中的每个路径方案与对应的第一货物分配方案集合中的每个货物分配方案,以及该剩余货物路径方案和该剩余货物分配方案进行整合评价, 以确定目标运输方案。
在本申请实施方式中,若存在不能装入集装箱的货物,可以根据实装率计算待运输货物中是否还包括未分配到集装箱的剩余货物,则可以根据对剩余货物进行路径规划以及货物分配,以得到剩余货物的路径方案以及货物分配方案,并将该剩余货物的路径方案以及货物分配方案、目标路径方案以及目标货物分配方案作为目标运输方案,以得到对待运输货物的完整运输方案。
本申请第二方面提供一种训练快速装载模型的方法,可以包括:
首先获取离线仿真数据,该离线仿真数据包括在离线仿真时通过三维装载计算得到的历史装载方案与历史实装率;然后从从该离线仿真数据中获取特征向量,该特征向量包括该历史装载方案对应的历史运输货物的特征值;将该特征向量装换为预置格式的训练数据;通过该训练数据训练预测模型,以得到快速装载模型,该快速装载模型用于输出每个运输路径的货物分配方案集合中每个货物分配方案的实装率,该实装率为该每个货物分配方案中装入集装箱的货物占用该集装箱的比例。
在本申请实施方式中,可以通过离线仿真数据训练快速装载模型,该快速装载模型用于快速得到货物分配方案的实装率,可以提高确定目标运输方案的效率。
结合本申请第二方面,在本申请第二方面的第一种实施方式中,该预置格式为:(特征向量,历史实装率)。
结合本申请第二方或本申请第二方面的第一种实施方式,在本申请第二方面的第二种实施方式中,该预测模型可以包括但不限于:线性回归模型、岭回归模型、LASSO模型、支持向量机模型、随机森林模型、XgBoost模型或人工神经网络模型等。
结合本申请第二方面,本申请第二方面的第一种实施方式或本申请第二方面的第二种实施方式,在本申请第二方面的第三种实施方式中,该获取模块,可以包括:
首先获取至少一个历史路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合,该至少一个历史路径方案中的每个路径方案为针对历史运输货物进行运输而规划的运输路径,其中,一个路径方案可以包括至少一条运输路径,该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合包括至少一个货物分配方案,该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该历史运输货物进行分配的方案;根据三维装载算法确定该至少一个历史路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率以及装载方案,该实装率为某一货物分配方案中装入集装箱的货物占用该集装箱的比例;通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,该目标运输方案包括目标路径方案与该目标路径方案对应的目标货物分配方案。
在本申请实施例中,根据通过离线仿真数据进行离线训练的三维装载算法模型确定每个路径方案对应的货物分配方案的实装率,可以准确地获取到每个路径方案对应的货物分配方案的实装率,提高确定离线仿真数据的准确性。
结合本申请第二方面,在本申请第二方面的第四种实施方式中,获取至少一个历史路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合,可以包括:
首先获取历史货运单,该历史货运单包括运输节点信息以及历史运输货物信息,该运输节点信息包括货运起点、货运终点以及M个提货点,该历史运输货物信息包括分布在该M个提货点的该历史运输货物的信息,该M为正整数;然后根据该运输节点信息中的运输节点确定该至少一个历史路径方案,其中,一个路径方案可以包括至少一条运输路径,该至少一条运输路径中的每条运输路径包括货运起点、货运终点以及M个提货点中N个提货点,该N为正整数,且N≤M,为完成分布在该M个提货点的历史运输货物,该至少一个历史路径方案中的每个路径方案均覆盖该M个提货点;为该至少一个历史路径方案中的每个路径方案中的每条运输路径进行该历史运输货物的分配,以得到该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案。
在本申请实施方式中,在获取历史货运单后,根据该历史货运单中提供的信息进行路径规划以及货物分配,在确定路径方案时,可直接根据运输节点进行路径规划,可以减少进行路径搜索的时长,提高路径规划的效率,在路径规划完成之后再根据规划得到的路径方案进行货物分配,以得到每个路径方案的历史货物分配方案集合,后续再对每个路径方案以及每个路径方案对应的历史货物分配方案集合中的每个货物分配方案进行整合评价,以得到目标运输方案,可以提高得到目标运输方案的整体效率。
结合本申请第二方面的第四种实施方式,在本申请第二方面的第五种实施方式中,该根据该运输节点信息确定该至少一个历史路径方案,可以包括:
若历史路径数据的数量大于第一阈值,则基于该历史路径数据对该M个提货点的转移超参数进行初始化,以得到超参数矩阵,该历史路径数据包括针对历史待运输货物进行运输时的历史路径方案;通过该超参数矩阵确定该M个提货点的转移概率分布,该转移概率分布包括运输路径中的集装箱在该货运起点与该M个提货点之间、该货运终点与该M个提货点之间或该M个提货点之间的转移概率;根据该转移概率分布确定该至少一个历史路径方案中的每个路径方案中的每条运输路径,以得到该至少一个历史路径方案。
在本申请实施方式中,可以通过历史路径数据进行路径规划,具体包括利用历史路径数据对M个提货点的转移超参数进行初始化,然后根据转移超参数确定提货点转移概率分布,该概率分布为路径方案中的每条运输路径的集装箱在提货点以及港口之间的转移概率。应理解,历史路径数据中一个跳转产生的次数越多,那么该跳转对应的概率越高,可以根据得到的提货点转移概率分布确定至少一个历史路径方案中每个路径方案的每条运输路径,可以进一步提高获取该至少一个历史路径方案的效率,且通过历史路径数据计算提货点转移超参数,可以使得到的路径方案更准确。
结合本申请第二方面的第五种实施方式,在本申请第二方面的第六种实施方式中,该方法还可以包括:
若该历史路径数据的数量不大于该第一阈值,则基于启发式算法对该M个提货点的转移超参数进行初始化,以得到该超参数矩阵。
当历史路径数据数量不足时,此时无法通过历史路径数据对提货点的转移超参数进行初始化,可以选择启发式算法对提货点的转移超参数进行初始化,增加了一种确定提货点超参数的方式。
结合本申请第二方面的第三种实施方式至本申请第二方面的第六种实施方式中的任一实施方式,在本申请第二方面的第七种实施方式中,该为该至少一个历史路径方案中的每个路径方案中的每条运输路径进行该历史运输货物的分配,以得到该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案,可以包括:
对从历史货运单中获取到的M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,该聚类条件可以包括货物的长度、宽度、高度以及重量,此外,该聚类条件还可以包括材质、承压系数或最小面积等,通过该M个提货点中每个提货点的第一货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第一货物分配方式集合,该M个提货点中每个提货点的第一货物分配超参数为对该M个提货点中每个提货点进行货物分配的超参数,该M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该第一货物分配超参数可以是均匀分布的超参数,也可以是重复进行货物分配时通过上一次货物分配方案更新得到;从该M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案。
在本申请实施方式中,在对提货点的货物进行分配时,可以参考提货点所分布的货物的特征进行聚类,包括长度、宽度、高度或重量等特征,可以使用精确聚类,也可以使用模糊聚类,具体可根据实际需求调整,可以快速地将提货点的货物进行分类,从而快速地进行货物分配得到该至少一个历史路径方案中的每个路径方案对应的历史货物分配方案集合中的每个货物分配方案。
结合本申请第二方面、本申请第二方面的第三种实施方式至本申请第二方面的第六种实施方式中的任一实施方式,在本申请第二方面的第七种实施方式中,该通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,可以包括:
通过预置的评价函数以及该实装率,对所获取到的所有货物分配方案进行得分计算;若该所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从该得分高于该第二阈值的货物分配方案中确定该目标货物分配方案,以及将该目标货物分配方案对应的路径方案作为该目标路径方案;通过该目标货物分配方案与该目标路径方案确定该目标运输方案。
在本申请实施方式中,可以通过预置的评价函数以及实装率对所获取到的所有货物分配方案进行得分计算得到每个货物分配方案的得分,若该所有货物分配方案中不存在得分高于第二阈值的货物分配方案,则从该得分高于第二阈值的货物分配中确定目标货物分配方案,若高于第二阈值的货物分配方案的数量为一个,则确定该一个货物分配方案为目标 货物分配方案,若高于第二阈值的货物分配方案的数量为至少两个,则可以从该至少两个高于第二阈值的货物分配方案中随机确定一个或确定得分最高的货物分配方案为目标货物分配方案,以及确定该目标货物分配方案对应的路径方案为目标路径方案,以得到目标运输方案,本申请实施方式中通过对每个货物分配方案进行打分以确定目标货物分配方案,可以得到最优的目标运输方案。
结合本申请第二方面的第七种实施方式,在本申请第二方面的第八种实施方式中,该评价函数包括:
Figure PCTCN2018108534-appb-000007
其中,该
Figure PCTCN2018108534-appb-000008
为路径方案向量,m为集装箱的数量,
Figure PCTCN2018108534-appb-000009
为m个集装箱的体积实装率向量,
Figure PCTCN2018108534-appb-000010
为该m个集装箱的载重实装率向量;该α、该β与该γ为权重参数,该r Vi为第i个集装箱的体积实装率,该r Wi为第i个集装箱的载重实装率,该
Figure PCTCN2018108534-appb-000011
为该m个集装箱的平均体积实装率,该
Figure PCTCN2018108534-appb-000012
为该m个集装箱的平均载重实装率。
本申请实施方式增加了一种对货物分配方案以及路径方案进行评价的评价函数,可以根据该评价函数得到最优的目标运输方案。
结合本申请第二方面的第七种实施方式或本申请第二方面的第八种实施方式,在本申请第二方面的第九种实施方式中,该方法还包括:
若该所有货物分配方案中不包括得分高于该第二阈值的货物分配方案,则通过该M个提货点中每个提货点的第二货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第二货物分配方式集合,该M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该M个提货点中每个提货点的第二货物分配超参数为通过该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案对该M个提货点中每个提货点的第一货物分配超参数进行更新得到;从该M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个历史路径方案中的每个路径方案对应的第二历史货物分配方案集合中的每个货物分配方案,该至少一个历史路径方案中的每个路径方案对应的第二历史货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该历史运输货物进行分配的方案;通过该评价函数以及该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案的实装率,对该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案进行得分计算,该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案的实装率由该三维装载算法得到。
本申请实施方式中,若该所有分配方案中不包括得分高于该第二阈值的货物分配方案,则可以通过第一历史货物分配方案集合中的每个货物分配方案对每个提货点的第一货物分配超参数进行更新,得到每个提货点的第二货物分配超参数,然后根据该第二货物分配超 参数对每个提货点的货物重新进行货物分配,以得到每个路径方案的第二历史货物分配方案集合中的每个货物分配方案,之后继续对该第二历史货物分配方案集合中的每个货物分配方案进行进一步地整合评价,直到达到停止条件为止,例如,得到得分大于该第二阈值的货物分配方案,或迭代的次数达到预置的次数。因此,本申请实施方式中通过对货物分配方案进行重复分配以及整合评价,可以得到更优的目标货物分配方案以及目标路径方案。
应理解,对货物分配方案进行重复分配时,还可以重新对路径方案进行规划,也可以直接通过该至少一个历史路径方案重新进行货物分配。
结合本申请第二方面、本申请第二方面的第三种实施方式至本申请第二方面的第九种实施方式中的任一实施方式,在本申请第二方面的第十种实施方式中,在该通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案之前,该方法还包括:
若该实装率确定该M个提货点中的L个提货点还包括未分配到该集装箱的剩余货物,则为该剩余货物确定剩余货物路径方案和剩余货物分配方案,该L≤该M,该L为正整数;
该通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,可以包括:
通过该实装率对该至少一个历史路径方案中的每个路径方案与对应的第一历史货物分配方案集合中的每个货物分配方案,以及该剩余货物路径方案和该剩余货物分配方案进行整合评价,以确定目标运输方案。
在本申请实施方式中,若存在不能装入集装箱的货物,可以根据实装率计算历史运输货物中是否还包括未分配到集装箱的剩余货物,则可以根据对剩余货物进行路径规划以及货物分配,以得到剩余货物的路径方案以及货物分配方案,并将该剩余货物的路径方案以及货物分配方案、目标路径方案以及目标货物分配方案作为目标运输方案,以得到对历史运输货物的完整运输方案。
本申请第三方面提供一种确定装置,其特征在于,包括:
获取模块,用于获取至少一个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合,该至少一个路径方案中的每个路径方案为针对待运输货物进行运输而规划的运输路径,该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合包括至少一个货物分配方案,该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该待运输货物进行分配的方案;
快速装载模块,用于根据快速装载模型确定该至少一个路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,该快速装载模型为通过离线仿真数据进行离线训练得到,该离线仿真数据包括通过三维装载算法计算得到的历史装载方案,该实装率为某一货物分配方案中装入集装箱的货物占用该集装箱的比例;
评价模块,通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路 径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,该目标运输方案包括目标路径方案与该目标路径方案对应的目标货物分配方案。
结合本申请第三方面,在本申请第三方面的第一种实施方式中,该获取模块,包括:
获取子模块,用于获取目标货运单,该目标货运单包括运输节点信息以及待运输货物信息,该运输节点信息包括货运起点、货运终点以及M个提货点,该待运输货物信息包括分布在该M个提货点的该待运输货物的信息,该M为正整数;
路径规划子模块,用于根据该运输节点信息确定该至少一个路径方案,该至少一个路径方案中的每个路径方案包括至少一条运输路径,该至少一条运输路径中的每条运输路径包括货运起点、货运终点以及该M个提货点中N个提货点,该至少一个路径方案中的每个路径方案均覆盖该M个提货点,该N为正整数,且N≤M;
货物分配子模块,用于为该至少一个路径方案中的每个路径方案中的每条运输路径进行该待运输货物的分配,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
结合本申请第三方面的第一种实施方式,在本申请第三方面的第二种实施方式中,该路径规划子模块,具体用于:
若历史路径数据的数量大于第一阈值,则基于该历史路径数据对该M个提货点的转移超参数进行初始化,以得到超参数矩阵;
通过该超参数矩阵确定该M个提货点的转移概率分布,该转移概率分布包括运输路径中的集装箱在该货运起点与该M个提货点之间、该货运终点与该M个提货点之间或该M个提货点之间的转移概率;
根据该转移概率分布确定该至少一个路径方案中的每个路径方案中的每条运输路径,以得到该至少一个路径方案。
结合本申请第三方面的第一种实施方式或本申请第三方面的第二种实施方式,在本申请第三方面的第三种实施方式中,该确定装置还包括:
初始化模块,用于若该历史路径数据的数量不大于该第一阈值,则基于启发式算法对该M个提货点的转移超参数进行初始化,以得到该超参数矩阵。
结合本申请第三方面的第一种实施方式至本申请第三方面的第三种实施方式中任一实施方式,在本申请第三方面的第四种实施方式中,该货物分配子模块,具体用于:
对该M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,该聚类条件包括货物的长度、宽度、高度以及重量;
通过该M个提货点中每个提货点的第一货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第一货物分配方式集合,该M个提货点中每个提货点的第一货物分配超参数为对该M个提货点中每个提货点进行货物分配的超参数,该M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式;
从该M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行 结合,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
结合本申请第三方面的第一种实施方式至本申请第三方面的第四种实施方式中任一实施方式,在本申请第三方面的第五种实施方式中,该快速装载模块,具体用于:
获取该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,该第一特征向量用于指示某一货物分配方案中待运输货物的特征值;
将该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第一特征向量输入该快速装载模型中,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中每个货物分配方案的实装率,该实装率包括体积实装率与载重实装率,该体积实装率包括该至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的体积占用集装箱荷载体积的比例,该载重实装率该至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的重量占用集装箱的荷重的比例。
结合本申请第三方面的第五种实施方式,在本申请第三方面的第六种实施方式中,该快速装载模块,具体用于:
获取该待运输货物中的每个货物的第二特征向量,该待运输货物中的每个货物的第二特征向量包括对应货物的长度、宽度、高度以及重量;
根据该待运输货物中每个货物的第二特征向量计算出该M个提货点中的每个提货点分布的货物针对该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量,该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量包括该待运输货物中的每个货物的第二特征向量的均值与协方差;
对该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量进行加权组合得到对应的该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量。
结合本申请第三方面、本申请第三方面的第一种实施方式至本申请第三方面的第六种实施方式中任一实施方式,在本申请第三方面的第七种实施方式中,该评价模块,具体用于:
通过预置的评价函数以及该实装率,对所获取到的所有货物分配方案进行得分计算;
若该所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从该得分高于该第二阈值的货物分配方案中确定该目标货物分配方案,以及将该目标货物分配方案对应的路径方案作为该目标路径方案;
通过该目标货物分配方案与该目标路径方案确定该目标运输方案。
结合本申请第三方面的第七种实施方式,在本申请第三方面的第八种实施方式中,该评价函数包括:
Figure PCTCN2018108534-appb-000013
其中,该
Figure PCTCN2018108534-appb-000014
为路径方案向量, m为集装箱的数量,
Figure PCTCN2018108534-appb-000015
为m个集装箱的体积实装率向量,
Figure PCTCN2018108534-appb-000016
为该m个集装箱的载重实装率向量;该α、该β与该γ为权重参数,该r Vi为第i个集装箱的体积实装率,该r Wi为第i个集装箱的载重实装率,该
Figure PCTCN2018108534-appb-000017
为该m个集装箱的平均体积实装率,该
Figure PCTCN2018108534-appb-000018
为该m个集装箱的平均载重实装率。
结合本申请第三方面的第七种实施方式或本申请第三方面的第八种实施方式,在本申请第三方面的第九种实施方式中该评价模块,还用于:
若该所有货物分配方案中不包括得分高于该第二阈值的货物分配方案,则通过该M个提货点中每个提货点的第二货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第二货物分配方式集合,该M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该M个提货点中每个提货点的第二货物分配超参数为通过该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案对该M个提货点中每个提货点的第一货物分配超参数进行更新得到;
从该M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案,该至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该待运输货物进行分配的方案;
通过该评价函数以及该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率,对该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案进行得分计算,该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率由该快速装载模型得到。
结合本申请第三方面、本申请第三方面的第一种实施方式至本申请第三方面的第九种实施方式中任一实施方式,在本申请第三方面的第十种实施方式中,该确定装置还包括:
后期处理模块,用于通过该实装率对该至少一个路径方案中的每个路径方案与该至少一个路径方案中的每个路径方案的货物分配方案进行整合评价,以确定目标运输方案之后,根据该目标货物分配方案和该目标路径方案,确定该目标路径方案中每条运输路径的集装箱的型号;
三维装载模块,用于根据该后期处理模块确定的该目标路径方案中每条运输路径的集装箱的型号以及三维装载算法生成装载方案,该装载方案为该待运输货物在该目标路径方案中每条运输路径中的集装箱内的装载方式。
结合本申请第三方面、本申请第三方面的第一种实施方式至本申请第三方面的第十种实施方式中任一实施方式,在本申请第三方面的第十一种实施方式中,该确定装置还可以包括:
确定模块,在该通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整 合评价,以确定目标运输方案之前,用于若该实装率确定该M个提货点中的L个提货点还包括未分配到该集装箱的剩余货物,则为该剩余货物确定剩余货物路径方案和剩余货物分配方案,该L≤该M,该L为正整数;
该评价模块,还用于通过该实装率对该至少一个路径方案中的每个路径方案与对应的第一货物分配方案集合中的每个货物分配方案,以及该剩余货物路径方案和该剩余货物分配方案进行整合评价,以确定目标运输方案。
本申请第四方面提供一种训练装置,其特征在于,包括:
获取模块,用于获取离线仿真数据,该离线仿真数据包括通过三维装载计算得到的历史装载方案与历史实装率;
该获取模块,还用于从该离线仿真数据中获取特征向量,该特征向量包括该历史装载方案对应的历史运输货物的特征值;
转换模块,用于将该特征向量转换为预置格式的训练数据;
训练模块,用于通过该训练数据训练预测模型,以得到快速装载模型,该快速装载模型用于输出每个运输路径的货物分配方案集合中每个货物分配方案的实装率,该实装率为该每个货物分配方案中装入集装箱的货物占用该集装箱的比例。
结合本申请第四方面,在本申请第四方面的第一种实施方式中,该预置格式为:(特征向量,历史实装率)。
结合本申请第四方面或本申请第四方面的第一种实施方式,在本申请第四方面的第二种实施方式中,该预测模型包括:线性回归模型、岭回归模型、LASSO模型、支持向量机模型、随机森林模型、XgBoost模型或人工神经网络模型。
结合本申请第四方面,本申请第四方面的第一种实施方式或本申请第四方面的第二种实施方式,在本申请第四方面的第三种实施方式中,该获取模块,可以包括:
获取子模块,用于获取至少一个历史路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合,该至少一个历史路径方案中的每个路径方案为针对历史运输货物进行运输而规划的运输路径,该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合包括至少一个货物分配方案,该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该历史运输货物进行分配的方案;
三维装载子模块,用于根据三维装载算法确定该至少一个历史路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,该实装率为某一货物分配方案中装入集装箱的货物占用该集装箱的比例;
评价子模块,通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,该目标运输方案包括目标路径方案与该目标路径方案对应的目标货物分配方案。
在本申请实施例中,可以在训练快速装载模型时使用三维装载算法进行计算,得到历史路径方案对应历史装载方案,可以准确输出历史路径方案对应的货物分配方案的实装率。
结合本申请第四方面的第三种实施方式,在本申请第四方面的第四种实施方式中,该获取子模块,包括:
获取单元,用于获取历史货运单,该历史货运单包括运输节点信息以及历史运输货物信息,该运输节点信息包括货运起点、货运终点以及M个提货点,该历史运输货物信息包括分布在该M个提货点的该历史运输货物的信息,该M为正整数;
路径规划单元,用于根据该运输节点信息确定该至少一个历史路径方案,该至少一个历史路径方案中的每个路径方案包括至少一条运输路径,该至少一条运输路径中的每条运输路径包括货运起点、货运终点以及该M个提货点中N个提货点,该至少一个历史路径方案中的每个路径方案均覆盖该M个提货点,该N为正整数,且N≤M;
货物分配单元,用于为该至少一个历史路径方案中的每个路径方案中的每条运输路径进行该历史运输货物的分配,以得到该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案。
结合本申请第四方面的第三种实施方式,在本申请第四方面的第五种实施方式中,该路径规划单元,具体用于:
若历史路径数据的数量大于第一阈值,则基于该历史路径数据对该M个提货点的转移超参数进行初始化,以得到超参数矩阵;
通过该超参数矩阵确定该M个提货点的转移概率分布,该转移概率分布包括运输路径中的集装箱在该货运起点与该M个提货点之间、该货运终点与该M个提货点之间或该M个提货点之间的转移概率;
根据该转移概率分布确定该至少一个历史路径方案中的每个路径方案中的每条运输路径,以得到该至少一个历史路径方案。
结合本申请第四方面的第四种实施方式,在本申请第四方面的第六种实施方式中,该训练装置还包括:
初始化模块,用于若该历史路径数据的数量不大于该第一阈值,则基于启发式算法对该M个提货点的转移超参数进行初始化,以得到该超参数矩阵。
结合本申请第四方面的第三种实施方式至本申请第四方面的第五种实施方式中的任一实施方式,在本申请第四方面的第六种实施方式中,该货物分配单元,具体用于:
对该M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,该聚类条件包括货物的长度、宽度、高度以及重量;
通过该M个提货点中每个提货点的第一货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第一货物分配方式集合,该M个提货点中每个提货点的第一货物分配超参数为对该M个提货点中每个提货点进行货物分配的超参数,该M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式;
从该M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案。
结合本申请第四方面的第三种实施方式至本申请第四方面的第六种实施方式中的任一实施方式,在本申请第四方面的第七种实施方式中,,该评价子模块,具体用于:
通过预置的评价函数以及该实装率,对所获取到的所有货物分配方案进行得分计算;
若该所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从该得分高于该第二阈值的货物分配方案中确定该目标货物分配方案,以及将该目标货物分配方案对应的路径方案作为该目标路径方案;
通过该目标货物分配方案与该目标路径方案确定该目标运输方案。
结合本申请第四方面的第七种实施方式,在本申请第四方面的第八种实施方式中,该评价函数包括:
Figure PCTCN2018108534-appb-000019
其中,该
Figure PCTCN2018108534-appb-000020
为路径方案向量,m为集装箱的数量,
Figure PCTCN2018108534-appb-000021
为m个集装箱的体积实装率向量,
Figure PCTCN2018108534-appb-000022
为该m个集装箱的载重实装率向量;该α、该β与该γ为权重参数,该r Vi为第i个集装箱的体积实装率,该r Wi为第i个集装箱的载重实装率,该
Figure PCTCN2018108534-appb-000023
为该m个集装箱的平均体积实装率,该
Figure PCTCN2018108534-appb-000024
为该m个集装箱的平均载重实装率。
结合本申请第四方面的第六种实施方式或本申请第四方面的第七种实施方式,在本申请第四方面的第十一种实施方式中,该评价子模块,还用于:
若该所有货物分配方案中不包括得分高于该第二阈值的货物分配方案,则通过该M个提货点中每个提货点的第二货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第二货物分配方式集合,该M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该M个提货点中每个提货点的第二货物分配超参数为通过该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案对该M个提货点中每个提货点的第一货物分配超参数进行更新得到;
从该M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个历史路径方案中的每个路径方案对应的第二历史货物分配方案集合中的每个货物分配方案,该至少一个历史路径方案中的每个路径方案对应的第二历史货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该历史运输货物进行分配的方案;
通过该评价函数以及该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案的实装率,对该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案进行得分计算,该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案的实装率由该三维装载子模型得到。
应理解,对货物分配方案进行重复分配时,还可以重新对路径方案进行规划,也可以 直接通过该至少一个历史路径方案重新进行货物分配。
结合本申请第四方面、本申请第四方面的第三种实施方式至本申请第四方面的第十一种实施方式中的任一实施方式,在本申请第四方面的第十二种实施方式中,该训练装置还包括:
确定模块,在该通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案之前,用于若该实装率确定该M个提货点中的L个提货点还包括未分配到该集装箱的剩余货物,则为该剩余货物确定剩余货物路径方案和剩余货物分配方案,该L≤该M,该L为正整数;
该评价子模块,还用于通过该实装率对该至少一个历史路径方案中的每个路径方案与对应的第一历史货物分配方案集合中的每个货物分配方案,以及该剩余货物路径方案和该剩余货物分配方案进行整合评价,以确定目标运输方案。
本申请第五方面提供一种确定装置,可以包括:
处理器、存储器、总线以及输入输出接口,该处理器、该存储器与该输入输出接口通过该总线连接;
该存储器,用于存储程序代码;
该处理器调用该存储器中的程序代码时执行本申请第一方面提供的方法的步骤。
本申请第六方面提供一种训练装置,可以包括:
处理器、存储器、总线以及输入输出接口,该处理器、该存储器与该输入输出接口通过该总线连接;
该存储器,用于存储程序代码;
该处理器调用该存储器中的程序代码时执行本申请第二方面提供的方法的步骤。
本申请实施例第七方面提供一种存储介质,其上存储有可编程指令,当所述可编程指令在计算机上运行时,使得计算机执行上述第一方面或第一方面任一实现方式中描述的方法。
该存储介质包括:U盘、移动硬盘、只读存储器(英文缩写ROM,英文全称:Read-Only Memory)、随机存取存储器(英文缩写:RAM,英文全称:Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
本申请实施例第八方面提供一种存储介质,其上存储有可编程指令,当所述可编程指令在计算机上运行时,使得计算机执行上述第二方面或第二方面任一实现方式中描述的方法。该存储介质包括:U盘、移动硬盘、只读存储器(英文缩写ROM,英文全称:Read-Only Memory)、随机存取存储器(英文缩写:RAM,英文全称:Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
本申请实施例第九方面提供了一种计算机程序产品,该计算机程序产品包括计算机软件指令,该计算机软件指令可通过处理器进行加载来实现上述第一方面的确定运输方案的方法中的流程。
本申请实施例第十方面提供了一种计算机程序产品,该计算机程序产品包括计算机软 件指令,该计算机软件指令可通过处理器进行加载来实现上述第二方面的训练快速装载模型的方法中的流程。
本申请实施例第十一方面提供了一种仿真系统,包括确定装置与训练装置,该确定装置用于执行本申请第一方面至第一方面中任一实施方式中的步骤;该训练装置执行本申请第二方面至第二方面中任一实施方式中的步骤。
从以上技术方案可以看出,本申请实施例具有以下优点:
在确定目标运输方案时,可以使用快速装载模型得到每个路径方案对应的货物分配方案的实装率,然后根据实装率确定目标运输方案,其中,快速装载模型为通过对离线仿真数据进行离线训练得到,该离线仿真数据为通过三维计算得到的历史装载方案,通过该快速装载模型可以快速得到货物分配方案的实装率,无需进行三维运算得到装载方式,可直接对实装率进行计算,可以快速得到货物分配方案的实装率,可以降低得到实装率所需的时长,进而提高得到目标运输方案的效率。
附图说明
图1为本申请实施例中确定运输方案的方法的一种应用场景示意图;
图2为本申请实施例中确定运输方案的方法的一种实施例示意图;
图3为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图4为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图5为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图6为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图7为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图8为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图9为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图10为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图11为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图12为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图13为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图14为本申请实施例中训练预测模型的一种实施例示意图;
图15为本申请实施例中确定运输方案的方法的另一种实施例示意图;
图16为本申请实施例中三维装载仿真的一种实施例示意图
图17为本申请实施例中确定装置的一种实施例示意图;
图18为本申请实施例中训练装置的一种实施例示意图;
图19为本申请实施例中确定装置的另一种实施例示意图;
图20为本申请实施例中训练装置的另一种实施例示意图。
具体实施方式
本申请实施例提供一种确定运输方案的方法、训练快速装载模型的方法及设备,用于 进行货物运输,特别在运输量大、复杂的场景下,可以快速得到目标运输方案,减小运输成本,提高运输效率。
随着物流行业的发展,货物运输在工业以及生活中都广泛应用,集装箱装载仿真是物流领域的核心问题,因此对集装箱的装载仿真需要达到高效性与准确性。高效性是指可以快速响应,可以通过输入的数据在短时间内输出装载结果,因此可以优先抢占物流资源,缩短发货的时间,保证货物的及时运输与交付。准确性是指输出的装载结果的有效,能够提高集装箱的利用率,降低运输成本。
本申请实施例应用的场景可以如图1所示,本申请实施例仅以2个港口,4个提货点,以及2个集装箱为例,在实际应用中,港口、提货点以及集装箱的数量可根据实际需求调整,具体此处不作限定。首先确定目标货运单,目标货运单中包括待运输货物,提货点以及港口,该待运输货物分布在提货点D1、提货点D2、提货点D3以及提货点D4,需要通过集装箱将该待运输货物运输到港口2,集装箱位于港口1。然后确定集装箱的数量与每个集装箱的路径,以及每个集装箱在途经提货点的货物分配方案和装载方案。本申请实施例提供的确定运输方案的方法可以快速生成集装箱的最佳运输路径以及装载方案,可以提高集装箱的实装率,以及运输的效率。在实际场景中,集装箱可以由货运车运输,一个集装箱可以由一辆货运车运输。
例如,目标货运单中要求将D1D2、D3以及D4的货物运输到港口2,可以确定路径方案为:集装箱1从港口1出发,经过D1以及D3,运输D1以及D3的货物,然后到达港口2,经过计算,集装箱1的体积实装率为95%,载重实装率为96%;集装箱2从港口1出发,经过D2以及D4,运输D2以及D4的货物,然后到达港口2,经过计算,集装箱2的体积实装率为97%,载重实装率为98%。因此,可通过本申请实施例提供的确定运输方案的方法得到货物运输的最佳路径,且提高集装箱的体积实装率与载重实装率。
下面对本申请实施例提供的确定运输方案的方法的流程进行说明,请参阅图2,本申请实施例中确定运输方案的方法的一种实施例示意图,包括:
201、获取目标货运单;
该目标货运单为待运输的货运单,目标货运单中包括运输节点以及待运输货物的信息,该运输节点包括港口以及M个提货点,其中,港口可以包括针对该待运输货物进行运输的货运起点与货运终点,该货运起点与货运终点可以是同一个港口,也可以是不同的港口,该待运输货物分布在该M个提货点中的各个提货点,该港口的数量可以是一个,也可以是多个,该M个提货点的数量可以是一个,也可以是多个,具体此处不作限定。在实际应用中,待运输货物单可以由用户输入得到,也可以是由实际运输系统生成得到。
202、根据运输节点对运输路径进行规划,以得到路径方案集合;
在获取到目标货运单后,可以通过该目标货运单获取运输的货运起点、货运终点以及待运输货物所分布的提货点,可根据历史路径数据对对运输路径进行规划,可以得到至少一个路径方案,该至少一个路径方案中的每个路径方案可以包括至少一条运输路径,即由多条运输路径组成一个完整的路径方案,该至少一个路径方案组成路径方案集合,该路径方案集合中包括至少一个路径方案。其中,一条运输路径可以对应一个集装箱,若完成待 运输货物需要多个集装箱,则可以确定多个集装箱的运输路径,若一个提货点的货物无法通过一个集装箱完成运输,则可以使用多个集装箱进行运输,因此一个提货点可以由多个集装箱经过。例如,目标货运单中的需求是将提货点D1、D2、D3以及D4的货物运输到港口2,集装箱从港口1出发,可对提货点转移超参数进行初始化,然后通过提货点转移超参数得到提货点转移概率,确定提货点之间、或港口与提货点之间的转移概率,可以根据该提货点转移概率得到多条运输路径,包括从港口1至D1、D2、D3、D4、至港口2,或从港口1至D2、D1、D4、D3、至港口2等至少一条运输路径,该至少一条运输路径可以组成至少一个路径方案,该至少一个路径方案组成路径方案集合。
具体地,目标货运单中包括运输节点以及分布在该M个提货点的待运输货物的信息,该运输节点包括港口,M个提货点,其中M为正整数,该港口包括货运起点与货运终点,其中,货运起点与货运终点可以是同一个港口,也可以是不同的港口。根据该运输节点进行路径规划,得到路径方案集合,该路径方案集合总包括至少一个路径方案,该至少一个路径方案中的每个路径方案包括至少一个运输路径,以一个路径方案为例,该一个路径方案中包括至少一条运输路径,该一个路径方案中的所有运输路径覆盖该M个提货点,该一个路径方案中的其中一条运输路径可以覆盖该M个提货点中的L个提货点,L≤M。
在实际应用中,可根据历史路径数据或预设的权重调整规划路径,例如,可以根据历史路径方案得到港口到提货点或提货点之间的转移概率,也可以是通过随机分布或均匀分布初始化该提货点之间的转移概率,然后根据提货点之间的转移概率生成路径方案集合中的每个路径方案的每条运输路径。相对于现有方案中通过规则搜索获取路径方案,本申请实施例可以通过提货点之间的转移概率快速生成路径方案集合中的每个路径方案,提高得到路径方案的效率。
该路径方案集合中的每个路径方案还需要经过详细对比,才能确定出最合适的路径方案,该确定最合适的路径方案的条件可以包括,路径的长度,路径方案中集装箱的实装率或路径方案中集装箱的体积实装率与载重实装率的差值,其中,路径的长度越短,该路径方案越优;路径方案中集装箱的实装率越大,该路径方案越优;路径方案中集装箱的体积实装率与载重实装率的差值越小,该路径方案越优。
203、根据路径方案集合进行货物分配,以得到路径方案集合中每个路径方案的货物分配方案集合;
该路径方案集合中包括至少一个路径方案,因此还需要对该至少一个路径方案中的每个路径方案进行进一步地货物分配,对该路径方案集合中的每个路径方案中的每条运输路径进行货物分配,即将待运输货物分配到每个路径方案中每条运输路径对应的集装箱中,得到该至少一个路径方案中的每个路径方案的货物分配方案集合,即第一货物分配方案集合。
具体地,在进行货物分配时,因存在不同类型的货物,不同类型的货物的分布概率也会有差异,因此不能简单地通过分布算法进行货物分配,需要通过计算实际的不同类型的货物的概率分布来定义货物分配,在本申请实施例中,进行货物分配之前,首先可以对每个提货点的货物进行聚类,然后对同一类货物用一个概率分布来确定分配方式。聚类的参 考条件可以是货物的长度、宽度、高度、重量、最小接触面积,材质以及承压系数等,为提高效率,可使用部分条件进行聚类,例如,可使用长度、宽度、高度以及重量进行聚类。其中,聚类可以包括精确聚类与模糊聚类。在本申请实施例中,以长度、宽度、高度以及重量进行聚类为例,精确聚类可以是将四个特征完全相同的货物归为一类,该聚类为小粒度的聚类方式,聚类的速度也快。以此种方式进行聚类,可以提高货物分配时的准确性。当精确聚类产生的货物类别数太大,会增加个体,即一个路径方案的评价运算成本,包括增加运算时间,降低运算的效率等。因此,当精确聚类的类别数量大于预置的阈值时,还可以通过K均值(k-means)聚类、高斯混合模型或层次聚类法等聚类算法进行聚类运算。因此,每个提货点的货物都可以通过聚类算法得到各个种类的分布,并通过该各个种类的分布得到至少一个路径方案中的每个路径方案对应的货物分配方案集合中的每个货物分配方案。
204、根据快速装载模型确定每个路径方案的实装率;
在确定该至少一个路径方案中的每个路径方案的货物分配方案后,可根据快速装载模型确定至少一个路径方案中的每个路径方案的实装率,该实装率可以包括体积实装率与载重实装率。该快速装载模型为通过离线仿真数据进行离线训练得到,该离线仿真数据可以是通过三维装载计算得到的历史装载方案。其中,体积实装率为分配到集装箱的货物占用集装箱载货体积的比例,载重实装率为分配到集装箱的货物占用集装箱荷载重量的比例。
在实际应用中,可以从历史装载方案中抽取特征,将离线仿真数据转换为预置格式的训练数据,然后使用该训练数据训练预测模型,可通过该预测模型对待运输货物的装载方案进行预测,输出实装率。相比于现有技术使用三维装载进行在线计算,本申请实施例中的快速装载模型可以快速输出至少一个路径方案中的每个路径方案的实装率,提高得到目标运输方案的效率。
205、通过实装率对每个路径方案以及每个路径方案对应的货物分配方案进行整合评价,以确定目标运输方案;
在得到货物分配方案集合中每个路径方案的货物分配方案后,需要通过至少一个路径方案中的每个路径方案与对应的货物分配方案进行对比评价,选出目标路径方案,即该至少一个路径方案中的每个路径方案中目标的路径方案。具体评价的方式可以是,路径方案的路程越短越优,集装箱的实装率越大越优,集装箱的体积实装率与载重实装率差值越小越优,可通过不同条件,包括路径方案的长度、集装箱的实装率与集装箱体积实装率与载重实装率的差值,通过协同评价确定目标路径方案与对应的目标货物分配方案,即目标运输方案,该目标运输方案可以包括目标货运单所对应的最优或次优的运输方案。
在实际应用中,可通过对路径方案集合中的路径方案进行迭代的路径规划、货物分配、快速装载以及整合评价,在达到停止条件后,例如,在迭代次数达到阈值,或得到最优的路径方案与货物分配方案等后,输出目标运输方案。
在本申请实施例中,在获取到目标货运单后,根据该目标货运单进行路径规划,可以得到路径方案集合,该路径方案集合中包括至少一个路径方案,该至少一个路径方案中的每个路径方案都由至少一条运输路径组成,即至少一条运输路径组成完整的路径方案,然 后对该路径方案集合中的每个路径方案中的每条运输路径进行货物分配,得到至少一个路径方案中的每个路径方案的货物分配方案。相对于现有方案中使用禁忌搜索方法进行大量路径搜索,本申请实施例可以减小路径搜索的次数,提高得到路径方案的效率。之后通过快速装载模型快速输出货物分配方案集合中的每个路径方案的实装率,包括体积实装率与载重实装率,之后对货物分配方案以及路径方案集合中的至少一个路径方案中的每个路径方案进行进一步整合评价,得到目标路径方案与该目标路径方案对应的目标货物分配方案,该目标路径方案与目标货物分配方案组成目标运输方案。其中,快速装载模型为通过离线仿真数据得到,该离线仿真数据包括通过三维装载计算得到的历史装载方案,相比于现有方案中使用三维装载进行在线运算得到装载方案,本申请实施例可以快速输出实装率,并根据该实装率对至少一个路径方案中的每个路径方案以及对应的货物分配方案进行整合评价,以得到目标运输方案,可以提高得到目标路径方案与目标货物分配方案,即确定目标运输方案的效率。
前述对本申请实施例中确定运输方案的方法的流程就行了说明,下面对本申请实施例中确定运输方案的方法进行更详细的说明,请参阅图3,本申请实施例中确定运输方案的方法的另一个实施例示意图。
其中,确定运输方案的方法的流程可以是,在接收到目标货运单301后,进行超参数初始化302,对提货点转移超参数进行初始化,提货点转移超参数初始化完成后,可以根据该提货点转移超参数得到提货点转移概率分布,并根据该提货点转移概率分布进行路径规划303,得到路径方案集合,然后对路径方案集合中的每个路径方案进行货物分配304,得到路径方案集合中的每个路径方案的货物分配方案集合,然后进行快速装载306,对路径方案以及货物分配方案进行货物的快速装载,输出实装率,之后进行个体评价305,对路径方案与货物分配方案进行整合以及评价,进一步对路径方案集合以及货物分配方案进行整合评价得到目标路径方案与对应的目标货物分配方案,即仿真结果309。其中,快速装载步骤中的快速装载模型由对离线仿真307得到的离线仿真数据308,进行离线训练得到。在实际应用中,可以重复进行步骤303-步骤306,进行重复的路径方案探索与货物分配探索,直到停止条件达到,得到目标路径方案与对应的目标货物分配方案,也可以是直接将该路径方案集合中的其中一个路径方案以及对应的货物分配方案作为目标路径方案与对应的目标货物分配方案,具体可以根据实际设计需求进行调整,具体此处不作限定。
下面对本申请实施例中的步骤进行具体说明。
301、目标货运单。
首先,获取到一个目标货运单301,可通过该目标货运单中获知运输节点以及待运输货物,该运输节点包括港口以及提货点,该待运输货物分布在每个提货点,其中,港口可以包括货运起点以及货运终点,货运起点与货运终点可以是一个港口,也可以是不同的港口,例如,该目标货运单可以是指示港口1为货运起点,将分布在提货点1与提货点2的待运输货物运输到港口2。
302、超参数初始化。
随后进行提货点转移超参数初始化,根据初始化后的提货点转移超参数得到提货点转移概率分布,该提货点转移概率分布包括集装箱从港口到提货点或提货点之间转移的概率,例如,从提货点D1到提货点D2的概率。
实际应用中,贝叶斯估计算法的具体流程可以是,首先为待估计量分配一个先验分布,然后结合实验数据,根据贝叶斯公式计算得到后验分布,之后由后验分布获取待估计量的估计值。因此,本申请实施例中确定货物分配方案的方法也可以通过贝叶斯估计算法计算提货点转移超参数的。其中,先验分布可以由历史数据或用户经验得到,在实际的业务系统中,积累了大量的历史路径数据,可以从这些历史路径数据中提取大量待估计量的样本,并利用这些样本对超参数进行估计。该历史路径数据可以作为先验数据,此外,在实际应用中,还可以根据实际调度人员的经验对先验数据进行调整。而当从历史路径数据中获取的待估计量的样本数量低于预设的阈值时,无法对超参数进行估计,可以使用启发式算法进行估计,下面分别对本申请实施例中可以使用的贝叶斯估计算法以及启发式算法进行详细说明。
1、贝叶斯估计算法。
先验分布可以通过采样多项分布,或二项分布等来生成的,例如,当目标货运单中包含k个提货点D 1,D 2,D 3…,D k以及一个港口Port,港口到提货点之间的转移概率即一个多项分布,参数θ=(θ 12,...,θ k),从港口到提货点的转移概率如图4所示,一个港口至一个提货点,或两个提货点之间表示一种转移模式,起点为当前起点,终点为当前跳转点,如果当前跳转点属于目标货运单中需求的提货点,且不同与当前起点,那么这种转移模式为有效转移模式,即根据目标货运单中的提货点确定该转移模式为有效的转移模式。预设参数θ服从Dirichlet(α)分布,α即为超参数,则后验分布也服从Dirichlet分布。先验分布与后验分布间仅仅是超参数发生了变化,因此可以简化后验分布的计算,在本申请实施例中可以将θ看作服从Dirichlet(α)分布。
每一条历史路径数据可以表示历史的运输路径,例如,现有一条历史路径数据为:Port→D 1→D 3→D 6→Port,即表示一条运输路径为,从港口出发,依次经过D 1,D 3,D 6后回到港口。具体的初始化流程可以包括:首先选定目标货运单中的一个提货点或港口作为当前起点,筛选出历史路径数据,如果货运单中存在k个提货点,则可以确定k+1份历史路径数据。例如,若选择港口作为当前起点,则货运单中任意一个提货点作为当前跳转点都可以组成一种有效转移模式,筛选出的历史路径数据中需存在至少一种对应的有效转移模式,例如,目标货运单中,选定从港口至D 1,则筛选出的历史路径数据中至少有一条路径中包括从港口至D 1,如图5所示,假设目标货运单中包括港口以及两个提货点D 1以及D 4,则筛选出的历史路径数据中包含从港口转移到D 1或从港口转移到D 4的历史路径都作为有效转移模式。
筛选后的历史路径数据,按照预置的顺序排列,然后按照预设的窗口大小,依次取S条历史路径数据,如图6所示,可以预设S=3,然后每个窗口3条历史路径,若最后的数量不足3条,则可以归为前一个窗口中。若当前窗口的数量少于预设的第一阈值N min,则可以通过启发式算法进行计算。此处先对窗口数量不少于第一阈值N min的情况进行介绍,对少于第一阈值N min的情况,即启发式算法另行进行介绍。
在各个窗口中的历史路径数据分别统计有效转移模式出现的次数,并基于此计算对应多项式分布的参数,假设有t个窗口,t≥N min,则可以获取t个θ的样本。具体的统计过程可以如图7所示,选择其中一个窗口的路径数据,对该窗口的历史路径数据进行统计,例如,当前起点为港口,当前跳转点为D 1,则统计从港口至D 1的历史路径数量。在统计港口至提货点或提货点之间的转移次数后,进行归一化计算,得到多项分布的参数θ。
在统计完所有的窗口之后,基于得到的t个θ样本,进行极大似然估计,计算出超参数α的估计值。
2、启发式算法。
若当前窗口的数量少于预设的阈值N min,则可以通过启发式算法进行超参数计算。若目标货运单中有k个提货点,以其中一个提货点D i为例,货物的总体积为V i,总重量为W i,一个集装箱能装载的最大体积为V,最大载重为W。
启发式算法的具体的算法流程可以包括:
首先计算全部装载完提货点D i的货物所需的最少车辆数,此处将一个车辆看作一个集装箱,min_car=Max(V i/V,W i/W);在计算出最少车辆min_car后,计算向上取整后的最少车辆数,min_car_ceiled=Ceil(min_car)。
然后计算提货点D i的能量系数P Di,该能量系数为提货点对应的多项分布的超参数,
Figure PCTCN2018108534-appb-000025
若不存在保税仓,保税仓即需要交税才能出的仓库,且只能空的集装箱进入,提货点D i对应的多项式分布的超参数为:
Figure PCTCN2018108534-appb-000026
港口对应的多项分布超参数为:
Figure PCTCN2018108534-appb-000027
在实际场景中,提货点中还可能存在保税仓,若提货点中存在保税仓,假设该保税仓为D i,则将港口对应的多项式分布的超参数设置为: α Port=(ε (1),...,ε (i-1),1,ε (i+1)),提货点D i对应的超参数为1,其余提货点的超参数可以设置为ε,且该其余提货点对应的ε可以设置为一个非常小的数,例如,0.00001,0.0000001等。
在实际应用中,积累了大量的高质量历史路径数据,可以为超参数初始化提供有效的数据基础,本申请实施例提供的确定运输方案的方法可以使用历史路径数据进行超参数初始化,通过历史路径数据可以得到更准确的路径方案,并且可以提高后续得到路径方案的效率。
303、路径规划。
在提货点转移超参数初始化完成后,进行路径规划。需要对运输的路径进行规划,确定运输的路径方案以及集装箱的数量。超参数初始化完成后,可以得到提货点转移超参数矩阵,可以利用该提货点转移超参数矩阵生成提货点转移概率矩阵,包括港口与提货点之间,或提货点之间的转移概率分布。其中,转移概率为从一个港口转移到一个提货点或从一个提货点转移到另一个提货点的概率,可以基于历史路径数据得到,例如,如图8所示,从图中可知,从提货点A转移到提货点B的概率为0.2,从提货点A转移到提货点C的概率为0.6,以此类推。通过贝叶斯估计算法或启发式算法计算得到超参数矩阵后,本申请实施例可以使用Dirichlet分布取期望值的方式获取转移概率矩阵的估计值,取期望值得到转移概率矩阵。在实际应用中,若需要偏向与其中一条路径,则可以提高该条路径中的转移概率所占的权重,可以通过自行设置转移概率以提高该条路径中的转移概率所占的权重。
得到转移概率矩阵后,通过采样生成路径方案集合,即路径方案种群。采样得到路径方案的示例请参阅图9,通过转移概率矩阵可以获取到港口与提货点或提货点之间转移的概率,然后通过该转移概率矩阵确定路径方案集合。例如,当前起点为港口,首先进行步骤1,从港口选择下一个跳转点,通过采样计算,得到的转移概率为0.23,确定为提货点A,之后的提货点确定方式与确定提货点A类似,随后步骤2与步骤3依次确定提货点C以及港口,得到路径方案为:港口→提货点A→提货点C→港口。
可以通过计算得到每个方路径方案被选中的概率P i,i表示第i个个体,即一个路径方案,可以通过评价函数对至少一个路径方案中的每个路径方案进行评价,具体计算个体概率的公式可以是:
Figure PCTCN2018108534-appb-000028
其中,j表示第j个个体,n表示个体的总数,f(i)为个体的得分,该个体得分函数为个体评价步骤中的评价函数,将在步骤305的个体评价步骤中详细描述,此处不作阐述。因此,通过该个体概率的计算公式可知,路径方案被选中的概率与路径方案的评价有关,路径方案评价越高,被选中的概率越大,可以理解为,路径方案的评价越高,该路径方案越优。
然后基于计算得到的至少一个路径方案中的每个路径方案的概率,选出m个个体,更新路径方案集合,此外,除了更新路径方案集合,还可以对超参数矩阵进行更新,以便之 后提高之后对超参数计算的效率,且若规划得到的路径评价都不高,则可以继续通过更新后的超参数矩阵继续进行路径规划,以得到更准确的路径方案。利用选出的该m个个体进行贝叶斯估计更新超参数矩阵,更新超参数矩阵的具体示例如图10所示,例如,m个个体中提货点A→提货点B的转移模式出现了1次,则对应超参数矩阵中从提货点A转移至提货点B的值加1,原始值为0.6,那么增加后的值为1.6,其他的转移计算以此类推。在更新完超参数矩阵后,若选出的m个个体中在经过整合评价后不存在合适的路径方案,则可以继续通过该超参数矩阵得到提货点转移概率矩阵,然后重新进行路径规划。
例如,在超参数初始化完成后,根据初始化后的提货点转移超参数得到提货点转移概率矩阵,并根据该提货点转移概率矩阵确定第一路径方案集合,该第一路径方案集合中包括至少一个路径方案,然后根据该第一路径方案集合中的每个路径方案进行货物分配,得到该第一路径方案集合中的每个路径方案的货物分配方案集合中的每个货物分配方案,之后对该第一路径方案集合中的每个路径方案以及该第一路径方案集合中的每个路径方案对应的第一路径方案集合中的每个路径方案进行整合评价,根据整合评价的结果未得到合适的目标运输方案,此时,可以根据第一路径方案集合中的每个路径方案,对提货点转移超参数进行更新,并通过更新后的提货点转移超参数获取更新后的提货点转移概率矩阵,根据更新后的提货点转移概率矩阵重新对运输路径进行规划,得到第二路径方案集合,该第二路径方案集合包括至少一个路径方案,之后再对该第二路径方案集合中的每个路径方案货物分配以及整合评价,以得到目标运输方案。
304、货物分配。
在进行路径规划,得到路径方案集合中的每个路径方案之后,可以对该路径方案集合中的每个路径方案中的每条运输路径进行货物分配,对路径方案集合中的每个路径方案中的每条运输路径进行货物分配,以得到每条运输中集装箱所装载的货物。
一个路径方案集合可以看作为一个种群,种群中的每个个体代表一个路径方案,当个体生成后,路径方案中的每条运输路径都已确认,因此所需的集装箱数量,以及每个集装箱的路径都已经确定,在实际应用中,可以看作集装箱与运输路径一一对应,以及集装箱与货运车一一对应。那么,此时还需要对装入集装箱中的货物进行分配,确定每个集装箱所装载的货物。
在对货物进行分配时,由于存在类型不相同的货物,例如,长度不同,宽度不同,高度不同,或重量不同等。因此无法用简单的分布来描述提货点的货物分配,本申请实施例可以将货物进行聚类,然后通过聚类对货物进行分配。聚类的参考条件可以是货物的长度、宽度、高度、重量、最小接触面积,材质以及承压系数等,为提高效率,可使用部分条件进行聚类,例如,可使用长度、宽度、高度以及重量进行聚类。其中,聚类可以包括精确聚类与模糊聚类。在本申请实施例中,以长度、宽度、高度以及重量进行聚类为例,精确聚类可以是将四个特征完全相同的货物归为一类,该聚类为小粒度的聚类方式,聚类的速度也快。以此种方式进行聚类,可以提高货物分配时的准确性。当精确聚类产生的货物类别数太大,例如,货物的长度可分为多个类别,宽度也可以分为多个类别等,会增加个体的运算成本。因此,当精确聚类的类别数量大于预置的阈值时,还可以通过K均值 (k-means)聚类、高斯混合模型或层次聚类法等聚类算法进行聚类运算。在本申请实施中,通过聚类的方法对货物进行分类,在分类后对货物进行分配,可以提高后续得到货物装载方案的准确性。
例如,货物分配的示意图可以如图11所示,其中,每个提货点都有等待运输的货物,例如图11中A中的待运输货物的种类包括货物类1、货物类2或货物类m A,分别对应提货点第一货物分配超参数θ A1、θ A2以及
Figure PCTCN2018108534-appb-000029
然后通过采样对该每个提货点的待运输货物进行货物分配,为每个提货点都生成一个代表货物分配方式集合的类别种群,该类别种群中的个体即表示该提货点中的货物分配方式。例如,提货点A有两个集装箱经过,图11中,那么种群1中的个体就表示提货点A中的货物分配到该两个集装箱的方案,即表示每个类别的货物在每个集装箱上分配的数量。由于一个种群只能代表一个提货点的货物分配方式,不能代表完整的货物分配方案,因此,需要将所有提货点的货物分配方式进行结合,形成完整方案后才能对方案进行评价,即协同评价,以确定每个路径方案的货物分配方案。例如,在评价种群1中的个体时,需要种群2以及种群3都提供一个代表个体,种群1中的个体首先与种群2与种群3的代表个体合并成一个完整的货物分配方案,然后再执行评价的步骤。代表个体可以是在种群中随机选取的,也可以是种群中的最优个体,还可以是在随机选取进行多次整合后确定的最优个体。可以通过个体评价,即步骤305中的评价函数对货物分配方案进行评价,根据评价函数计算货物分配方案集合中的每个路径方案的得分,该评价函数将在以下步骤305的详细说明中介绍,此处不作阐述。
例如,以一个路径方案为例,在形成第一货物分配方案集合,并对第一货物分配方案集合中的每个货物分配方案进行评价进行打分后,若该第一货物分配方案中不包括得分高于第二阈值的货物分配方案,则可以通过得到的第一货物分配方案集合更新提货点的第一货物分配超参数,以得到提货点的第二货物分配超参数,并通过该第二货物分配超参数重新进行对待运输货物进行分配,并得到该路径方案对应的第二货物分配方案集合,继续对该第二货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标货物分配方案与对应的目标路径方案。在实际应用中,可以重复进行货物分配,直到达到停止条件,例如,迭代次数达到预置次数,整合评价的结果得分高于第二阈值的货物分配方案达到预置的数量等。确定目标货物分配方案后停止。
具体地,为提高本申请实施例中确定运输方案的方法的效率,在得到个体得分后,采用与路径规划303步骤中类似的算法进行个体选择概率的计算,然后根据个体选择概率选取n个个体,并以此更新各个提货点的原始提货点超参数矩阵,得到目标提货点的超参数矩阵。即在确定每个提货点的货物分配方案后,还可以包括进一步地进行学习的过程,可用于重复对货物进行分配。更新货物分配超参数的具体过程可以如图12所示,在确定n个货物分配方案,且确定该货物分配方案都能将货物装入集装箱后,通过该n个货物分配方案更新。例如,货物分配方案中,标号为1的为同一类别,标号为2的为同一类别,标号为3的为同一类别,类1,即类别1在第一个集装箱,即车1上分配了2个,则将对应的车1、类1的超参数加2,原始提货点超参数矩阵中车1、类1的超参数为1.0,那么加2 后的目标提货点超参数矩阵中车1、类1的超参数为3.0,其他的类别以及集装箱超参数以此类推。
305、个体评价。
在完成货物分配,得到货物分配方案后,还需要进行个体评价,个体评价即对路径方案以及货物分配方案进行整合,进行对路径方案以及货物分配方案的评价。评价的指标包括:路径方案的路径长度,集装箱的实装率,集装箱的体积实装率与载重实装率的差值等。其中,路径方案的路径越短,代表车辆行驶的路径,即集装箱的运输路径越短,可以节省运输成本,提高运输效率;集装箱的实装率越大,即集装箱装载的货物越多,即相同货物所需集装箱的数量越少,也可以节约运输成本以及提高运输效率;集装箱的体积实装率与载重实装率差值不能超过阈值,例如,如图13所示,相同重量的货物,当其中一个集装箱的体积实装率为35%,载重实装率为95%,另一个集装箱的体积实装率为95%,载重实装率为35%,此外还有一个集装箱装载剩余的货物,因此需要3个集装箱;当其中一个集装箱的体积实装率为75%,载重实装率为85%,另一个集装箱的体积实装率为80%,载重实装率为75%,则仅需两个集装箱。因此,集装箱的体积实装率与载重实装率的差值越小,越可以节约集装箱资源。在实际应用中,集装箱的实装率可以由快速装载,即步骤306得到。
可以通过观察历史数据上的集装箱的装载记录,获取到实装率不均衡的集装箱,体积实装率与载重实装率都分布在货运单所有货物的平均体积实装率
Figure PCTCN2018108534-appb-000030
和平均载重实装率相宜的两侧,即
Figure PCTCN2018108534-appb-000031
其中,
Figure PCTCN2018108534-appb-000032
Figure PCTCN2018108534-appb-000033
r Vi为第i个集装箱的体积实装率,r Wi为第i个集装箱的载重实装率。
因此,本申请实施例提供了一种评价函数,在评价路径的长度与集装箱的实装率的同时,还能对体积实装率与载重实装率的差值进行对比。
该评价函数为:
Figure PCTCN2018108534-appb-000034
其中,
Figure PCTCN2018108534-appb-000035
为路径方案向量,包括m个集装箱,
Figure PCTCN2018108534-appb-000036
为该m个集装箱的体积实装率向量,
Figure PCTCN2018108534-appb-000037
为该m个集装箱的载重实装率向量;α,β与γ为权重参数,r Vi为第i个集装箱的体积实装率,r Wi为第i个集装箱的载重实装率,
Figure PCTCN2018108534-appb-000038
为该m个集装箱的平均体积实装率,
Figure PCTCN2018108534-appb-000039
为该m个集装箱的平均载重实装率。
可通过该评价函数对路径方案进行评价,筛选出一些无效或评价低的路径方案,更新 路径方案集合。
306、快速装载。
在进行步骤305个体评价时,可以通过快速装载模型对路径方案集合中的每个路径方案进行快速装载仿真得到每个路径方案对应的货物分配方案集合中的每个路径方案的实装率。可以根据货物分配得到的货物分配方案,即装入集装箱的货物集合,快速得到该集装箱的实装率。快速装载模型为根据离线仿真数据308训练得到,该离线仿真数据包括通过三维装载得到的历史装载方案,该离线仿真数据可以通过离线仿真307得到,离线仿真307的步骤与在本申请实施例中确定目标运输方案的步骤类似。
在实际应用中,除了通过快速装载模型确认路径方案集合中每个路径方案进行快速装载得到至少一个路径方案中的每个路径方案对应的每个货物分配的实装率,还可以根据实装率对装入集装箱的货物进行估计,得到可行解,即判断每个货物是否能装入集装箱,若存在不能装入集装箱的剩余货物,则需要对剩余货物进行进一步地货物分配,以得到能够对待运输货物进行完整运输的目标运输方案。
因此,步骤304、步骤305与步骤306共同组成货物分配方案探索的步骤,该货物分配方案探索与步骤303组成路径方案探索的步骤,在确定目标运输方案,即目标运输路径与该目标运输路径对应的货物分配方案后,输出目标运输方案,科根据该目标运输方案对待运输货物进行运输。
在实际应用中,可以通过重复步骤303-步骤306得到目标路径方案与对应的目标货物分配方案,即目标运输方案,并通过该目标路径方案与对应的目标货物分配方案对待运输货物进行运输。
具体地,快速装载模型与装载方案的具体获取流程可以如图14所示,分为离线训练部分与在线预测部分。
首先对离线训练部分进行说明,具体如图14所示,通过离线仿真获取大量的高质量的离线仿真数据,对该离线仿真数据中特征抽取,即集装箱的货物分配方案转化为一组特征向量,利用从离线仿真数据训练预测模型,该预测模型用于输出输入数据的体积实装率与载重实装率。该预测模型可以包括:线性回归模型,岭回归模型,LASSO模型,支持向量机模型,随机森林模型,XgBoost模型或人工神经网络模型等。
离线训练与在线预测都需要进行特征抽取,离线训练与在线预测的特征抽取过程类似,区别在于离线训练是从离线仿真数据中抽取特征,在线预测是从分配给集装箱的货物集合,即从每个货物分配方案中抽取特征。抽取特征的具体流程包括,首先抽取单个货物的特征,得到单个货物的特征向量,即第二特征向量,该单个货物的特征包括:货物的长度、宽度、高度以及重量,此外还可以包括最小接触面积,材质,承压系数等。其中,由于材质与承压系数是类别型变量,即与货物的类别有关系,且与类别维度不大,因此本申请实施例中可以采用one-hot编码方式来表示材质与承压系数。例如,有4种材质,其中一个货物属于材质1,则以one-hot编码方式的形式标识为:(1,0,0,0)。
Figure PCTCN2018108534-appb-000040
数量和该提货点的量化值t i
接下来对集装箱经过各提货点的
Figure PCTCN2018108534-appb-000041
进行加权组合,得到最终的特征向量,即该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量
Figure PCTCN2018108534-appb-000042
其中,n表示提货点的数量,r i表示提货点i在路径方案中的顺序的量化值,r i与t i为基于历史路径的数据,由对应分析的方法进行分析得到。
在离线训练时,将离线仿真数据中的每项历史装载方案数据都转换为预置格式的训练数据,该预置格式可以是(特征向量,实装率),然后进行模型训练,以得到快速装载模型。需要训练两个模型,包括一个预测体积实装率的模型以及预测载重实装率的模型,即快速装载模型包括预测体积实装率的模型以及预测载重实装率的模型。
在进行在线预测时,将分配给集装箱的货物集合转换为预置格式的特征向量,然后输入训练好的模型中,得到对应的输出值。
在实际应用中,分配给集装箱的货物存在不能装入集装箱的情况,且在更新超参数时,只能考虑装入集装箱的部分,即可行解,用装入的货物数量来更新超参数。货物分配时确定的货物分配方案中包括装入集装箱的货物集合,实装率预测模型只能预测该货物集合的实装率,而无法确定该货物集合中的货物能否装入集装箱。因此需要根据实装率的信息预估能装入集装箱的货物集合,在本申请实施例中,可以使用线性规划来实现该问题。
例如,有n个货物分配至一个集装箱,第i个货物的体积为v i,重量为w i,p i为第i个货物能装入集装箱的概率,由实装率预测模型输出的体积实装率为r v,载重实装率为r w,装入集装箱的货物的体积与重量的预测值分别为V和W,可以定义线性规划为:
Figure PCTCN2018108534-appb-000043
求解以上公式可以得到一组p i,按照p i由大到小的顺序货物集合中的货物进行排序,然后按照顺序依次取货物,知道去除的货物总体积超过V,或总重量超过W,则给出的货物集合即为一个可行解的估计,即可装入集装箱的货物。
在本申请实施例中,在线仿真得到可行解后,还可以进行后期处理,即在确定目标运输方案之后,还可以进一步确定合适的集装箱型号并生成最终装载方案。例如,可以使用集装箱的箱型可以是40HQ,利用图3中的货物分配探索与路径方案探索可以获取到每个集装箱装载的货物体积与重量,则可以选择的箱型的体积与载重不能小于货物的体积与重量。在实际应用中,确定集装箱的箱型之后,也可以利用三维装载确定待运输货物在集装箱内的最终装载方案,以便将所有货物准确地装入集装箱中,提高实际装载货物时对货物进行装载的效率,该集装箱的箱型可以选择能够成功装入且成本最低的型号,以降低集装箱的成本。
若没有剩余的货物,即无法装入集装箱的货物,则确定的合适箱型与最终装载方案为完整方案,若有剩余的货物,即存在无法装入集装箱的货物,则可以将剩余的货物重新生 成一个虚拟货运单,重复进行货物分配方案的探索,得到对剩余货物的运输方案,以完成剩余的货物的运输。
在本申请实施例中,训练快速装载模型的离线仿真数据可以由离线仿真得到,离线仿真的具体流程与在线仿真类似,区别包括,离线仿真使用历史货运单进行仿真,在线仿真使用当前货运单进行仿真;离线仿真使用三维装载生成装载方案,在线仿真通过离线训练的快速装载输出实装率等,具体请参阅图15,本申请实施例中确定运输方案的方法的另一种实施例示意图。
在实际应用中,由于在线仿真的中间数据占用内存大,无法进行保存,因此在进行离线训练快速装载模型时,需要对历史货运单重新进行路径规划、货物分配以及个体评价的步骤。在进行离线仿真时,首先获取历史货运单,该历史货运单中包括历史提货点信息以及历史待运输货物的信息,然后根据该历史货运单进行路径规划,以得到该历史货运单对应的历史路径方案,然后根据该历史路径方案进行货物分配得到每个历史路径方案对应的历史货物分配方案集合,之后通过三维装载运算得到么给了是路径方案对应的历史货物分配方案集合中每个货物分配方案的实装率以及装载方式,然后根据每个货物分配方案的实装率对该至少一个历史路径方案中的每个路径方案以及对应的历史货物分配方案集合中的每个货物分配方案进行整合评价,得到历史运输方案。
其中,离线仿真中的步骤包括,超参数初始化、路径规划、货物分配以及个体评价的步骤与前述图3中在线仿真的超参数初始化、路径规划、货物分配以及个体评价的步骤类似,具体此处不再赘述。下面对离线仿真与在线仿真的区别步骤进行阐述。
在离线仿真时,为得到更准确的数据,可以使用三维装载算法进行实装率的计算,即得到具体的装载方式。当进行货物分配完成货物的分发,即为集装箱分配货物后,即可使用三维装载运算得到集装箱的体积实装率与载重实装率,并可使用该体积实装率与载重实装率对路径方案进行评价,对路径方案进行评价的步骤与前述图3中的个体评价步骤305类似。本申请实施例可以使用基于Corner Point和Extreme Point的启发式算法完成货物的模拟装载,如图16所示,在装载货物之前,首先会确定集装箱的空间状态,然后获取一系列的候选放置点,之后会逐一进行尝试放置,直到找到合适的放置点。与Corner Point算法相比,Extreme Point算法因会扫描被货物架空的区域,因此会产生更多的候选点,可以得到精确的实装率与装载方案,可以提高集装箱的使用率。
而三维装载时一个序列过程,即只能按顺序对货物进行模拟装载,不能并行处理多个货物,因此在货物较多时,需要耗费更多的时间进行模拟装载,因此,本申请实施例在离线仿真与后期处理时使用三维装载仿真,可以提高得到货物装载方案以及输出实装率的效率,在离线仿真时使用三维装载算法对实装率进行计算,可以得到更准确的实装率,在后期处理时使用三维装载算法得到装载方案,可以在运输时获知货物的装载方式,提高运输的效率。
前述对本申请实施例提供的确定运输路径的方法进行了详细说明,下面对本申请实施例提供的装置进行说明,首先对确定装置进行说明,请参阅图17,该确定装置可以包括:
获取模块1701,用于获取至少一个路径方案以及该至少一个路径方案中的每个路径方 案对应的第一货物分配方案集合,该至少一个路径方案中的每个路径方案为针对待运输货物进行运输而规划的运输路径,该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合包括至少一个货物分配方案,该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该待运输货物进行分配的方案;
快速装载模块1702,用于根据快速装载模型确定该至少一个路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,该快速装载模型为通过离线仿真数据进行离线训练得到,该离线仿真数据包括通过三维装载算法计算得到的历史装载方案,该实装率为某一货物分配方案中装入集装箱的货物占用该集装箱的比例;
评价模块1703,通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,该目标运输方案包括目标路径方案与该目标路径方案对应的目标货物分配方案。
可选地,在一些可能的实施方式中,该获取模块1701,可以包括:
获取子模块17011,用于获取目标货运单,该目标货运单包括运输节点信息以及待运输货物信息,该运输节点信息包括货运起点、货运终点以及M个提货点,该待运输货物信息包括分布在该M个提货点的该待运输货物的信息,该M为正整数;
路径规划子模块17012,用于根据该运输节点信息确定该至少一个路径方案,该至少一个路径方案中的每个路径方案包括至少一条运输路径,该至少一条运输路径中的每条运输路径包括货运起点、货运终点以及该M个提货点中N个提货点,该至少一个路径方案中的每个路径方案均覆盖该M个提货点,该N为正整数,且N≤M;
货物分配子模块17013,用于为该至少一个路径方案中的每个路径方案中的每条运输路径进行该待运输货物的分配,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
可选地,在一些可能的实施方式中,该路径规划子模块17012,具体用于:
若历史路径数据的数量大于第一阈值,则基于该历史路径数据对该M个提货点的转移超参数进行初始化,以得到超参数矩阵;
通过该超参数矩阵确定该M个提货点的转移概率分布,该转移概率分布包括运输路径中的集装箱在该货运起点与该M个提货点之间、该货运终点与该M个提货点之间或该M个提货点之间的转移概率;
根据该转移概率分布确定该至少一个路径方案中的每个路径方案中的每条运输路径,以得到该至少一个路径方案。
可选地,在一些可能的实施方式中,该确定装置还可以包括:
初始化模块1704,用于若该历史路径数据的数量不大于该第一阈值,则基于启发式算法对该M个提货点的转移超参数进行初始化,以得到该超参数矩阵。
可选地,在一些可能的实施方式中,该货物分配子模块17013,具体用于:
对该M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,该聚 类条件包括货物的长度、宽度、高度以及重量;
通过该M个提货点中每个提货点的第一货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第一货物分配方式集合,该M个提货点中每个提货点的第一货物分配超参数为对该M个提货点中每个提货点进行货物分配的超参数,该M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式;
从该M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
可选地,在一些可能的实施方式中,该快速装载模块1702,具体用于:
获取该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,该第一特征向量用于指示某一货物分配方案中待运输货物的特征值;
将该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第一特征向量输入该快速装载模型中,以得到该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中每个货物分配方案的实装率,该实装率包括体积实装率与载重实装率,该体积实装率包括该至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的体积占用集装箱荷载体积的比例,该载重实装率该至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的重量占用集装箱的荷重的比例。
可选地,在一些可能的实施方式中,该快速装载模块1702,具体用于:
获取该待运输货物中的每个货物的第二特征向量,该待运输货物中的每个货物的第二特征向量包括对应货物的长度、宽度、高度以及重量;
根据该待运输货物中每个货物的第二特征向量计算出该M个提货点中的每个提货点分布的货物针对该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量,该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量包括该待运输货物中的每个货物的第二特征向量的均值与协方差;
对该至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量进行加权组合得到对应的该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量。
可选地,在一些可能的实施方式中,其特征在于,该评价模块1703,具体用于:
通过预置的评价函数以及该实装率,对所获取到的所有货物分配方案进行得分计算;
若该所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从该得分高于该第二阈值的货物分配方案中确定该目标货物分配方案,以及将该目标货物分配方案对应的路径方案作为该目标路径方案;
通过该目标货物分配方案与该目标路径方案确定该目标运输方案。
可选地,在一些可能的实施方式中,该评价函数包括:
Figure PCTCN2018108534-appb-000044
其中,该
Figure PCTCN2018108534-appb-000045
为路径方案向量,m为集装箱的数量,
Figure PCTCN2018108534-appb-000046
为m个集装箱的体积实装率向量,
Figure PCTCN2018108534-appb-000047
为该m个集装箱的载重实装率向量;该α、该β与该γ为权重参数,该r Vi为第i个集装箱的体积实装率,该r Wi为第i个集装箱的载重实装率,该
Figure PCTCN2018108534-appb-000048
为该m个集装箱的平均体积实装率,该
Figure PCTCN2018108534-appb-000049
为该m个集装箱的平均载重实装率。
可选地,在一些可能的实施方式中,该评价模块1703,还用于:
若该所有货物分配方案中不包括得分高于该第二阈值的货物分配方案,则通过该M个提货点中每个提货点的第二货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第二货物分配方式集合,该M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该M个提货点中每个提货点的第二货物分配超参数为通过该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案对该M个提货点中每个提货点的第一货物分配超参数进行更新得到;
从该M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案,该至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该待运输货物进行分配的方案;
通过该评价函数以及该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率,对该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案进行得分计算,该至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率由该快速装载模型得到。
可选地,在一些可能的实施方式中,该确定装置还可以包括:
后期处理模块1705,用于通过该实装率对该至少一个路径方案中的每个路径方案与该至少一个路径方案中的每个路径方案的货物分配方案进行整合评价,以确定目标运输方案之后,根据该目标货物分配方案和该目标路径方案,确定该目标路径方案中每条运输路径的集装箱的型号;
三维装载模块1706,用于根据该后期处理模块1705确定的该目标路径方案中每条运输路径的集装箱的型号以及三维装载算法生成装载方案,该装载方案为该待运输货物在该目标路径方案中每条运输路径中的集装箱内的装载方式。
可选地,在一些可能的实施方式中,该确定装置还可以包括:
确定模块1707,在该通过该实装率对该至少一个路径方案中的每个路径方案以及该至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案之前,用于若该实装率确定该M个提货点中的L个提货点还包括未分配到该集装箱的剩余货物,则为该剩余货物确定剩余货物路径方案和剩余货 物分配方案,该L≤该M,该L为正整数;
该评价模块1703,还用于通过该实装率对该至少一个路径方案中的每个路径方案与对应的第一货物分配方案集合中的每个货物分配方案,以及该剩余货物路径方案和该剩余货物分配方案进行整合评价,以确定目标运输方案。
下面对本申请实施例中的训练装置进行说明,请参阅图18,本申请实施例中训练装置的一个实施例示意图,可以包括:
获取模块1801,用于获取离线仿真数据,该离线仿真数据包括通过三维装载计算得到的历史装载方案与历史实装率;
该获取模块1801,还用于从该离线仿真数据中获取特征向量,该特征向量包括该历史装载方案对应的历史运输货物的特征值;
转换模块1802,用于将该特征向量转换为预置格式的训练数据;
训练模块1803,用于通过该训练数据训练预测模型,以得到快速装载模型,该快速装载模型用于输出每个运输路径的货物分配方案集合中每个货物分配方案的实装率,该实装率为该每个货物分配方案中装入集装箱的货物占用该集装箱的比例。
可选地,在一些可能的实施方式中,该预置格式为:(特征向量,历史实装率)。
可选地,在一些可能的实施方式中,该预测模型包括:线性回归模型、岭回归模型、LASSO模型、支持向量机模型、随机森林模型、XgBoost模型或人工神经网络模型。
可选地,在一些可能的实施方式中,该获取模块1801,可以包括:
获取子模块18011,用于获取至少一个历史路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合,该至少一个历史路径方案中的每个路径方案为针对历史运输货物进行运输而规划的运输路径,该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合包括至少一个货物分配方案,该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该历史运输货物进行分配的方案;
三维装载子模块18012,用于根据三维装载算法确定该至少一个历史路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,该实装率为某一货物分配方案中装入集装箱的货物占用该集装箱的比例;
评价子模块18013,通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,该目标运输方案包括目标路径方案与该目标路径方案对应的目标货物分配方案。
在本申请实施例中,可以在训练快速装载模型时使用三维装载算法进行计算,得到历史路径方案对应历史装载方案,可以准确输出历史路径方案对应的货物分配方案的实装率。
可选地,在一些可能的实施方式中,该获取子模块18011,包括:
获取单元180111,用于获取历史货运单,该历史货运单包括运输节点信息以及历史运输货物信息,该运输节点信息包括货运起点、货运终点以及M个提货点,该历史运输货物 信息包括分布在该M个提货点的该历史运输货物的信息,该M为正整数;
路径规划单元180112,用于根据该运输节点信息确定该至少一个历史路径方案,该至少一个历史路径方案中的每个路径方案包括至少一条运输路径,该至少一条运输路径中的每条运输路径包括货运起点、货运终点以及该M个提货点中N个提货点,该至少一个历史路径方案中的每个路径方案均覆盖该M个提货点,该N为正整数,且N≤M;
货物分配单元180113,用于为该至少一个历史路径方案中的每个路径方案中的每条运输路径进行该历史运输货物的分配,以得到该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案。
可选地,在一些可能的实施方式中,该路径规划单元180112,具体用于:
若历史路径数据的数量大于第一阈值,则基于该历史路径数据对该M个提货点的转移超参数进行初始化,以得到超参数矩阵;
通过该超参数矩阵确定该M个提货点的转移概率分布,该转移概率分布包括运输路径中的集装箱在该货运起点与该M个提货点之间、该货运终点与该M个提货点之间或该M个提货点之间的转移概率;
根据该转移概率分布确定该至少一个历史路径方案中的每个路径方案中的每条运输路径,以得到该至少一个历史路径方案。
可选地,在一些可能的实施方式中,该训练装置还包括:
初始化模块1804,用于若该历史路径数据的数量不大于该第一阈值,则基于启发式算法对该M个提货点的转移超参数进行初始化,以得到该超参数矩阵。
可选地,在一些可能的实施方式中,该货物分配单元180113,具体用于:
对该M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,该聚类条件包括货物的长度、宽度、高度以及重量;
通过该M个提货点中每个提货点的第一货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第一货物分配方式集合,该M个提货点中每个提货点的第一货物分配超参数为对该M个提货点中每个提货点进行货物分配的超参数,该M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式;
从该M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案。
可选地,在一些可能的实施方式中,该评价子模块18013,具体用于:
通过预置的评价函数以及该实装率,对所获取到的所有货物分配方案进行得分计算;
若该所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从该得分高于该第二阈值的货物分配方案中确定该目标货物分配方案,以及将该目标货物分配方案对应的路径方案作为该目标路径方案;
通过该目标货物分配方案与该目标路径方案确定该目标运输方案。
可选地,在一些可能的实施方式中,该评价函数包括:
Figure PCTCN2018108534-appb-000050
其中,该
Figure PCTCN2018108534-appb-000051
为路径方案向量,m为集装箱的数量,
Figure PCTCN2018108534-appb-000052
为m个集装箱的体积实装率向量,
Figure PCTCN2018108534-appb-000053
为该m个集装箱的载重实装率向量;该α、该β与该γ为权重参数,该r Vi为第i个集装箱的体积实装率,该r Wi为第i个集装箱的载重实装率,该
Figure PCTCN2018108534-appb-000054
为该m个集装箱的平均体积实装率,该
Figure PCTCN2018108534-appb-000055
为该m个集装箱的平均载重实装率。
可选地,在一些可能的实施方式中,该评价子模块18013,还用于:
若该所有货物分配方案中不包括得分高于该第二阈值的货物分配方案,则通过该M个提货点中每个提货点的第二货物分配超参数对该聚类结果进行采样计算,以得到该M个提货点中每个提货点的第二货物分配方式集合,该M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,该M个提货点中每个提货点的第二货物分配超参数为通过该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案对该M个提货点中每个提货点的第一货物分配超参数进行更新得到;
从该M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到该至少一个历史路径方案中的每个路径方案对应的第二历史货物分配方案集合中的每个货物分配方案,该至少一个历史路径方案中的每个路径方案对应的第二历史货物分配方案集合中的每个货物分配方案为针对对应的路径方案对该历史运输货物进行分配的方案;
通过该评价函数以及该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案的实装率,对该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案进行得分计算,该至少一个历史路径方案中的每个路径方案的第二历史货物分配方案集合中的每个货物分配方案的实装率由该三维装载子模型得到。
应理解,对货物分配方案进行重复分配时,还可以重新对路径方案进行规划,也可以直接通过该至少一个历史路径方案重新进行货物分配。
可选地,在一些可能的实施方式中,该训练装置还包括:
确定模块1805,在该通过该实装率对该至少一个历史路径方案中的每个路径方案以及该至少一个历史路径方案中的每个路径方案对应的第一历史货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案之前,用于若该实装率确定该M个提货点中的L个提货点还包括未分配到该集装箱的剩余货物,则为该剩余货物确定剩余货物路径方案和剩余货物分配方案,该L≤该M,该L为正整数;
该评价子模块18013,还用于通过该实装率对该至少一个历史路径方案中的每个路径方案与对应的第一历史货物分配方案集合中的每个货物分配方案,以及该剩余货物路径方案和该剩余货物分配方案进行整合评价,以确定目标运输方案。
图19是本申请实施例提供的一种确定装置结构示意图,该确定装置1900可因配置或性能不同而产生比较大的差异,可以包括一个或一个以上中央处理器(central processing units,CPU)1922(例如,一个或一个以上处理器)和存储器1932,一个或一个以上存储应用程序1942或数据1944的存储介质1930(例如一个或一个以上海量存储设备)。其中,存储器1932和存储介质1930可以是短暂存储或持久存储。存储在存储介质1930的程序可以包括一个或一个以上模块(图示没标出),每个模块可以包括对确定装置中的一系列指令操作。更进一步地,中央处理器1922可以设置为与存储介质1930通信,在确定装置1900上执行存储介质1930中的一系列指令操作。
确定装置1900还可以包括一个或一个以上电源1926,一个或一个以上有线或无线网络接口1950,一个或一个以上输入输出接口1958,和/或,一个或一个以上操作系统1941,例如Windows ServerTM,Mac OS XTM,UnixTM,LinuxTM,FreeBSDTM等等。
上述图2-图16的实施例中确定运输方案的方法中的步骤可以基于该图19所示的确定装置结构执行。
图20是本申请实施例提供的一种训练装置结构示意图,该训练装置2000可因配置或性能不同而产生比较大的差异,可以包括一个或一个以上中央处理器(central processing units,CPU)2022(例如,一个或一个以上处理器)和存储器2032,一个或一个以上存储应用程序2042或数据2044的存储介质2030(例如一个或一个以上海量存储设备)。其中,存储器2032和存储介质2030可以是短暂存储或持久存储。存储在存储介质2030的程序可以包括一个或一个以上模块(图示没标出),每个模块可以包括对训练装置中的一系列指令操作。更进一步地,中央处理器2022可以设置为与存储介质2030通信,在训练装置2000上执行存储介质2030中的一系列指令操作。
训练装置2000还可以包括一个或一个以上电源2026,一个或一个以上有线或无线网络接口2050,一个或一个以上输入输出接口2058,和/或,一个或一个以上操作系统2041,例如Windows ServerTM,Mac OS XTM,UnixTM,LinuxTM,FreeBSDTM等等。
上述图2-图16的实施例中由进行离线训练的步骤可以基于该图20所示的训练装置结构执行。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的系统,装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统,装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本申请各个实施例图2至图16所述方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(ROM,Read-Only Memory)、随机存取存储器(RAM,Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,以上实施例仅用以说明本申请的技术方案,而非对其限制;尽管参照前述实施例对本申请进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (33)

  1. 一种确定运输方案的方法,其特征在于,包括:
    获取至少一个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合,所述至少一个路径方案中的每个路径方案为针对待运输货物进行运输而规划的运输路径,所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合包括至少一个货物分配方案,所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案为针对对应的路径方案对所述待运输货物进行分配的方案;
    根据快速装载模型确定所述至少一个路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,所述快速装载模型为通过离线仿真数据进行离线训练得到,所述离线仿真数据包括通过三维装载算法计算得到的历史装载方案,所述实装率为某一货物分配方案中装入集装箱的货物占用所述集装箱的比例;
    通过所述实装率对所述至少一个路径方案中的每个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,所述目标运输方案包括目标路径方案与所述目标路径方案对应的目标货物分配方案。
  2. 根据权利要求1所述的方法,其特征在于,所述获取至少一个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合,包括:
    获取目标货运单,所述目标货运单包括运输节点信息以及待运输货物信息,所述运输节点信息包括货运起点、货运终点以及M个提货点,所述待运输货物信息包括分布在所述M个提货点的所述待运输货物的信息,所述M为正整数;
    根据所述运输节点信息确定所述至少一个路径方案,所述至少一个路径方案中的每个路径方案包括至少一条运输路径,所述至少一条运输路径中的每条运输路径包括货运起点、货运终点以及所述M个提货点中N个提货点,所述至少一个路径方案中的每个路径方案均覆盖所述M个提货点,所述N为正整数,且N≤M;
    为所述至少一个路径方案中的每个路径方案中的每条运输路径进行所述待运输货物的分配,以得到所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
  3. 根据权利要求2所述的方法,其特征在于,所述根据所述运输节点信息确定所述至少一个路径方案,包括:
    若历史路径数据的数量大于第一阈值,则基于所述历史路径数据对所述M个提货点的转移超参数进行初始化,以得到超参数矩阵;
    通过所述超参数矩阵确定所述M个提货点的转移概率分布,所述转移概率分布包括运输路径中的集装箱在所述货运起点与所述M个提货点之间、所述货运终点与所述M个提货点之间或所述M个提货点之间的转移概率;
    根据所述转移概率分布确定所述至少一个路径方案中的每个路径方案中的每条运输路径,以得到所述至少一个路径方案。
  4. 根据权利要求3所述的方法,其特征在于,所述方法还包括:
    若所述历史路径数据的数量不大于所述第一阈值,则基于启发式算法对所述M个提货点的转移超参数进行初始化,以得到所述超参数矩阵。
  5. 根据权利要求2所述的方法,其特征在于,所述为所述至少一个路径方案中的每个路径方案中的每条运输路径进行所述待运输货物的分配,以得到所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案,包括:
    对所述M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,所述聚类条件包括货物的长度、宽度、高度以及重量;
    通过所述M个提货点中每个提货点的第一货物分配超参数对所述聚类结果进行采样计算,以得到所述M个提货点中每个提货点的第一货物分配方式集合,所述M个提货点中每个提货点的第一货物分配超参数为对所述M个提货点中每个提货点进行货物分配的超参数,所述M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式;
    从所述M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行结合,以得到所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
  6. 根据权利要求2所述的方法,其特征在于,所述根据快速装载模型确定所述至少一个路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,包括:
    获取所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,所述第一特征向量用于指示某一货物分配方案中待运输货物的特征值;
    将所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第一特征向量输入所述快速装载模型中,以得到所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中每个货物分配方案的实装率,所述实装率包括体积实装率与载重实装率,所述体积实装率包括所述至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的体积占用集装箱荷载体积的比例,所述载重实装率所述至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的重量占用集装箱的荷重的比例。
  7. 根据权利要求6所述的方法,其特征在于,所述获取所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,包括:
    获取所述待运输货物中的每个货物的第二特征向量,所述待运输货物中的每个货物的第二特征向量包括对应货物的长度、宽度、高度以及重量;
    根据所述待运输货物中每个货物的第二特征向量计算出所述M个提货点中的每个提货点分布的货物针对所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量,所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量包括所述待运输货物中的每个货物的 第二特征向量的均值与协方差;
    对所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量进行加权组合得到对应的所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量。
  8. 根据权利要求1-7中任一项所述的方法,其特征在于,所述通过所述实装率对所述至少一个路径方案中的每个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,包括:
    通过预置的评价函数以及所述实装率,对所获取到的所有货物分配方案进行得分计算;
    若所述所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从所述得分高于所述第二阈值的货物分配方案中确定所述目标货物分配方案,以及将所述目标货物分配方案对应的路径方案作为所述目标路径方案;
    通过所述目标货物分配方案与所述目标路径方案确定所述目标运输方案。
  9. 根据权利要求8所述的方法,其特征在于,所述评价函数包括:
    Figure PCTCN2018108534-appb-100001
    其中,所述
    Figure PCTCN2018108534-appb-100002
    为路径方案向量,m为集装箱的数量,
    Figure PCTCN2018108534-appb-100003
    为m个集装箱的体积实装率向量,
    Figure PCTCN2018108534-appb-100004
    为所述m个集装箱的载重实装率向量;所述α、所述β与所述γ为权重参数,所述r Vi为第i个集装箱的体积实装率,所述r Wi为第i个集装箱的载重实装率,所述
    Figure PCTCN2018108534-appb-100005
    为所述m个集装箱的平均体积实装率,所述
    Figure PCTCN2018108534-appb-100006
    为所述m个集装箱的平均载重实装率。
  10. 根据权利要求8或9所述的方法,其特征在于,所述方法还包括:
    若所述所有货物分配方案中不包括得分高于所述第二阈值的货物分配方案,则通过所述M个提货点中每个提货点的第二货物分配超参数对所述聚类结果进行采样计算,以得到所述M个提货点中每个提货点的第二货物分配方式集合,所述M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,所述M个提货点中每个提货点的第二货物分配超参数为通过所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案对所述M个提货点中每个提货点的第一货物分配超参数进行更新得到;
    从所述M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到所述至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案,所述至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案为针对对应的路径方案对所述待运输货物进行分配的方案;
    通过所述评价函数以及所述至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率,对所述至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案进行得分计算,所述至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率由所述快速装载模型得 到。
  11. 根据权利要求1至10任一项所述的方法,其特征在于,所述通过所述实装率对所述至少一个路径方案中的每个路径方案与所述至少一个路径方案中的每个路径方案的货物分配方案进行整合评价,以确定目标运输方案之后,所述方法还包括:
    根据所述目标货物分配方案和所述目标路径方案,确定所述目标路径方案中每条运输路径的集装箱的型号;
    根据所述目标路径方案中每条运输路径的集装箱的型号以及三维装载算法生成装载方案,所述装载方案为所述待运输货物在所述目标路径方案中每条运输路径中的集装箱内的装载方式。
  12. 根据权利要求1至11中任一项所述的方法,其特征在于,在所述通过所述实装率对所述至少一个路径方案中的每个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案之前,所述方法还包括:
    若所述实装率确定所述M个提货点中的L个提货点还包括未分配到所述集装箱的剩余货物,则为所述剩余货物确定剩余货物路径方案和剩余货物分配方案,所述L≤所述M,所述L为正整数;
    所述通过所述实装率对所述至少一个路径方案中的每个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,包括:
    通过所述实装率对所述至少一个路径方案中的每个路径方案与对应的第一货物分配方案集合中的每个货物分配方案,以及所述剩余货物路径方案和所述剩余货物分配方案进行整合评价,以确定目标运输方案。
  13. 一种训练快速装载模型的方法,其特征在于,包括:
    获取离线仿真数据,所述离线仿真数据包括通过三维装载计算得到的历史装载方案与历史实装率;
    从所述离线仿真数据中获取特征向量,所述特征向量包括所述历史装载方案对应的历史运输货物的特征值;
    将所述特征向量装换为预置格式的训练数据;
    通过所述训练数据训练预测模型,以得到快速装载模型,所述快速装载模型用于输出每个运输路径的货物分配方案集合中每个货物分配方案的实装率,所述实装率为所述每个货物分配方案中装入集装箱的货物占用所述集装箱的比例。
  14. 根据权利要求13所述的方法,其特征在于,所述预置格式为:特征向量,历史实装率。
  15. 根据权利要求13或14所述的方法,其特征在于,所述预测模型包括:线性回归模型、岭回归模型、LASSO模型、支持向量机模型、随机森林模型、XgBoost模型或人工神经网络模型。
  16. 一种确定装置,其特征在于,包括:
    获取模块,用于获取至少一个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合,所述至少一个路径方案中的每个路径方案为针对待运输货物进行运输而规划的运输路径,所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合包括至少一个货物分配方案,所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案为针对对应的路径方案对所述待运输货物进行分配的方案;
    快速装载模块,用于根据快速装载模型确定所述至少一个路径方案中的每个路径方案对应的第一货物分配集合中的每个货物分配方案的实装率,所述快速装载模型为通过离线仿真数据进行离线训练得到,所述离线仿真数据包括通过三维装载算法计算得到的历史装载方案,所述实装率为某一货物分配方案中装入集装箱的货物占用所述集装箱的比例;
    评价模块,通过所述实装率对所述至少一个路径方案中的每个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案,其中,所述目标运输方案包括目标路径方案与所述目标路径方案对应的目标货物分配方案。
  17. 根据权利要求16所述的确定装置,其特征在于,所述获取模块,包括:
    获取子模块,用于获取目标货运单,所述目标货运单包括运输节点信息以及待运输货物信息,所述运输节点信息包括货运起点、货运终点以及M个提货点,所述待运输货物信息包括分布在所述M个提货点的所述待运输货物的信息,所述M为正整数;
    路径规划子模块,用于根据所述运输节点信息确定所述至少一个路径方案,所述至少一个路径方案中的每个路径方案包括至少一条运输路径,所述至少一条运输路径中的每条运输路径包括货运起点、货运终点以及所述M个提货点中N个提货点,所述至少一个路径方案中的每个路径方案均覆盖所述M个提货点,所述N为正整数,且N≤M;
    货物分配子模块,用于为所述至少一个路径方案中的每个路径方案中的每条运输路径进行所述待运输货物的分配,以得到所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
  18. 根据权利要求17所述的确定装置,其特征在于,所述路径规划子模块,具体用于:
    若历史路径数据的数量大于第一阈值,则基于所述历史路径数据对所述M个提货点的转移超参数进行初始化,以得到超参数矩阵;
    通过所述超参数矩阵确定所述M个提货点的转移概率分布,所述转移概率分布包括运输路径中的集装箱在所述货运起点与所述M个提货点之间、所述货运终点与所述M个提货点之间或所述M个提货点之间的转移概率;
    根据所述转移概率分布确定所述至少一个路径方案中的每个路径方案中的每条运输路径,以得到所述至少一个路径方案。
  19. 根据权利要求18所述的确定装置,其特征在于,所述确定装置还包括:
    初始化模块,用于若所述历史路径数据的数量不大于所述第一阈值,则基于启发式算法对所述M个提货点的转移超参数进行初始化,以得到所述超参数矩阵。
  20. 根据权利要求17所述的确定装置,其特征在于,所述货物分配子模块,具体用于:
    对所述M个提货点中每个提货点的货物根据聚类条件进行聚类,以得到聚类结果,所述聚类条件包括货物的长度、宽度、高度以及重量;
    通过所述M个提货点中每个提货点的第一货物分配超参数对所述聚类结果进行采样计算,以得到所述M个提货点中每个提货点的第一货物分配方式集合,所述M个提货点中每个提货点的第一货物分配超参数为对所述M个提货点中每个提货点进行货物分配的超参数,所述M个提货点中每个提货点的第一货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式;
    从所述M个提货点中每个提货点的第一货物分配方式集合中分别选取货物分配方式进行结合,以得到所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案。
  21. 根据权利要求17所述的确定装置,其特征在于,所述快速装载模块,具体用于:
    获取所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量,所述第一特征向量用于指示某一货物分配方案中待运输货物的特征值;
    将所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第一特征向量输入所述快速装载模型中,以得到所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中每个货物分配方案的实装率,所述实装率包括体积实装率与载重实装率,所述体积实装率包括所述至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的体积占用集装箱荷载体积的比例,所述载重实装率所述至少一个路径方案中的每个路径方案中的每条运输路径所分配的货物的重量占用集装箱的荷重的比例。
  22. 根据权利要求21所述的确定装置,其特征在于,所述快速装载模块,具体用于:
    获取所述待运输货物中的每个货物的第二特征向量,所述待运输货物中的每个货物的第二特征向量包括对应货物的长度、宽度、高度以及重量;
    根据所述待运输货物中每个货物的第二特征向量计算出所述M个提货点中的每个提货点分布的货物针对所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量,所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量包括所述待运输货物中的每个货物的第二特征向量的均值与协方差;
    对所述至少一个路径方案中的每个路径方案对应的第一货物分配方案中的每个货物分配方案的第三特征向量进行加权组合得到对应的所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案的第一特征向量。
  23. 根据权利要求16-22中任一项所述的确定装置,其特征在于,所述评价模块,具体用于:
    通过预置的评价函数以及所述实装率,对所获取到的所有货物分配方案进行得分计算;
    若所述所有货物分配方案中包括得分高于第二阈值的货物分配方案,则从所述得分高于所述第二阈值的货物分配方案中确定所述目标货物分配方案,以及将所述目标货物分配 方案对应的路径方案作为所述目标路径方案;
    通过所述目标货物分配方案与所述目标路径方案确定所述目标运输方案。
  24. 根据权利要求23所述的确定装置,其特征在于,所述评价函数包括:
    Figure PCTCN2018108534-appb-100007
    其中,所述
    Figure PCTCN2018108534-appb-100008
    为路径方案向量,m为集装箱的数量,
    Figure PCTCN2018108534-appb-100009
    为m个集装箱的体积实装率向量,
    Figure PCTCN2018108534-appb-100010
    为所述m个集装箱的载重实装率向量;所述α、所述β与所述γ为权重参数,所述r Vi为第i个集装箱的体积实装率,所述r Wi为第i个集装箱的载重实装率,所述
    Figure PCTCN2018108534-appb-100011
    为所述m个集装箱的平均体积实装率,所述
    Figure PCTCN2018108534-appb-100012
    为所述m个集装箱的平均载重实装率。
  25. 根据权利要求23或24所述的确定装置,其特征在于,所述评价模块,还用于:
    若所述所有货物分配方案中不包括得分高于所述第二阈值的货物分配方案,则通过所述M个提货点中每个提货点的第二货物分配超参数对所述聚类结果进行采样计算,以得到所述M个提货点中每个提货点的第二货物分配方式集合,所述M个提货点中每个提货点的第二货物分配方式集合中的每个货物分配方式为针对对应的路径方案对分布在提货点的货物进行分配的方式,所述M个提货点中每个提货点的第二货物分配超参数为通过所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案对所述M个提货点中每个提货点的第一货物分配超参数进行更新得到;
    从所述M个提货点中每个提货点的第二货物分配方式集合中分别选取货物分配方式进行结合,以得到所述至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案,所述至少一个路径方案中的每个路径方案对应的第二货物分配方案集合中的每个货物分配方案为针对对应的路径方案对所述待运输货物进行分配的方案;
    通过所述评价函数以及所述至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率,对所述至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案进行得分计算,所述至少一个路径方案中的每个路径方案的第二货物分配方案集合中的每个货物分配方案的实装率由所述快速装载模型得到。
  26. 根据权利要求16至25任一项所述的确定装置,其特征在于,所述确定装置还包括:
    后期处理模块,用于通过所述实装率对所述至少一个路径方案中的每个路径方案与所述至少一个路径方案中的每个路径方案的货物分配方案进行整合评价,以确定目标运输方案之后,根据所述目标货物分配方案和所述目标路径方案,确定所述目标路径方案中每条运输路径的集装箱的型号;
    三维装载模块,用于根据所述后期处理模块确定的所述目标路径方案中每条运输路径的集装箱的型号以及三维装载算法生成装载方案,所述装载方案为所述待运输货物在所述 目标路径方案中每条运输路径中的集装箱内的装载方式。
  27. 根据权利要求16至26中任一项所述的确定装置,其特征在于,所述确定装置还包括:
    确定模块,在所述通过所述实装率对所述至少一个路径方案中的每个路径方案以及所述至少一个路径方案中的每个路径方案对应的第一货物分配方案集合中的每个货物分配方案进行整合评价,以确定目标运输方案之前,用于若所述实装率确定所述M个提货点中的L个提货点还包括未分配到所述集装箱的剩余货物,则为所述剩余货物确定剩余货物路径方案和剩余货物分配方案,所述L≤所述M,所述L为正整数;
    所述评价模块,还用于通过所述实装率对所述至少一个路径方案中的每个路径方案与对应的第一货物分配方案集合中的每个货物分配方案,以及所述剩余货物路径方案和所述剩余货物分配方案进行整合评价,以确定目标运输方案。
  28. 一种训练装置,其特征在于,包括:
    获取模块,用于获取离线仿真数据,所述离线仿真数据包括通过三维装载计算得到的历史装载方案与历史实装率;
    所述获取模块,还用于从所述离线仿真数据中获取特征向量,所述特征向量包括所述历史装载方案对应的历史运输货物的特征值;
    转换模块,用于将所述特征向量转换为预置格式的训练数据;
    训练模块,用于通过所述训练数据训练预测模型,以得到快速装载模型,所述快速装载模型用于输出每个运输路径的货物分配方案集合中每个货物分配方案的实装率,所述实装率为所述每个货物分配方案中装入集装箱的货物占用所述集装箱的比例。
  29. 根据权利要求28所述的训练装置,其特征在于,所述预置格式为:(特征向量,历史实装率)。
  30. 根据权利要求28或29所述的训练装置,其特征在于,所述预测模型包括:线性回归模型、岭回归模型、LASSO模型、支持向量机模型、随机森林模型、XgBoost模型或人工神经网络模型。
  31. 一种计算机可读存储介质,包括指令,当所述指令在计算机上运行时,使得计算机执行如权利要求1-15中任意一项所述的方法。
  32. 一种确定装置,其特征在于,包括:
    存储器,用于存储程序;
    处理器,用于执行所述存储器存储的所述程序,当所述程序被执行时,所述处理器用于执行如权利要求1-12中任一所述的步骤。
  33. 一种训练装置,其特征在于,包括:
    存储器,用于存储程序;
    处理器,用于执行所述存储器存储的所述程序,当所述程序被执行时,所述处理器用于执行如权利要求13-15中任一所述的步骤。
PCT/CN2018/108534 2018-02-06 2018-09-29 确定运输方案的方法、训练快速装载模型的方法及设备 WO2019153759A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP18905663.3A EP3739530A4 (en) 2018-02-06 2018-09-29 METHOD FOR DETERMINING A TRANSPORT SCHEME, AND METHOD AND DEVICE FOR LEARNING A QUICK LOADING MODEL
US16/986,508 US20200364664A1 (en) 2018-02-06 2020-08-06 Method for determining transportation scheme, method for training fast loading model, and device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810118531.9 2018-02-06
CN201810118531.9A CN109902987B (zh) 2018-02-06 2018-02-06 确定运输方案的方法、训练快速装载模型的方法及设备

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/986,508 Continuation US20200364664A1 (en) 2018-02-06 2020-08-06 Method for determining transportation scheme, method for training fast loading model, and device

Publications (1)

Publication Number Publication Date
WO2019153759A1 true WO2019153759A1 (zh) 2019-08-15

Family

ID=66943197

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/108534 WO2019153759A1 (zh) 2018-02-06 2018-09-29 确定运输方案的方法、训练快速装载模型的方法及设备

Country Status (4)

Country Link
US (1) US20200364664A1 (zh)
EP (1) EP3739530A4 (zh)
CN (1) CN109902987B (zh)
WO (1) WO2019153759A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110533234A (zh) * 2019-08-20 2019-12-03 集美大学 结合避碰策略的agv优化控制方法、终端设备及存储介质
CN111382907A (zh) * 2020-03-10 2020-07-07 上海东普信息科技有限公司 规划车线的方法、系统、计算机设备和存储介质
CN113537506A (zh) * 2020-04-22 2021-10-22 百度在线网络技术(北京)有限公司 针对机器学习效果的测试方法、装置、设备和介质
CN114358665A (zh) * 2022-03-21 2022-04-15 天津港第二集装箱码头有限公司 考虑art前置和后置缓冲区的地面锁站动态调度方法

Families Citing this family (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA3084343C (en) * 2017-12-11 2024-03-12 Banyan Technology Customized integrated pricing packages for freight shipment
CN112308280A (zh) * 2019-08-02 2021-02-02 菜鸟智能物流控股有限公司 物流调度管理方法、装置、电子设备和存储介质
CN112395903A (zh) * 2019-08-12 2021-02-23 顺丰科技有限公司 一种空间特征的确定方法、装置、网络设备及存储介质
CN111178810B (zh) * 2019-12-31 2023-06-20 北京百度网讯科技有限公司 用于生成信息的方法和装置
CN111709576A (zh) * 2020-06-15 2020-09-25 深圳市前海沃德太客供应链技术有限公司 一种运输信息处理系统
CN111784265A (zh) * 2020-07-29 2020-10-16 上汽安吉物流股份有限公司 整车物流运输智能配载系统及方法
CN111898954B (zh) * 2020-07-31 2024-01-12 沙师弟(重庆)网络科技有限公司 一种基于改进高斯混合模型聚类的车辆匹配方法
CN112183954B (zh) * 2020-09-10 2024-05-24 华为技术有限公司 一种获取货物的装载方案的方法和装置
CN112365207A (zh) * 2020-11-10 2021-02-12 上海汽车集团股份有限公司 一种装箱方法、装置及计算机可读存储介质
US11783271B2 (en) * 2021-01-04 2023-10-10 Bank Of America Corporation System for directing resource transfers based on resource distribution data
CN112766611A (zh) * 2021-02-08 2021-05-07 日日顺供应链科技股份有限公司 一种三维装车方法及装置
CN112801593A (zh) * 2021-02-09 2021-05-14 北京橙心无限科技发展有限公司 生成配送方案的方法和装置
CN112966997A (zh) * 2021-03-04 2021-06-15 上海东普信息科技有限公司 快递货物的运输方法、装置、设备及存储介质
JP2023013359A (ja) * 2021-07-15 2023-01-26 株式会社日立製作所 物流作業最適化装置および物流作業最適化方法
CN113988720B (zh) * 2021-12-27 2022-06-07 北京京东振世信息技术有限公司 分流路径生成方法、装置、电子设备和计算机可读介质
US20230245045A1 (en) * 2022-01-30 2023-08-03 Walmart Apollo, Llc Systems and methods for vehicle routing
CN115049334B (zh) * 2022-06-16 2023-04-07 四川省交通勘察设计研究院有限公司 一种水路运输需求统计分析方法、设备及存储介质
US20240005258A1 (en) * 2022-06-30 2024-01-04 Uber Technologies, Inc. IIntelligent Load Clusters for Freight
CN115994689B (zh) * 2023-03-22 2023-05-30 交通运输部水运科学研究所 一种智慧码头港区危险货物安全运输的管理方法及装置
CN116402424B (zh) * 2023-04-07 2024-01-23 广州拓威天海国际物流有限公司 一种物流配置优化调节方法及系统
CN117114560B (zh) * 2023-10-25 2024-01-05 中科云谷科技有限公司 用于确定物品运输路径的方法、装置、存储介质及处理器

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103473617A (zh) * 2013-09-17 2013-12-25 四川航天系统工程研究所 多种物资放入多规格货包的三维装载全局优化方法及系统
CN104866911A (zh) * 2014-02-21 2015-08-26 日本电气株式会社 用于优化物流配载的设备和方法
KR20160119633A (ko) * 2015-04-06 2016-10-14 (주)한국화물운송 화물 연계 운송 관리 방법 및 시스템, 컴퓨터 프로그램, 그 기록매체

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6937992B1 (en) * 2000-12-29 2005-08-30 Arrowstream, Inc. Transport vehicle capacity maximization logistics system and method of same
CA2370084C (en) * 2002-02-01 2017-12-12 Canadian National Railway Company System and method for on-line ordering of a transporation service providing route selection capability
US20090048987A1 (en) * 2007-08-13 2009-02-19 Jeff Sweeney System and Method for Optimizing the Loading of Shipping Containers
CN101381033A (zh) * 2008-10-15 2009-03-11 江苏科技大学 一种基于蚁群算法的集装箱装载方法
CN102142107B (zh) * 2010-01-29 2016-03-09 国际商业机器公司 物流信息处理方法和装置
US8554702B2 (en) * 2010-12-17 2013-10-08 Oracle International Corporation Framework for optimized packing of items into a container
US20140180954A1 (en) * 2012-12-20 2014-06-26 Oracle International Corporation Concurrent determination of shipping mode and ship unit packing during transportation planning
CN103455841B (zh) * 2013-07-17 2016-12-28 大连海事大学 基于改进蚁群算法和启发式算法的集装箱装载方法
US20150248639A1 (en) * 2014-02-28 2015-09-03 SynchroNet Logistics, Inc. Determining optimal routes and rates for cargo transportation
CN111626577B (zh) * 2016-09-30 2023-12-26 杭州数梦工场科技有限公司 一种车辆调度方法和装置
CN106600180B (zh) * 2016-11-04 2018-01-02 中联物流(中国)有限公司 物流管理系统的运输路径优化方法和系统

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103473617A (zh) * 2013-09-17 2013-12-25 四川航天系统工程研究所 多种物资放入多规格货包的三维装载全局优化方法及系统
CN104866911A (zh) * 2014-02-21 2015-08-26 日本电气株式会社 用于优化物流配载的设备和方法
KR20160119633A (ko) * 2015-04-06 2016-10-14 (주)한국화물운송 화물 연계 운송 관리 방법 및 시스템, 컴퓨터 프로그램, 그 기록매체

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3739530A4

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110533234A (zh) * 2019-08-20 2019-12-03 集美大学 结合避碰策略的agv优化控制方法、终端设备及存储介质
CN110533234B (zh) * 2019-08-20 2022-09-23 集美大学 结合避碰策略的agv优化控制方法、终端设备及存储介质
CN111382907A (zh) * 2020-03-10 2020-07-07 上海东普信息科技有限公司 规划车线的方法、系统、计算机设备和存储介质
CN113537506A (zh) * 2020-04-22 2021-10-22 百度在线网络技术(北京)有限公司 针对机器学习效果的测试方法、装置、设备和介质
CN113537506B (zh) * 2020-04-22 2023-08-29 百度在线网络技术(北京)有限公司 针对机器学习效果的测试方法、装置、设备和介质
CN114358665A (zh) * 2022-03-21 2022-04-15 天津港第二集装箱码头有限公司 考虑art前置和后置缓冲区的地面锁站动态调度方法
CN114358665B (zh) * 2022-03-21 2022-07-01 天津港第二集装箱码头有限公司 考虑art前置和后置缓冲区的地面锁站动态调度方法

Also Published As

Publication number Publication date
CN109902987B (zh) 2023-12-08
EP3739530A4 (en) 2020-12-23
EP3739530A1 (en) 2020-11-18
CN109902987A (zh) 2019-06-18
US20200364664A1 (en) 2020-11-19

Similar Documents

Publication Publication Date Title
WO2019153759A1 (zh) 确定运输方案的方法、训练快速装载模型的方法及设备
CN111191732B (zh) 一种基于全自动学习的目标检测方法
Jiang et al. Mainstream: Dynamic {Stem-Sharing} for {Multi-Tenant} Video Processing
US9355067B1 (en) Distribution of parameter calculation for iterative optimization methods
CN107203789B (zh) 分配模型建立方法、分配方法以及相关装置
US9563822B2 (en) Learning apparatus, density measuring apparatus, learning method, computer program product, and density measuring system
CN109460793A (zh) 一种节点分类的方法、模型训练的方法及装置
CN110515732A (zh) 一种基于资源受限机器人深度学习推理的任务分配方法
CN111915060A (zh) 组合优化任务的处理方法以及处理装置
CN111523918B (zh) 商品聚类的方法、装置、设备和存储介质
Alzaqebah et al. Whale optimization algorithm for requirements prioritization
WO2020140818A1 (zh) 订单处理方法、装置、设备及存储介质
CN115329683B (zh) 航空行李在线装载规划方法、装置、设备及介质
CN116594748A (zh) 针对任务的模型定制处理方法、装置、设备和介质
CN113391894A (zh) 一种基于rbp神经网络的最优超任务网优化方法
CN113704389A (zh) 一种数据评估方法、装置、计算机设备及存储介质
Gupta et al. Image enhancement using ant colony optimization
Nalepa et al. Adaptive guided ejection search for pickup and delivery with time windows
CN101226521A (zh) 一种用于多义性数据对象预测建模的机器学习方法
CN116703129B (zh) 基于人员数据画像的智能任务匹配调度方法及系统
CN115936184A (zh) 一种适应多用户类型的负荷预测匹配方法
CN113822301B (zh) 分拣中心的分类方法及装置、存储介质、电子设备
CN115112137A (zh) 路径规划方法、装置、电子设备及可读存储介质
Fagerlund Comparison of machine learning algorithms for real-time vehicle selection in transport management
CN117952397B (zh) 物流订单分析方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18905663

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018905663

Country of ref document: EP

Effective date: 20200812