WO2020019901A1 - 通行时间确定方法、装置、计算机设备及存储介质 - Google Patents

通行时间确定方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2020019901A1
WO2020019901A1 PCT/CN2019/091311 CN2019091311W WO2020019901A1 WO 2020019901 A1 WO2020019901 A1 WO 2020019901A1 CN 2019091311 W CN2019091311 W CN 2019091311W WO 2020019901 A1 WO2020019901 A1 WO 2020019901A1
Authority
WO
WIPO (PCT)
Prior art keywords
time
state data
road section
data
transit time
Prior art date
Application number
PCT/CN2019/091311
Other languages
English (en)
French (fr)
Inventor
刘雨亭
Original Assignee
腾讯科技（深圳）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 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2020019901A1 publication Critical patent/WO2020019901A1/zh
Priority to US17/004,442 priority Critical patent/US11846519B2/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/34—Route searching; Route guidance
    • G01C21/3453—Special cost functions, i.e. other than distance or default speed limit of road segments
    • G01C21/3492—Special cost functions, i.e. other than distance or default speed limit of road segments employing speed data or traffic data, e.g. real-time or historical
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/34—Route searching; Route guidance
    • G01C21/3407—Route searching; Route guidance specially adapted for specific applications
    • G01C21/3415—Dynamic re-routing, e.g. recalculating the route when the user deviates from calculated route or after detecting real-time traffic data or accidents
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION 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
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats

Definitions

  • the embodiments of the present application relate to the field of computer technology, and in particular, to a method, a device, a computer device, and a storage medium for determining a transit time.
  • navigation functions are widely used in daily life.
  • the navigation function of the mobile terminal can be used to predict the transit time of the route and thus the time to reach the destination. Point, bring a lot of convenience for users to travel.
  • a point-in-time prediction model is generally used to determine a point in time when a user arrives at a destination.
  • sample data of at least one sample route is obtained.
  • the sample data includes route description data and historical passage data corresponding to the sample route.
  • the route description data is used to describe the geographical situation of the sample route.
  • the historical passage data includes at least the sample.
  • the transit time of the route. Training is performed based on the obtained multiple sample data to obtain a point-in-time prediction model, which can be used to predict the estimated arrival time of any route.
  • the route description data of the target route and the current time point can be input into the point-in-time prediction model, and the estimated arrival time of the target route is determined based on the point-in-time prediction model, that is, the user's arrival The point in time of the destination of the target route.
  • the point-in-time prediction model is trained only based on the global information of the route, and the local information of the route is not taken into account.
  • the time-based prediction model can only determine the estimated arrival time based on the global information of the route, and the prediction is not accurate enough.
  • a method, an apparatus, a computer device, and a storage medium for determining a transit time are provided.
  • a method for determining transit time is provided. The method includes:
  • the computer equipment obtains a target route to be passed, and the target route includes a plurality of road segments arranged in sequence;
  • the computer equipment starts from the first of a plurality of road sections, and for the first and second road sections adjacent to each other at any two positions, according to the first state data of the first road section, based on the transit time selection model and state data prediction
  • the model determines the transit time of the first road section and the second status data of the second road section after passing the first road section according to the travel time under the first state data;
  • the computer equipment continues to determine the transit time of the second road segment based on the transit time selection model and the status data prediction model, until the transit time of each of the multiple road segments is determined;
  • the transit time selection model is used to determine the transit time of any link based on the status data of any link
  • the status data prediction model is used to determine the status of the next link of any link based on the status data and transit time of any link. data.
  • a transit time determining device in another aspect, includes:
  • a route acquisition module configured to obtain a target route to be passed, the target route includes a plurality of road segments arranged in sequence;
  • a strategy determination module which is used to select a model based on the transit time of the first road segment and the second road segment adjacent to each other at any two positions, starting from the first road segment of the plurality of road segments.
  • state data prediction model determine the transit time of the first road section, and the second state data of the second road section after passing the first road section according to the travel time under the first state data;
  • the strategy determination module is further configured to continue to determine the transit time of the second road segment based on the transit time selection model and the status data prediction model based on the second status data, until the transit time of each of the multiple road segments is determined;
  • the transit time selection model is used to determine the transit time of any link based on the status data of any link
  • the status data prediction model is used to determine the status of the next link of any link based on the status data and transit time of any link. data.
  • a computer device includes a memory and a processor.
  • the memory stores computer-readable instructions.
  • the processor causes the processor to perform the following steps:
  • the target route includes a plurality of road segments arranged in order;
  • the second state data continue to determine the transit time of the second road segment based on the transit time selection model and the status data prediction model until the transit time of each of the multiple road segments is determined;
  • the transit time selection model is used to determine the transit time of any link based on the status data of any link
  • the status data prediction model is used to determine the status of the next link of any link based on the status data and transit time of any link. data.
  • a computer-readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, the processor causes the processor to perform the following steps:
  • the target route includes a plurality of road segments arranged in order;
  • the second state data continue to determine the transit time of the second road segment based on the transit time selection model and the status data prediction model until the transit time of each of the multiple road segments is determined;
  • the transit time selection model is used to determine the transit time of any link based on the status data of any link
  • the status data prediction model is used to determine the status of the next link of any link based on the status data and transit time of any link. data.
  • FIG. 1A is an application environment diagram of a method for determining transit time provided by an embodiment of the present application
  • FIG. 1 is a flowchart of a model training method according to an embodiment of the present application
  • FIG. 2 is a schematic diagram of real-time traffic speed prediction provided by an embodiment of the present application.
  • FIG. 3 is a schematic diagram of real-time traffic speed prediction of a remaining road section provided by an embodiment of the present application.
  • FIG. 4 is a schematic diagram of historical traffic data provided by an embodiment of the present application.
  • FIG. 6 is a flowchart of a method for determining transit time provided by an embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of a transit time determining device according to an embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of a terminal according to an embodiment of the present application.
  • FIG. 9 is a schematic structural diagram of a server according to an embodiment of the present application.
  • FIG. 10 is a schematic structural diagram of a computer device according to an embodiment of the present application.
  • Reinforcement learning framework Including Agent (agent), state, action, incentive, value, and Markov decision-making process and other members.
  • a prediction model is trained during the training phase.
  • the prediction model is essentially a mapping relationship between route characteristics and ETA. Training the sample data can make the prediction model match the sample data as much as possible, and the accuracy is higher.
  • the ETA corresponding to the route characteristics can be obtained according to the route characteristics and the mapping relationship determined in the prediction model.
  • the reinforcement learning framework is different from the general supervised learning framework.
  • the agent repeatedly tries to make different actions in different states through the feedback mechanism, and according to the benefits of each action made, step by step Optimize the feedback mechanism, and finally find the decision sequence that can get the maximum benefit in the Markov Decision Process (MDP). Therefore, the training result of the reinforcement learning framework is not a numerical value, but a distribution of benefits in the state-action space.
  • the input is a target route including a plurality of road sections
  • the output is a decision optimal strategy, which includes the travel time of each road section.
  • Section The smallest unit used to describe a route.
  • a route consists of multiple sections. Each section is described by a set of structured physical description data, including but not limited to the length, width, and number of traffic lights. , Road grade, etc.
  • Status data of the road section it can include at least one of the following:
  • Initial speed refers to the real-time traffic speed of the first section at the beginning of the target route, and the initial speeds corresponding to different sections of the same route are equal.
  • Historical statistical speed Perform statistics based on the historical traffic data of the road section to obtain the statistical value of the road speed at a certain point in time. For the same road section, the historical statistical speed at different time points may be different.
  • Real-time data of road sections including real-time traffic speed, statistical data and physical description data;
  • the real-time speed of a road section refers to the real-time speed of the road section at the beginning of the road section, and the real-time speed of the same road section at different time points may be different.
  • the statistical data of a road segment refers to the statistics of the historical traffic conditions of the road segment, including but not limited to the speed of the road segment when the traffic is clear and multiple historical statistical speeds within a certain time interval.
  • the physical description data of the road segment is used to describe the geographical situation of the road segment, including the length, width, number of traffic lights contained in the road segment, and road grade.
  • Real-time data of the remaining road sections including the real-time traffic speed, statistical data, and physical description data of each remaining road section after the road section.
  • the specific data format is similar to the real-time data of the road section, and is not repeated here.
  • Action refers to the action of passing a certain section according to a certain transit time, and the action of each section is expressed by the transit time.
  • the status of the Agent will change.
  • the changes include: the historical statistical speed will be switched to the historical statistical speed of the next section, and the cumulative transit time will increase.
  • the real-time data of the newly passed link and the real-time data of the link will be switched to the real-time data of the next link, and the real-time data of the remaining link will be removed from the real-time data of the newly passed link.
  • Final incentive value After passing a route to the destination of the route, the feedback for the transit time of the entire route is represented by R finish .
  • State data revenue value refers to the expectation of future revenue when the state is in a certain state at a certain point in time. This revenue value can be used to measure the accuracy of the state data prediction. The more the state of the situation is consistent with the actual situation, the smaller the error brought by it, and the more accurate the strategy formulated.
  • Value of transit time refers to the expectation of future revenue under the condition of passing a current section according to a certain transit time at a certain state at a certain point in time. The higher the value of the revenue, the current time point is in the current state. In the case of passing the current road section according to the transit time, the greater the expected gain, that is, the more the situation is consistent with the actual situation, the smaller the error brought by it, and the more accurate the formulated strategy is.
  • Markov decision process MDP ⁇ S, A, P, R, ⁇ >, where S represents a set of state data, A represents a set of transit time, P represents a state transition probability matrix, and each of the state transition probability matrices Each element represents the probability of transitioning from the previous set of state data to the next set of state data, R is the incentive, and ⁇ is the discount factor, which is used to calculate the cumulative return value.
  • a Markov decision process can be as follows:
  • the Agent When the Agent is in a certain state s in the S set, it may perform n actions a in the A set. For each different action a, the Agent simulates that after performing action a, it will have an effect on state s and reach a new state s'. In this process, the Agent will receive the immediate incentive corresponding to action a, and calculate the revenue value of the new state s', and finally select the action with the largest sum of the instant incentive and revenue values among the n actions to execute.
  • Travel time selection model A model for determining the travel time of the road section according to the status data of the road section.
  • State data prediction model A model that predicts the state data of the next road segment based on the state data and travel time of the previous road segment.
  • the state data prediction model may include at least one of a first speed prediction model and a second speed prediction model.
  • the first speed prediction model is used to predict the real-time traffic of the next link when the next link is reached based on the real-time passing speed and travel time of the link.
  • Speed the second speed prediction model is used to predict the real-time passing speed of the remaining links after the next link when the next link is reached based on the real-time passing speed of the current link and the remaining links after the current link.
  • Return value prediction model A model that obtains return values based on state data. The return value is used to represent the expected return in the current state.
  • a machine learning scheme based on supervised learning, which can be used to determine the transit time of a route.
  • a point-in-time prediction model is trained based on sample data of multiple sample routes. For a target route to be passed by a user, the estimated arrival time of the target route can be determined based on the point-in-time prediction model.
  • the sample route can be described from a global perspective, but the sample route cannot be sampled from a local perspective. Considering the local information of the route, this will cause the point-in-time prediction model to only determine the estimated arrival time based on the global information of the route, but the local information of the route is lost, so the prediction is not accurate enough.
  • an embodiment of the present application proposes a solution for determining travel time.
  • a travel time selection model and a state data prediction model in units of road sections are trained.
  • the target route can be
  • the travel time selection model and status data prediction model are used to determine the travel time of each link.
  • the local information of each link in the target route is fully considered, and the travel time of each link is predicted separately, which improves the forecast. Accuracy.
  • the embodiments of the present application are applied to scenarios in which the travel time of a target route is predicted, for example, in a scenario of map navigation, when a user wants to set off to a destination, multiple routes may be determined according to the current location of the user and the destination
  • the method provided in the embodiment of the present application is used to predict the transit time of each section of each route, thereby predicting the total transit time of each route, and the user selects a route with a shorter total transit time.
  • the embodiments of the present application can also be applied to other scenarios in which the travel time of a target route needs to be predicted.
  • the terminal may install a map navigation application, and the method provided in the embodiment of the present application may be used in the map navigation application to predict the travel time of the target route.
  • FIG. 1A is an application environment diagram of a method for determining transit time in an embodiment.
  • the transit time determination method is applied to a transit time determination system.
  • the transit time determination system includes a terminal 110 and a server 120.
  • the terminal 110 and the server 120 are connected through a network.
  • the terminal 110 may be a desktop terminal or a mobile terminal, and the mobile terminal may be at least one of a mobile phone, a tablet computer, and a notebook computer.
  • the server 120 may be implemented by an independent server or a server cluster composed of multiple servers.
  • the terminal 110 sends a target route to be passed to the server 120, where the target route includes a plurality of road segments arranged in order.
  • the server 120 obtains the target route to be passed, starting from the first one of the multiple road sections, and for the first and second road sections adjacent to each other at any two positions, based on the first state data of the first road section, based on
  • the transit time selection model and status data prediction model determine the transit time of the first road segment, and the second status data of the second road segment after passing the first road segment according to the transit time under the first status data, and based on the second status data, continue to be based on
  • the transit time selection model and status data prediction model determine the transit time of the second road section until the transit time of each of the multiple road sections is determined.
  • the transit time selection model is used to determine the transit time of any link based on the status data of any link
  • the status data prediction model is used to determine the status of the next link of any link based on the status data and transit time of any link. data.
  • the server 120 may return the transit time of each of the plurality of road segments to the terminal 110.
  • FIG. 1 is a flowchart of a model training method according to an embodiment of the present application.
  • the execution body of the model training method is a training device, and the process of training a model is described.
  • the training device may be a terminal or server such as a mobile phone or computer with a navigation function. Referring to Figure 1, the method includes:
  • Each sample route includes a plurality of road segments arranged in order, and the historical traffic data of each road segment in each sample route can be collected according to the movement process of the sample equipment.
  • the sample device may include various types of devices such as a mobile phone, an on-board computer, and a tablet computer.
  • an electronic map can be obtained.
  • the electronic map includes multiple sections.
  • the sample device can be located, the location of the sample device can be determined in real time, and according to the electronic map and the sample device, Position and the corresponding time point, collect the historical traffic data of each sample road section of the sample equipment, so as to obtain the historical traffic data of each road section in a sample route.
  • This collection method can be used to collect historical traffic data for each sample segment of multiple sample routes for multiple sample devices.
  • the historical traffic data of the road section includes the passage time and status data of the road section.
  • the transit time is the time it takes for the sample equipment to pass the road section.
  • the status data may include at least one of the initial speed, historical statistical speed, cumulative travel time, real-time data of the road section, and real-time data of the remaining road sections. Includes other data that can indicate the current status of the sample device.
  • V 0 represents the initial speed, that is, the real-time traffic speed at the beginning of the route
  • V hts, i represents the historical statistical speed of road segment i at the time point from the starting point of the route when it is at the starting point of road segment i;
  • T sum, i represents the current starting point of the route, when it is at the starting point of link i, the total transit time accumulated by the link before link i is the sum of the transit time of each link before link i;
  • L i represents the real-time data of link i, including real-time traffic speed, statistical data, and physical description data.
  • the real-time traffic speed refers to the real-time traffic speed of the road segment i starting from the starting point of the route and at the time point of the road segment i.
  • the statistical data includes the traffic speed of the road segment i when it is clear and within at least a period of time. Multiple historical statistical speeds and physical description data are used to describe the geographical situation of road segment i, which may include the length, width, number of traffic lights contained in the road segment i, and road grade.
  • L left, i represents the real-time data of each road segment after the road segment i, including the real-time speed, statistical data and physical description data of each subsequent road segment.
  • the historical traffic data of each road segment directly collected by the training device includes the time point and speed of the sample equipment passing through a certain place. For each road segment in the electronic map, according to the time when the sample equipment passes the starting point of the road segment The time point and the time point when the end point of the link is passed can determine the passage time of the link. In addition, according to the historical traffic data of the road segment and the historical traffic data of each road segment after the road segment, various status data at the beginning of the road segment can be obtained, that is, the status data of the road segment.
  • Each sample route constructs a plurality of sets of first sample data according to the historical traffic data of each link in the sample route.
  • Each set of the first sample data includes a set of status data and a set of status data corresponding to the set of status data.
  • the travel time of the road segment; training is performed based on multiple sets of first sample data to obtain a travel time selection model.
  • the transit time selection model is used to determine the transit time of any link according to the status data of any link. For any road segment, based on the transit time selection model, the travel time of the segment can be predicted based on the status data of the segment.
  • the status data and travel time of the road section can be obtained, that is, the corresponding relationship between the status data and the travel time is obtained, and the corresponding relationship is used as a set of first One sample of data to obtain multiple sets of first sample data.
  • i represents the previous link
  • i + 1 represents the next link
  • i is an integer
  • the state data s i, t of the link i and the link at the time point t of the start point of the link i can be obtained.
  • the transit time a i of i .
  • random values are used to set the transit time to select the model parameters of the model.
  • the first training algorithm is used, and the state data in the first sample data is used as the model input.
  • the transit time in this data is used as the output of the model, and the transit time selection model can be obtained by training according to the first sample data.
  • the first training algorithm can be various types of algorithms such as deep network training algorithms, recurrent neural network algorithms, decision tree algorithms, etc.
  • the trained transit time selection model can be deep network models, recurrent neural network models, Various types of models such as decision tree models.
  • the transit time selection model is used to determine the probability of multiple transit times according to the status data of any road section, that is, a set of probability distributions is determined, and each probability representation in the probability distribution corresponds to the corresponding The probability that the transit time passes the road section. The greater the probability, the more likely it is to pass the road section according to the transit time.
  • the transit time of the road section can be determined based on the probability of multiple transit times.
  • Each set of the second sample data includes a set of status data, a transit time of the link corresponding to the set of status data, and a set of states.
  • the next set of state data of the data training based on multiple sets of second sample data to obtain a state data prediction model.
  • the state data prediction model is used to determine the state data of the next road section of any road section according to the state data and travel time of any road section. For any road segment, the prediction model based on the state data can predict the state data of the next road segment of the road segment based on the state data and travel time of the road segment.
  • the status data at the starting point of each link, the transit time of the link, and the status data at the end of the link after passing the link according to the transit time can be obtained. That is, the corresponding relationship between the status data of the road segment, the travel time of the road segment and the status data of the next road segment of the road segment is obtained, and the corresponding relationship is used as a set of second sample data to obtain multiple sets of second sample data.
  • i represents the previous link
  • i + 1 represents the next link
  • i is an integer
  • the state data s i, t of the link i and the link at the time point t of the start point of the link i can be obtained.
  • the random parameters are used to set the model parameters of the state data prediction model.
  • a second training algorithm is used for each set of second sample data.
  • the state data and transit time in the second sample data are used as the model input.
  • the next set of state data in the sample data is used as the output of the model, and training is performed on the second sample data to obtain a state data prediction model.
  • the second training algorithm can be various types of algorithms such as deep network training algorithms, recurrent neural network algorithms, decision tree algorithms, etc.
  • the trained state data prediction model can be deep network models, recurrent neural network models, Various types of models such as decision tree models.
  • a string of segments with constant topological structure characteristics has an association relationship on the state data. Therefore, using a recurrent neural network algorithm to train the model will be more in line with the actual situation, and it will be possible to learn the status of the front and back segments at different points in time. Data conversion, which improves the accuracy of the model.
  • the training process can obtain a large number of sample routes, and each sample route is composed of multiple road segments, this can obtain a large number of two road segments that are adjacent to each other.Because there are too many road segment combinations, the sample data directly based on a large number of road segment combinations Performing training will cause too much calculation and exceed the limits of memory space and operation efficiency. And through observation, it is found that the difference between the status data of two adjacent road sections in the position is only the real-time speed of the road section and the real-time speed of the remaining road sections. Therefore, in addition to these two status data, the next road section The status data of other items can be calculated from the corresponding status data of the previous road section, and no model prediction is required. To this end, the model can be trained for two state data, the real-time speed of the road section and the real-time speed of the remaining road section.
  • the status data of any road section includes real-time data of the road section, and the real-time data refers to the real-time passing speed of the road section.
  • the state data prediction model includes a first speed prediction model, and the first speed prediction model is used to determine the speed of the road segment at the time point at the beginning of the road segment and the passage time of the road segment to determine the time after passing the road segment according to the passage time.
  • the real-time traffic speed of the next link at the time of the end of the link.
  • step 103 may include: constructing multiple sets of sample data based on the obtained historical traffic data, each set of sample data including the real-time speed of a road segment, the time of the road segment and the real-time speed of the next road segment, according to By training each set of sample data, a first speed prediction model can be obtained.
  • the state data prediction model by setting a first speed prediction model in the state data prediction model, it can be ensured that when making predictions based on the state data prediction model, according to the state data of any road segment and the state of the next road segment other than the real-time traffic speed, The data can predict the real-time traffic speed of the next road segment, so as to integrate the other item status data of the next road segment with the real-time traffic speed to obtain the complete status data of the next road segment.
  • the status data of any road section includes real-time data of the remaining road sections, and the real-time data of the remaining road sections refers to the real-time passing speed of each road section after the road section, that is, at the starting point of the road section, the The real-time traffic speed of each link after the link.
  • the state data prediction model includes a second speed prediction model, and the second speed prediction model is used to determine according to the travel time according to the real-time passing speed of each road segment after the road segment and the travel time of the road segment at the time point of the start of the road segment. The real-time traffic speed of each link after passing the link at the point in time when the link ends.
  • step 103 may include: constructing multiple sets of sample data according to the obtained historical traffic data, each group of sample data including the real-time passing speed of each road segment after a road segment, the passage time of the road segment, and the downgrade of the road segment.
  • the real-time traffic speed of each road section after a road section is trained according to each set of sample data to obtain a second speed prediction model.
  • the real-time speed of each road section is divided according to the state data of any road section and the next road section.
  • the status data of other items can predict the real-time traffic speed of each road segment after the next road segment, so as to integrate the status data of other items of the next road segment with the real-time traffic speed of each subsequent road segment to obtain the completeness of the next road segment.
  • Status data can predict the real-time traffic speed of each road segment after the next road segment, so as to integrate the status data of other items of the next road segment with the real-time traffic speed of each subsequent road segment to obtain the completeness of the next road segment.
  • the model may not be trained, and other methods may be used to predict these state data in the subsequent prediction process.
  • the global excitation value is used to measure the accuracy of the transit time of the sample route.
  • the higher the global excitation value the more accurate the transit time predicted by the sample route is, and the more consistent with the actual situation.
  • the local excitation value is used to measure the accuracy of the travel time of the corresponding road section. The higher the local excitation value is, the more accurate the predicted travel time of the road section is, the more consistent with the actual situation.
  • the step 104 may include the following steps 1041-1043:
  • a transit time selection model and a state data prediction model can be trained. For each link in the sample route, a model can be selected based on the transit time, and the predicted transit time of the link is determined based on the status data of the link.
  • the data prediction model can determine the status data of the next link according to the travel time of the link, and so on, to determine the predicted travel time of each link in the sample route.
  • the traffic strategy considering that if the traffic strategy is formulated only in the direction of the best value, it is likely to fall into a local optimum and lose other opportunities to obtain the maximum benefits. Therefore, in order to prevent the local optimal problem, when predicting the transit time for each link, first determine the optimal transit time of the link based on the transit time selection model, and then introduce noise in conjunction with other factors of the link to determine the other transit time of the link. , Such as suboptimal transit times, other possible transit times, etc. This can expand the search scope, ensure a more comprehensive search, and make the final traffic strategy more reasonable. For example, a Monte Carlo tree search strategy can be added to search, and the idea of random sampling is used to reduce the search space as much as possible and improve the search efficiency.
  • the sum of the predicted transit time of each link in the sample route is the predicted total transit time of the sample route.
  • the historical transit data of the sample route includes the actual total transit time of the sample route. The larger the first error between the predicted total transit time and the actual total transit time is, the more inaccurate the total transit time of the sample route is, so the global The excitation value is inversely proportional to the first error, and the global excitation value can be determined according to the first error.
  • the first error is determined based on the predicted total travel time and actual total travel time of the sample route:
  • mape Traj represents the first error
  • T represents the predicted total transit time
  • Traj represents the actual total transit time
  • abs represents the function of finding the absolute value after rounding
  • R finish represents the global excitation value
  • ⁇ represents the weight coefficient
  • the revenue value is calculated based on the global incentive value, which can ensure that the model can achieve the revenue value as much as possible at the link level. Accurate predictions.
  • the historical transit data of the sample route includes the actual transit time of each link.
  • the difference between the predicted transit time and the actual transit time is the second error.
  • the larger the error the less accurate the predicted transit time of the road section, so the local excitation value is inversely proportional to the second error, and the local excitation value can be determined based on the second error.
  • R each represents a local excitation value
  • ⁇ represents a weight coefficient
  • mape link represents a second error.
  • the revenue value is calculated based on the local incentive value, which can guarantee that the model can achieve the revenue value as much as possible at the link level. Accurate predictions.
  • the return value of the last state data in any sample route is equal to the sum of the global incentive value of the sample route and the local incentive value of each link in the sample route.
  • the benefit value of the last state data can be determined according to the sum of the global incentive value of the sample route and the local incentive value of each road segment.
  • the probability of the first section's transit time can be determined due to the selection of the model based on the transit time and the status data of the first segment, based on the status data
  • the prediction model can determine the status data of the second road segment based on the status data and travel time of the first road segment. That is, the condition that the probability of transitioning from the state data of the first link to the state data of the second link is equal to the probability of the transit time of the first link, and the condition of transitioning from the state data of the first link to the state data of the second link
  • the return value of the state data of the lower second road segment is equal to the local excitation value of the first road segment.
  • the revenue value of the next sample state data can be obtained, and the revenue value of the first sample state data can be obtained, so as to obtain the revenue of each state data in the sample route. Value.
  • the following formula is used to obtain the revenue value of the first sample state data:
  • si represents the first sample state data
  • V ⁇ (s i ) represents the revenue value of the first sample state data
  • s i + 1 represents the next sample state data
  • V ⁇ (s i + 1 ) represents the next A return value of a sample state data
  • S represents a set of all the next sample state data of the first sample state data in at least one sample route
  • ⁇ represents a traffic strategy composed of the transit time of multiple road sections
  • s i , a i ) represents the probability of transitioning from the first sample state data to the next sample state data, and is equal to the following when converting from the first sample state data to the next sample state data
  • s i , a i ) represents the return value of the next sample state data under the condition of transitioning from the first sample state data to the next sample state data, and is equal to the value from the first sample state data.
  • the local excitation value of a road segment when one sample state data is converted to the next sample state data, and ⁇ represents a discount factor.
  • the return value of the state data s under the ⁇ strategy is the expectation of the cumulative sum of the product of the return value and ⁇ j at each subsequent step, taking the first state data s0 as an example, the formula can be written as:
  • V ⁇ (s) E ⁇ [V 0 + ⁇ V 1 + ⁇ 2 V 2 + ⁇ 3 V 3 + ...
  • s s 0 ]
  • the historical traffic data of road segments 1 and 2 in multiple sample road segments is shown in Fig. 4.
  • passage 1 is passed according to the passage time a1, and then 6 types of passage times appear to pass through road segment 2.
  • the probability of transit time varies, and the sum is 100%.
  • a state transition diagram shown in FIG. 5 can be obtained.
  • each sample route includes road segment 1 and road segment 2.
  • the incentive value can be calculated as the global incentive value R
  • the sum of the finish and the local incentive values R each of the two road sections, and the return value of the last state data is equal to the incentive value, so the return values of the state data s11 to s16 can be calculated.
  • the return value of the state data s1 can be calculated using the following formula:
  • V (s1) P (s11
  • s1, a21) is the probability that the state data s1 passes through section 1 according to the transit time a21 and is converted to the state data s11, that is, the probability of the transit time a21
  • s1, a21) is the state data s1 according to The transit time a21 is converted to the state data s11 under the condition of the road segment 1, and the value of the state data s11, that is, the value of s11.
  • the other sections are similar and will not be repeated here.
  • the revenue value of the state data s0 can also be calculated in a similar manner, and then the revenue value of each state data is obtained.
  • the corresponding relationship between the state data and the return value can be determined, and the corresponding relationship is taken as a set of sample data, and then training is performed according to each set of sample data to obtain a return value prediction model.
  • the return value prediction model is used to A state data obtains the revenue value of the state data, and the revenue value is used to represent the expected future revenue under the current state.
  • a deep neural network algorithm can be used to train a numerical predictive model of revenue.
  • the revenue value prediction model initially uses random value presets. Through continuous trial and error learning, the revenue value prediction model can learn the rules of the revenue value corresponding to the state data, and update the model parameters in the revenue value prediction model. Constantly updating through trial and error learning makes the accuracy of the model constantly improve, and then continuously optimizes from a randomly initialized model until convergence, at which time the decisions made based on the numerical value prediction model will become more optimal.
  • the method provided in the embodiment of the present application obtains historical traffic data of each link in at least one sample route, can describe the sample route from a local perspective, and trains based on the acquired historical traffic data to obtain a transit time selection model and
  • the state data prediction model based on the transit time selection model and the status data prediction model, can predict the transit time and status data in units of road sections. It fully considers the local information of the route and improves the prediction accuracy.
  • a revenue numerical prediction model can be trained. Based on the revenue numerical prediction model, the revenue value of each state data can be predicted in units of road segments. The accuracy of the status data is measured by the revenue value, so that a more reasonable transit time is determined according to the status data, which improves the accuracy.
  • the prediction model adopted by the related technology basically has no ability to predict the traffic condition of the route. It only predicts based on the traffic condition at the departure time point, and there will be a large discrepancy between the transit time to the end point and the actual transit time.
  • the real-time traffic speed of the road section can be predicted.
  • the real-time traffic speed can indicate the traffic condition of the road section when it reaches the road section, so The impact of real-time traffic conditions can be taken into account during transit time, making predictions more accurate.
  • FIG. 6 is a flowchart of a transit time determination method provided by an embodiment of the present application.
  • the execution time of the transit time determination method is a prediction device, and the process of predicting the transit time of a target route will be described.
  • the prediction device may be a terminal or server such as a mobile phone, a computer, and the like, and the prediction device may be the same device as the training device in the above embodiment, or may be a different device.
  • the training device can provide the trained model to the prediction device for use by the prediction device. Referring to FIG. 6, the method includes:
  • the target route includes a plurality of road sections arranged in order, and any two road sections adjacent to each other are connected, and the end point of the previous road section is the starting point of the next road section.
  • the target route may be selected by the user or selected by the prediction device in the electronic map according to the starting point and the ending point. For example, when the user is going to a certain destination, the prediction device may use the user ’s current location as the starting point of the route, the destination as the ending point of the route, and one or more routes from the starting point to the ending point in the electronic map as Target routes to predict the total travel time for each target route.
  • the time selection model determines the probability of multiple transit times, and selects multiple alternative transit times for the first road segment from multiple transit times based on the probability of multiple transit times.
  • the transit time selection model is used to determine the probability of multiple transit times according to the status data of the road section, and then the first status data of the first road segment is input into the transit time selection model, and the multi-time selection model is used to determine multiple transit time probabilities.
  • Probability of multiple transit times, at this time multiple alternative transit times can be selected from multiple transit times according to the probability of multiple transit times. For example, a preset number of transit times is selected in order of the probability from large to small as the alternative transit time to obtain multiple alternative transit times. The preset number can be determined according to the accuracy requirements, or based on multiple transit times. The quantity and fixed selection ratio are determined.
  • inputting the state data s i, t of the link i into the transit time selection model can obtain the alternative transit time a i of the link i .
  • the state of data link i S i, t a i and alternative input to the transit time prediction model state data can be obtained through link i, at time t a i in accordance with an alternative transit time after the time point t +
  • the candidate state data s i + 1, t + 1 of the link i + 1 can be obtained through link i, at time t a i in accordance with an alternative transit time after the time point t +.
  • the status data of any road section includes real-time data of the road section, and the real-time data refers to the real-time speed of the road section, that is, the real-time speed of the road section at the beginning of the road section.
  • the real-time travel speed of the first road section and the alternative travel time of the first road section at the time point of the start of the first road section are input to the first speed prediction model, and based on the first speed prediction model, it is determined to pass according to the alternative travel time.
  • i represents the previous road segment
  • i + 1 represents the next road segment
  • i is an integer
  • the real-time passing speed V i of the road segment i at the time point t at the starting point of the road segment i, t and the transit time a i of the link i predicting the real-time transit speed V i + of the link i + 1 when the link i is reached at the time point t + 1 after passing the link i according to the transit time a i 1, t + 1 .
  • the real-time travel speed of the first road section and the real-time travel speed of the second road section at the time point of the start of the first road section, and the alternative travel time of the first road section may also be input.
  • the real-time passing speed of the second road section is determined at the time point when the first road section ends after passing the first road section according to the alternative travel time.
  • the real-time speed V i, t of the link i, the real-time speed V i + 1, t of the link i + 1, and the Passing time a i predicting the real-time passing speed V i + 1, t + 1 of link i + 1 when passing through link i according to the passing time a i and reaching the starting point of link i + 1 at time point t + 1 .
  • the real-time data of the remaining road sections refers to the real-time passing speed of each road section after the road section, that is, at the starting point of the road section, The real-time speed of each link after the link. Then, the real-time travel speed of each road segment after the first road segment and the alternative travel time of the first road segment at the time point of the start of the first road segment are input to the second speed prediction model. The real-time traffic speed of each link after the gated time passes the first link at the point in time when the first link ends.
  • i represents the previous road segment
  • i + 1 represents the next road segment
  • i is an integer
  • each road segment after the road segment i can pass in real time
  • the speed V left, i, t and the transit time a i of the link i are predicted after passing the link i according to the transit time a i and reaching the starting point of the link i + 1 at time point t + 1, after the link i + 1
  • the real-time speed of each road segment after the first road segment and the real-time speed of each road segment after the second road segment, and the first time is input to the second speed prediction model, and based on the second speed prediction model, the time of each road segment after the second road segment after passing the first road segment in accordance with the alternative travel time and at the end point of the first road segment is determined.
  • Real-time traffic speed is input to the second speed prediction model, and based on the second speed prediction model, the time of each road segment after the second road segment after passing the first road segment in accordance with the alternative travel time and at the end point of the first road segment.
  • the real-time traffic speed V left, i, t of each link after link i and the real-time traffic speed of each link after link i + 1 at time t at the starting point of link i V left, i + 1, t and the transit time a i of the link i It is predicted that after passing the link i according to the transit time a i and reaching the starting point of the link i + 1 at time point t + 1, the link i + 1 The real-time traffic speed V left, i + 1, t + 1 for each road section thereafter.
  • the initial speed in the second state data and the initial state data of the first state data is equal, so the initial speed is not changed. Because the road segment and time point both change, the historical statistical speed is replaced by the historical statistical speed of the second road segment at the time point at the end of the first road segment. The passage time of the first road section is increased based on the time.
  • the candidate state data corresponding to each candidate transit time is input into a revenue numerical prediction model, and based on the revenue numerical prediction model, the revenue value of each candidate state data is obtained, and the revenue value is used to measure
  • the accuracy of the prediction of the corresponding state data is to select the candidate state data with the largest revenue value from the candidate state data corresponding to multiple candidate transit times, determine it as the second state data, and select the candidate corresponding to the second state data.
  • the transit time is determined as the transit time of the first section.
  • the prediction device obtains the target route, for the first road section and the second road section which are adjacent at any two positions, according to the first state data of the first road section, based on the transit time selection model and the state data prediction model , Determine the transit time of the first road segment, and the second status data of the second road segment after passing the first road segment according to the transit time under the first status data.
  • the above steps 602-604 are optional steps, and the prediction device may also use other methods to determine the transit time of the first road section and the second status data.
  • the transit time selection model is used to determine the transit time of any road segment according to the status data of any road segment. Therefore, the first status data is input into the transit time selection model and the model is selected based on the transit time Determine the travel time of the first road segment, input the first status data and travel time into the status data prediction model, and determine the second status data based on the status data prediction model.
  • the second state data continue to determine the transit time of the second road segment based on the transit time selection model and the state data prediction model, until the transit time of each of the multiple road segments is determined.
  • the sum of the transit times of multiple sections in the target route is determined as the total transit time of the target route, and the current time point is taken as the departure time point of the user.
  • the time point reached after the total transit time of the departure time point is the arrival time point.
  • the method provided in the embodiment of the present application obtains a transit time selection model and a status data prediction model for a road section through training. Based on the transit time selection model and a status data prediction model, the travel time and status data can be predicted in units of road sections, and fully considered. The local information of the route is improved, the prediction accuracy is improved, and it can make up for the shortcomings of traditional prediction models.
  • a revenue numerical prediction model for road sections is obtained through training. Based on the revenue numerical prediction model, the revenue value of each state data can be predicted by the road segment unit, and the accuracy of the state data is measured by the revenue value, so that the more accurate Improved accuracy for reasonable transit times.
  • the prediction model adopted by the related technology basically has no ability to predict the traffic condition of the route. It only predicts based on the traffic condition at the departure time point, and there will be a large discrepancy between the transit time to the end point and the actual transit time.
  • the real-time traffic speed of the road section can be predicted.
  • the real-time traffic speed can indicate the traffic condition of the road section when it reaches the road section, so The impact of real-time traffic conditions can be taken into account during transit time, making predictions more accurate.
  • the transit time selection model and the status data prediction model together form a strategic network in the prediction scheme.
  • the transit time selection model is used to determine how much travel time is estimated for the current section in the current state. It is appropriate, and the state data prediction model selects the output of the model based on the transit time, and predicts the next state data, predicting what the next state may be the most, and what the specific state data is.
  • a numerical value prediction model is used to form a valuation network, and the profit of the state data reached at each transit time is estimated to obtain the correlation between the valuation network and the optimal strategy.
  • the model learns the relationship between changes in time, road segments, and real-time traffic speeds and changes in road capacity, and then uses the model to predict the travel time for each road segment.
  • the changes learned are taken into account, and the results given by the strategy network and the valuation network are used to determine the prediction results.
  • the embodiments of the present application provide a prediction scheme based on a reinforcement learning framework by using a transit time selection model, a state data prediction model, and a profit value prediction model.
  • the best decision process is obtained through training on sample data, and finally the prediction result can be output.
  • This forecasting scheme does not require detailed rule design. Through the level-level modeling, it can retain local information of the route, and has a certain reasoning ability. It can theoretically solve the problem of predicting changes in traffic conditions.
  • the reinforcement learning framework can be updated online, has good sensitivity to sample data, and can update data such as user traffic, user distribution changes, and traffic conditions in real time.
  • FIG. 7 is a schematic structural diagram of a transit time determining device according to an embodiment of the present application.
  • the device includes:
  • a route obtaining module 701 configured to perform the steps of obtaining a target route in the foregoing embodiment
  • the policy determining module 702 is configured to perform the steps of determining the transit time of the first road section and the second status data in the foregoing embodiment
  • the policy determining module 702 is further configured to perform the steps of determining the transit time of the second road segment according to the second status data in the foregoing embodiment, until the transit time of each of the multiple road segments is determined.
  • the policy determination module 702 includes:
  • a time determining unit configured to execute the step of determining the transit time of the first road segment based on the transit time selection model in the foregoing embodiment
  • the state determining unit is configured to execute the step of determining the second state data based on the state data prediction model in the foregoing embodiment.
  • the policy determination module 702 includes:
  • An alternative time determining unit configured to perform the steps of determining the probability of multiple transit times based on the transit time selection model in the above embodiment, and selecting multiple alternate transit times for the first road segment;
  • An alternative state determining unit configured to execute the above-mentioned embodiment, for each alternative transit time, based on the status data prediction model to determine the alternative status of the second link after passing the first link according to the alternative transit time under the first status data.
  • the policy determining unit is configured to perform the steps of determining the transit time of the first road segment and the second status data corresponding to the transit time according to the candidate status data corresponding to multiple candidate transit times in the foregoing embodiment.
  • the policy determination unit is further configured to execute the foregoing embodiment to obtain the return value of the candidate state data based on the return value prediction model, select the candidate state data with the largest return value, determine the second state data, and determine the first state data. Steps for the transit time of a road segment.
  • the policy determination unit is further configured to execute the foregoing embodiment to obtain the real-time passing speed of the first road section at the first time point, and to simulate and determine the real-time passing speed of the second road section at the second time point based on the first speed prediction.
  • the policy determination unit is further configured to obtain the real-time passing speed of each link after the first link at the first time point in the foregoing embodiment, and determine each link after the second link based on the second speed prediction model. Steps in real-time traffic speed at a second point in time.
  • the device further includes:
  • a sample acquisition module configured to perform the steps of acquiring historical traffic data of each link in at least one sample route in the foregoing embodiment
  • a first training module configured to perform the steps of constructing multiple sets of first sample data based on the acquired historical traffic data in the foregoing embodiment, and performing training to obtain a traffic time selection model
  • the second training module is configured to perform the steps of constructing a plurality of sets of second sample data based on the obtained historical traffic data and performing training to obtain a state data prediction model.
  • the device further includes:
  • An incentive acquisition module configured to perform the steps of acquiring the global excitation value of the sample route and the local excitation value of each link in the sample route for each sample route in the above embodiment
  • a revenue acquisition module configured to execute the first sample status data of any section of at least one sample route in the foregoing embodiment, and obtain a revenue value of the first sample status data until each of the at least one sample route is obtained Steps on the numerical value of status data;
  • the third training module is configured to perform the steps in the foregoing embodiment of training according to each state data and the return value of each state data to obtain a return value prediction model.
  • the incentive acquisition module includes:
  • a time prediction unit configured to perform the steps of selecting the model and the state data prediction model based on the currently trained travel time in the above embodiment, and determining the predicted travel time of each link in the sample route;
  • a global incentive obtaining unit is configured to execute the predicted total transit time of the sample route according to the predicted transit time of each road segment in the above embodiment, and calculate the first total transit time of the sample route and the actual total transit time of the sample route. An error, the step of determining the global excitation value of the sample route;
  • the local excitation obtaining unit is configured to perform the step of determining the local excitation value of each road segment according to the second error between the predicted travel time of each road segment and the actual travel time of each road segment in the above embodiment.
  • the gain obtaining module is configured to execute the above embodiment to obtain the value of the first sample state data using the following formula:
  • the transit time determination device when determining the transit time, only uses the division of the above functional modules as an example. In practical applications, the above functions may be allocated by different functional modules as required. That is, the internal structure of the training device or prediction device is divided into different functional modules to complete all or part of the functions described above.
  • the transit time determination device and the transit time determination method embodiments provided by the foregoing embodiments belong to the same concept. For specific implementation processes, refer to the method embodiments, and details are not described herein again.
  • FIG. 8 shows a structural block diagram of a terminal 800 provided by an exemplary embodiment of the present application.
  • the terminal 800 may be a portable mobile terminal, such as: smartphone, tablet, MP3 player (Moving Picture Experts Group Audio Layer III, moving image expert compression standard audio level 3), MP4 (Moving Picture Experts Group Audio Audio Layer IV, Image expert compression standard audio level 4) Player, laptop, desktop computer, head-mounted device, or any other smart terminal.
  • the terminal 800 may also be called other names such as user equipment, portable terminal, laptop terminal, desktop terminal, and the like.
  • the terminal 800 includes a processor 801 and a memory 802.
  • the processor 801 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like.
  • the processor 801 may use at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). achieve.
  • the processor 801 may also include a main processor and a co-processor.
  • the main processor is a processor for processing data in the awake state, also called a CPU (Central Processing Unit).
  • the co-processor is Low-power processor for processing data in standby.
  • the processor 801 may be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for rendering and drawing content required to be displayed on the display screen.
  • the processor 801 may further include an AI (Artificial Intelligence) processor, and the AI processor is configured to process computing operations related to machine learning.
  • AI Artificial Intelligence
  • the memory 802 may include one or more computer-readable storage media, which may be non-transitory.
  • the memory 802 may also include high-speed random access memory, and non-volatile memory, such as one or more disk storage devices, flash storage devices.
  • the non-transitory computer-readable storage medium in the memory 802 is configured to store at least one instruction, which is used by the processor 801 to implement the transit time provided by the method embodiment in the present application Determine the method.
  • the terminal 800 may optionally include a peripheral device interface 803 and at least one peripheral device.
  • the processor 801, the memory 802, and the peripheral device interface 803 may be connected through a bus or a signal line.
  • Each peripheral device can be connected to the peripheral device interface 803 through a bus, a signal line, or a circuit board.
  • the peripheral device includes at least one of a radio frequency circuit 804, a touch display screen 805, a camera 806, an audio circuit 807, a positioning component 808, and a power supply 809.
  • the peripheral device interface 803 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 801 and the memory 802.
  • the processor 801, the memory 802, and the peripheral device interface 803 are integrated on the same chip or circuit board; in some other embodiments, any one of the processor 801, the memory 802, and the peripheral device interface 803 or Two can be implemented on separate chips or circuit boards, which is not limited in this embodiment.
  • the radio frequency circuit 804 is used for receiving and transmitting an RF (Radio Frequency) signal, also called an electromagnetic signal.
  • the radio frequency circuit 804 communicates with a communication network and other communication devices through electromagnetic signals.
  • the radio frequency circuit 804 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals.
  • the radio frequency circuit 804 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and the like.
  • the radio frequency circuit 804 can communicate with other terminals through at least one wireless communication protocol.
  • the wireless communication protocol includes, but is not limited to: a metropolitan area network, various generations of mobile communication networks (2G, 3G, 4G, and 8G), a wireless local area network, and / or a WiFi (Wireless Fidelity) network.
  • the radio frequency circuit 804 may further include circuits related to Near Field Communication (NFC), which is not limited in this application.
  • NFC Near Field Communication
  • the display screen 805 is used to display a UI (User Interface).
  • the UI can include graphics, text, icons, videos, and any combination thereof.
  • the display screen 805 also has the ability to collect touch signals on or above the surface of the display screen 805.
  • the touch signal can be input to the processor 801 as a control signal for processing.
  • the display screen 805 may also be used to provide a virtual button and / or a virtual keyboard, which is also called a soft button and / or a soft keyboard.
  • the display screen 805 may be one, and the front panel of the terminal 800 is provided.
  • the display screen 805 may be at least two, which are respectively disposed on different surfaces of the terminal 800 or have a folded design. In still other embodiments, the display screen 805 may be a flexible display screen disposed on a curved surface or a folded surface of the terminal 800. Furthermore, the display screen 805 can also be set as a non-rectangular irregular figure, that is, a special-shaped screen.
  • the display screen 805 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).
  • the camera component 808 is used to capture images or videos.
  • the camera component 808 includes a front camera and a rear camera.
  • the front camera is disposed on the front panel of the terminal, and the rear camera is disposed on the back of the terminal.
  • the camera assembly 808 may further include a flash.
  • the flash can be a monochrome temperature flash or a dual color temperature flash.
  • a dual color temperature flash is a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.
  • the audio circuit 807 may include a microphone and a speaker.
  • the microphone is used to collect sound waves of the user and the environment, and convert the sound waves into electrical signals and input them to the processor 801 for processing, or input them to the radio frequency circuit 804 to implement voice communication.
  • the microphone can also be an array microphone or an omnidirectional acquisition microphone.
  • the speaker is used to convert electrical signals from the processor 801 or the radio frequency circuit 804 into sound waves.
  • the speaker can be a traditional film speaker or a piezoelectric ceramic speaker.
  • the speaker When the speaker is a piezoelectric ceramic speaker, it can not only convert electrical signals into sound waves audible to humans, but also convert electrical signals into sound waves inaudible to humans for ranging purposes.
  • the audio circuit 807 may further include a headphone jack.
  • the positioning component 808 is used to locate the current geographic position of the terminal 800 to implement navigation or LBS (Location Based Service).
  • the positioning component 808 may be a positioning component based on the United States' GPS (Global Positioning System), the Beidou system in China, the Granas system in Russia, or the Galileo system in the European Union.
  • the power supply 809 is used to power various components in the terminal 800.
  • the power source 809 may be an alternating current, a direct current, a disposable battery, or a rechargeable battery.
  • the rechargeable battery may support wired charging or wireless charging.
  • the rechargeable battery can also be used to support fast charging technology.
  • the terminal 800 further includes one or more sensors 810.
  • the one or more sensors 810 include, but are not limited to, an acceleration sensor 811, a gyroscope sensor 812, a pressure sensor 813, a fingerprint sensor 814, an optical sensor 815, and a proximity sensor 816.
  • the acceleration sensor 811 can detect the magnitude of acceleration on three coordinate axes of the coordinate system established by the terminal 800.
  • the acceleration sensor 811 may be used to detect components of the acceleration of gravity on three coordinate axes.
  • the processor 801 may control the touch display screen 805 to display the user interface in a horizontal view or a vertical view according to the gravity acceleration signal collected by the acceleration sensor 811.
  • the acceleration sensor 811 may also be used for collecting motion data of a game or a user.
  • the gyro sensor 812 can detect the body direction and rotation angle of the terminal 800, and the gyro sensor 812 can cooperate with the acceleration sensor 811 to collect a 3D motion of the user on the terminal 800. Based on the data collected by the gyro sensor 812, the processor 801 can implement the following functions: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.
  • the pressure sensor 813 may be disposed on a side frame of the terminal 800 and / or a lower layer of the touch display screen 805.
  • a user's holding signal to the terminal 800 can be detected, and the processor 801 can perform left-right hand recognition or quick operation according to the holding signal collected by the pressure sensor 813.
  • the processor 801 operates according to the user's pressure on the touch display screen 805 to control the operable controls on the UI interface.
  • the operability controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.
  • the fingerprint sensor 814 is used to collect a user's fingerprint, and the processor 801 recognizes the identity of the user based on the fingerprint collected by the fingerprint sensor 814, or the fingerprint sensor 814 recognizes the identity of the user based on the collected fingerprint. When identifying the user's identity as a trusted identity, the processor 801 authorizes the user to have related sensitive operations, which include unlocking the screen, viewing encrypted information, downloading software, paying, and changing settings.
  • the fingerprint sensor 814 may be provided on the front, back, or side of the terminal 800. When a physical button or a manufacturer's logo is set on the terminal 800, the fingerprint sensor 814 may be integrated with the physical button or the manufacturer's logo.
  • the optical sensor 815 is used to collect the ambient light intensity.
  • the processor 801 may control the display brightness of the touch display screen 805 according to the ambient light intensity collected by the optical sensor 815. Specifically, when the ambient light intensity is high, the display brightness of the touch display screen 805 is increased; when the ambient light intensity is low, the display brightness of the touch display screen 805 is decreased.
  • the processor 801 may also dynamically adjust the shooting parameters of the camera component 808 according to the ambient light intensity collected by the optical sensor 815.
  • the proximity sensor 816 also called a distance sensor, is usually disposed on the front panel of the terminal 800.
  • the proximity sensor 816 is used to collect the distance between the user and the front of the terminal 800.
  • the processor 801 controls the touch display screen 805 to switch from the bright screen state to the closed screen state; when the proximity sensor 816 detects When the distance between the user and the front of the terminal 800 gradually becomes larger, the processor 801 controls the touch display screen 805 to switch from the screen state to the bright screen state.
  • FIG. 8 does not constitute a limitation on the terminal 800, and may include more or fewer components than shown in the figure, or combine certain components, or adopt different component arrangements.
  • FIG. 9 is a schematic structural diagram of a server provided by an embodiment of the present application.
  • the server 900 may have a large difference due to different configurations or performance, and may include one or more processors (central processing units) (CPUs) 901 and one Or more than one memory 902, wherein the memory 902 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 901 to implement the methods provided by the foregoing method embodiments.
  • the server may also have components such as a wired or wireless network interface, a keyboard, and an input-output interface for input and output.
  • the server may also include other components for implementing device functions, and details are not described herein.
  • the server 900 may be configured to perform the steps performed by the prediction device in the foregoing transit time determination.
  • FIG. 10 shows an internal structure diagram of a computer device in one embodiment.
  • the computer device may specifically be the terminal 110 or the server 120 in FIG. 1A.
  • the computer device includes the computer device including a processor, a memory, a network interface, an input device, and a display screen connected through a system bus.
  • the memory includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium of the computer device stores an operating system and can also store computer-readable instructions.
  • the processor can implement a method for determining a transit time.
  • the internal memory may also store computer-readable instructions.
  • the processor may cause the processor to execute the transit time determination method.
  • the display screen of a computer device can be a liquid crystal display or an electronic ink display screen.
  • the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the computer device casing. It can be an external keyboard, trackpad, or mouse.
  • FIG. 10 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • the specific computer equipment may be Include more or fewer parts than shown in the figure, or combine certain parts, or have a different arrangement of parts.
  • the transit time determining device provided in this application may be implemented in the form of a computer-readable instruction, and the computer-readable instruction may run on a computer device as shown in FIG. 10.
  • the memory of the computer device may store various program modules constituting the transit time determination device, for example, the route acquisition module and the policy determination module shown in FIG. 7.
  • the computer-readable instructions constituted by the respective program modules cause the processor to execute the steps in the transit time determination method of the embodiments of the present application described in this specification.
  • the computer device shown in FIG. 10 may execute the route acquisition module in the transit time determination device shown in FIG. 7 to obtain a target route to be passed, and the target route includes a plurality of road segments arranged in order.
  • the computer device can execute the policy determination module from the first link of the multiple links. For the first link and the second link that are adjacent at any two positions, according to the first state data of the first link, based on the transit time
  • the model and the state data prediction model are selected to determine the travel time of the first road section, and the second state data of the second road section after passing the first road section according to the travel time under the first state data.
  • the computer device may also execute, based on the second status data, the policy determination module to continue to determine the transit time of the second road segment based on the transit time selection model and the status data prediction model until the transit time of each of the multiple road segments is determined.
  • a computer device which includes a memory and a processor.
  • the memory stores computer-readable instructions.
  • the processor causes the processor to perform the steps of the transit time determination method described above.
  • the steps of the transit time determination method herein may be the steps in the transit time determination methods of the foregoing embodiments.
  • a computer-readable storage medium which stores computer-readable instructions.
  • the processor is caused to execute the steps of the foregoing transit time determination method.
  • the steps of the transit time determination method herein may be the steps in the transit time determination methods of the foregoing embodiments.
  • Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Synchlink DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM dual data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM synchronous chain Synchlink DRAM
  • Rambus direct RAM
  • DRAM direct memory bus dynamic RAM
  • RDRAM memory bus dynamic RAM

Abstract

一种通行时间确定方法、装置、计算机设备及存储介质。该方法包括：计算机设备获取待通行的目标路线；计算机设备从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据；计算机设备根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间。

Description

通行时间确定方法、装置、计算机设备及存储介质
本申请要求于2018年07月23日提交中国专利局，申请号为2018108142901，申请名称为“通行时间确定方法、装置、终端及存储介质”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
本申请实施例涉及计算机技术领域，特别涉及一种通行时间确定方法、装置、计算机设备及存储介质。
背景技术
随着互联网技术的发展和移动终端的普及，导航功能广泛应用于日常生活中，当用户要通过一段路线时，利用移动终端的导航功能可以预测路线的通行时间，从而预测用户到达目的地的时间点，为用户的出行带来了很多便利。
相关技术中通常采用时间点预测模型来确定用户到达目的地的时间点。在模型训练阶段，获取至少一条样本路线的样本数据，样本数据中包括对应样本路线的路线描述数据和历史通行数据，该路线描述数据用于描述样本路线的地理情况，历史通行数据至少包括该样本路线的通行时间。根据获取到的多个样本数据进行训练，得到时间点预测模型，该时间点预测模型模型可以用于预测任一条路线的预计到达时间。那么，当用户要通过目标路线时，可以将该目标路线的路线描述数据和当前时间点输入至时间点预测模型中，基于该时间点预测模型确定该目标路线的预计到达时间，即为用户到达该目标路线的目的地的时间点。
但是，上述方案中仅根据路线的全局信息来训练时间点预测模型，未考虑到路线的局部信息，导致基于时间点预测模型仅能根据路线的全局信息来确定预计到达时间，预测不够准确。
发明内容
根据本申请提供的各种实施例，提供了一种通行时间确定方法、装置、计算机设备及存储介质。一方面，提供了一种通行时间确定方法，该方法包括：
计算机设备获取待通行的目标路线，目标路线包括按照顺序排列的多个路段；
计算机设备从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据；
计算机设备根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间；
其中，通行时间选取模型用于根据任一路段的状态数据确定任一路段的通行时间，状态数据预测模型用于根据任一路段的状态数据和通行时间，确定任一路段的下一路段的状态数据。
再一方面，提供了一种通行时间确定装置，该装置包括：
路线获取模块，用于获取待通行的目标路线，目标路线包括按照顺序排列的多个路段；
策略确定模块，用于从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据；
策略确定模块，还用于根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间；
其中，通行时间选取模型用于根据任一路段的状态数据确定任一路段的通行时间，状态数据预测模型用于根据任一路段的状态数据和通行时间，确定任一路段的下一路段的状态数据。
一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行以下步骤：
获取待通行的目标路线，目标路线包括按照顺序排列的多个路段；
从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据；
根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间；
其中，通行时间选取模型用于根据任一路段的状态数据确定任一路段的通行时间，状态数据预测模型用于根据任一路段的状态数据和通行时间，确定任一路段的下一路段的状态数据。
一种计算机可读存储介质，存储有计算机可读指令，所述计算机可读指令被处理器执行时，使得所述处理器执行以下步骤：
获取待通行的目标路线，目标路线包括按照顺序排列的多个路段；
从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据；
根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间；
其中，通行时间选取模型用于根据任一路段的状态数据确定任一路段的通行时间，状态数据预测模型用于根据任一路段的状态数据和通行时间，确定任一路段的下一路段的状态数据。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1A是本申请实施例提供的一种通行时间确定方法的应用环境图；
图1是本申请实施例提供的一种模型训练方法的流程图；
图2是本申请实施例提供的一种实时通行速度预测示意图；
图3是本申请实施例提供的一种剩余路段的实时通行速度预测示意图；
图4是本申请实施例提供的一种历史通行数据示意图；
图5是本申请实施例提供的一种状态转换示意图；
图6是本申请实施例提供的一种通行时间确定方法的流程图；
图7是本申请实施例提供的一种通行时间确定装置的结构示意图；
图8是本申请实施例提供的一种终端的结构示意图；
图9是本申请实施例提供的一种服务器的结构示意图；
图10是本申请实施例提供的一种计算机设备的结构示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
在对本申请实施例进行详细说明之前，首先对涉及到的概念进行如下介绍：
1、强化学习框架：包括Agent(智能体)、状态、动作、激励、价值和马尔科夫决策过程等成员。
一般的监督学习框架应用于预测ETA(Estimated Time of Arrival，预计到达时间)的场景时，在训练阶段会训练出一个预测模型，该预测模型实质上是一种路线特征与ETA的映射关系，通过样本数据进行训练可以使预测模型尽可能地与样本数据相符，准确度更高。在预测阶段，根据路线特征以及预测模型中确定的映射关系即可得到路线特征对应的ETA。
而强化学习框架不同于一般的监督学习框架，在强化学习框架中，Agent通过反馈机制，在不同的状态下反复尝试做出不同的动作，并根据做出的每个动作得到的收益，一步一步优化反馈机制，最终在马尔科夫决策过程(Markov Decision Process，简称MDP)中找到能够得到最大收益的决策序列。因此，强化学习框架的训练结果并不是一个数值，而是在状态-动作空间下的收益分布。
那么，将强化学习框架应用于预测ETA的场景时，其输入为包括多个路段的目标路线，其输出为决策出的最优策略，该最优策略中包括每个路段的通行时间。
2、路段：用来描述路线的最小单元，一条路线由多个路段组成，每个路段采用一组结构化的物理描述数据进行描述，包括但不限于路段的长度、宽度、包含红绿灯的个数、道路等级等。
3、路段的状态数据：可以包括以下至少一项：
3-1、初始速度：是指处于目标路线的起点时第一个路段的实时通行速度， 同一路线中的不同路段对应的初始速度相等。
3-2、历史统计速度：根据路段的历史通行数据进行统计，得到路段在某一时间点的通行速度的统计值。对于同一路段来说，不同时间点的历史统计速度可能不同。
3-3累积通行时间：在路段之前的一个或多个路段累积的总通行时间。
3-4、路段的实时数据：包括实时通行速度、统计数据和物理描述数据；
其中，路段的实时通行速度是指在路段起点时路段的实时通行速度，同一个路段在不同时间点的实时通行速度可能不同。
路段的统计数据是指针对路段的历史通行情况统计出的数据，包括但不限于路段在畅通通行时的通行速度以及在某段时间区间内的多个历史统计速度。
路段的物理描述数据用于描述路段的地理情况，包括路段的长度、宽度、包含的红绿灯个数、道路等级等。
3-5、剩余路段的实时数据：包括路段之后的每个剩余路段的实时通行速度、统计数据和物理描述数据，具体的数据格式与路段的实时数据类似，在此不再赘述。
4、动作：是指按照一定的通行时间来通过某一路段的动作，每个路段的动作采用通行时间来表示。
每次执行一个动作，即按照一定的通行时间通过某一路段后，Agent所处的状态会发生变化，该变化包括：历史统计速度会切换为下一个路段的历史统计速度、累积通行时间会增加新通过路段的通行时间、路段的实时数据会切换为下一个路段的实时数据、剩余路段的实时数据中会将新通过路段的实时数据去除。
5、即时激励数值：在通过某一路段后，针对该路段的通行时间反馈的激励，以R each来表示。
最终激励数值：通过一条路线、到达该路线的目的地后，针对该整条路线的通行时间反馈的激励，以R finish来表示。
状态数据的收益数值：是指在某一时间点处于某种状态时未来能获得收益的期望，该收益数值可以用来衡量状态数据的预测准确程度，收益数值越高，表示当前时间点处于当前状态的情况与实际情况越相符，带来的误差越小，所制定的策略越准确。
通行时间的收益数值：是指在某一时间点处于某种状态时按照某一通行时间通过当前路段的条件下，未来能获得收益的期望，该收益数值越高，表示当前 时间点处于当前状态的情况下，按照该通行时间通过当前路段时，期望获得的收益越大，即该情况与实际情况越相符，带来的误差越小，所制定的策略越准确。
根据状态数据的收益数值和通行时间的收益数值，可以决定是否要按照制定的通行时间通过当前路段。
6、马尔科夫决策过程：MDP<S，A，P，R，γ>，其中S表示状态数据的集合，A表示通行时间的集合，P表示状态转移概率矩阵，状态转移概率矩阵中的每个元素表示由前一组状态数据转移到下一组状态数据的概率，R表示激励，γ表示折扣因子，用来计算累积的收益数值。
在预测ETA的场景下，一个马尔科夫决策过程可以如下：
Agent处于S集合中的某一状态s时可能会执行A集合中的n个动作a。对于每一个不同的动作a，Agent模拟执行动作a后会对状态s产生影响，达到新状态s’。在此过程中，Agent会收到动作a对应的即时激励，并计算新状态s’的收益数值，最终选择n个动作中即时激励与收益数值的总和最大的动作来执行。
7、通行时间选取模型：根据路段的状态数据确定该路段通行时间的模型。
状态数据预测模型：根据上一个路段的状态数据和通行时间，预测下一个路段状态数据的模型。
状态数据预测模型可以包括第一速度预测模型和第二速度预测模型中的至少一个，第一速度预测模型用于根据路段的实时通行速度和通行时间预测当到达下一个路段时下一个路段的实时通行速度，第二速度预测模型用于根据当前路段的通行时间和当前路段之后的剩余路段的实时通行速度预测当到达下一个路段时，下一个路段之后的剩余路段的实时通行速度。
收益数值预测模型：根据状态数据获取收益数值的模型，以收益数值来表示在当前所处状态下未来期望获得的收益。
相关技术中提出了一种基于监督学习的机器学习方案，可以用于确定路线的通行时间。在该方案中，根据多条样本路线的样本数据训练出时间点预测模型，对于用户待通过的目标路线，可以基于该时间点预测模型确定该目标路线的预计到达时间。
但是，由于训练过程中采用的样本数据实仅能体现样本路线的全局信息，能够从全局的角度来描述样本路线，但是无法从局部的角度来样本路线，也即是训练时间点预测模型时未考虑路线的局部信息，这会导致基于时间点预测模型仅 能根据路线的全局信息来确定预计到达时间，而丢失了路线的局部信息，因此预测不够准确。
为了提高预测准确度，本申请实施例提出了一种确定通行时间的方案，先训练出以路段为单位的通行时间选取模型和状态数据预测模型，针对用户待通行的目标路线，可以根据目标路线中的每个路段，基于通行时间选取模型和状态数据预测模型，确定每个路段的通行时间，充分考虑了目标路线中每个路段的局部信息，分别预测每个路段的通行时间，提高了预测准确度。
本申请实施例应用于预测目标路线的通行时间的场景下，例如，在地图导航的场景下，当用户要出发去往目的地时，可以根据用户当前所在的位置和该目的地确定多条路线，并采用本申请实施例提供的方法预测每条路线中每个路段的通行时间，从而预测出每条路线的总通行时间，由用户选择总通行时间较短的路线。当然，本申请实施例还可以应用于其他需要预测目标路线的通行时间的场景下。
在一种实现方式中，终端可以安装地图导航应用，在地图导航应用中可以采用本申请实施例提供的方法，预测目标路线的通行时间。
图1A为一个实施例中通行时间确定方法的应用环境图。参照图1A，该通行时间确定方法应用于通行时间确定系统。该通行时间确定系统包括终端110和服务器120。终端110和服务器120通过网络连接。终端110具体可以是台式终端或移动终端，移动终端具体可以手机、平板电脑、笔记本电脑等中的至少一种。服务器120可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
具体地，终端110将待通行的目标路线发送至服务器120，其中目标路线包括按照顺序排列的多个路段。服务器120获取到待通行的目标路线，从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据，根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间。其中，通行时间选取模型用于根据任一路段的状态数据确定任一路段的通行时间，状态数据预测模型用于根据任一路段的状态数据和通行时间，确定任一路段的下一路段的状态数据。最后，服务器120可将多个路段中每个路段的通行 时间返回至终端110。
图1是本申请实施例提供的一种模型训练方法的流程图，该模型训练方法的执行主体为训练装置，对训练模型的过程进行说明。其中，该训练装置可以为具备导航功能的手机、计算机等终端或者服务器等。参见图1，该方法包括：
101、获取至少一条样本路线中每个路段的历史通行数据。
每条样本路线包括按照顺序排列的多个路段，每条样本路线中每个路段的历史通行数据可以根据对样本设备的移动过程进行收集得到。该样本设备可以包括手机、车载电脑、平板电脑等多种类型的设备。
收集过程中，可以获取电子地图，该电子地图包括多个路段，在任一样本设备移动的过程中，可以对样本设备进行定位，实时确定样本设备所在的位置，并根据该电子地图、样本设备所在的位置以及对应的时间点，收集该样本设备在每个路段的历史通行数据，从而得到一条样本路线中每个路段的历史通行数据。采用此种收集方式可以针对多个样本设备收集到多条样本路线中每个路段的历史通行数据。
其中，路段的历史通行数据包括路段的通行时间和状态数据。该通行时间即为样本设备通过该路段所耗费的时间，该状态数据可以包括初始速度、历史统计速度、累积通行时间、路段的实时数据、剩余路段的实时数据中的至少一项，另外还可以包括其他能够表示样本设备当前所处状态的数据。
例如，路段i的状态数据为S i＝<V 0，V hts，i，T sum，i，L i，L left，i>。
其中，V 0表示初始速度，即处于路线起点时的实时通行速度；
V hts，i表示当前从路线起点出发，当处于路段i的起点时的时间点路段i的历史统计速度；
T sum，i表示当前从路线的起点出发，当处于路段i的起点时的时间点，路段i之前的路段累积的总通行时间，即为路段i之前的每个路段的通行时间的总和；
L i表示路段i的实时数据，包括实时通行速度、统计数据和物理描述数据。其中实时通行速度是指当前从路线的起点出发，当处于路段i的起点时的时间点路段i的实时通行速度，统计数据包括路段i在畅通通行时的通行速度以及在至少一段时间区间内的多个历史统计速度，物理描述数据用于描述路段i的地理情况，可以包括路段i的长度、宽度、包含的红绿灯个数、道路等级等。
L left，i表示路段i之后每个路段的实时数据，包括之后每个路段的实时通行 速度、统计数据和物理描述数据。
实际上，训练装置直接收集到的每个路段的历史通行数据包括样本设备经过某一地点的时间点和通行速度，则对于电子地图中的每个路段来说，根据样本设备经过路段起点时的时间点与经过路段终点时的时间点可以确定路段的通行时间。并且，根据该路段的历史通行数据以及该路段之后的各个路段的历史通行数据，可以获取到在该路段起点时的各项状态数据，即为该路段的状态数据。
102、对于每条样本路线，根据样本路线中每个路段的历史通行数据，构造多组第一样本数据，每组第一样本数据中包括一组状态数据以及与一组状态数据对应的路段的通行时间；根据多组第一样本数据进行训练，得到通行时间选取模型。
其中，通行时间选取模型用于根据任一路段的状态数据确定任一路段的通行时间。对于任一路段，基于通行时间选取模型可以根据该路段的状态数据，预测出该路段的通行时间。
当获取到一条样本路线中每个路段的历史通行数据，即可获取到该路段的状态数据以及通行时间，也即是获取到了状态数据与通行时间的对应关系，将该对应关系作为一组第一样本数据，从而得到多组第一样本数据。例如，以i表示上一个路段，以i+1表示下一个路段，i为整数，根据历史通行数据可以获取到在路段i起点的时间点t时，路段i的状态数据s i,t和路段i的通行时间a i。
在训练最初采用随机值设置通行时间选取模型的模型参数，之后对于每组第一样本数据，采用第一训练算法，将第一样本数据中的状态数据作为模型的输入，将第一样本数据中的通行时间作为模型的输出，根据第一样本数据进行训练即可得到通行时间选取模型。后续过程中针对下一组样本数据，还可以继续对通行时间选取模型进行训练，提高通行时间选取模型的准确度。
其中，该第一训练算法可以为深度网络训练算法、循环神经网络算法、决策树算法等多种类型的算法，相应的，所训练的通行时间选取模型可以为深度网络模型、循环神经网络模型、决策树模型等多种类型的模型。
在一种实现方式中，该通行时间选取模型用于根据任一路段的状态数据确定多个通行时间的概率，也即是确定了一组概率分布，该概率分布中的每个概率表示按照对应通行时间通过该路段的概率，概率越大，表示越有可能按照该通行时间通过该路段，则根据多个通行时间的概率可以确定该路段的通行时间。
103、根据样本路线中每个路段的历史通行数据，构造多组第二样本数据， 每组第二样本数据中包括一组状态数据、与一组状态数据对应的路段的通行时间以及一组状态数据的下一组状态数据；根据多组第二样本数据进行训练，得到状态数据预测模型。
其中，状态数据预测模型用于根据任一路段的状态数据和通行时间，确定任一路段的下一路段的状态数据。对于任一路段，基于状态数据预测模型可以根据该路段的状态数据和通行时间，预测出该路段的下一个路段的状态数据。
当获取到一条样本路线中每个路段的历史通行数据，即可获取到在每个路段的起点的状态数据、该路段的通行时间以及按照该通行时间通过该路段后在该路段终点的状态数据，也即是获取到了路段的状态数据、该路段的通行时间与该路段的下一个路段的状态数据的对应关系，将该对应关系作为一组第二样本数据，从而得到多组第二样本数据。例如，以i表示上一个路段，以i+1表示下一个路段，i为整数，根据历史通行数据可以获取到在路段i起点的时间点t时，路段i的状态数据s i,t和路段i的通行时间a i，并且假设按照该通行时间a i通过路段i后，会在时间点t+1到达路段i的终点，即路段i+1的起点，此时会获取到路段i+1的状态数据s i+1,t+1。
在训练最初采用随机值设置状态数据预测模型的模型参数，之后对于每组第二样本数据，采用第二训练算法，将第二样本数据中的状态数据和通行时间作为模型的输入，将第二样本数据中的下一组状态数据作为模型的输出，根据第二样本数据进行训练，即可得到状态数据预测模型。后续过程中针对下一组样本数据，还可以继续对状态数据预测模型进行训练，提高状态数据预测模型的准确度。
其中，该第二训练算法可以为深度网络训练算法、循环神经网络算法、决策树算法等多种类型的算法，相应的，所训练的状态数据预测模型可以为深度网络模型、循环神经网络模型、决策树模型等多种类型的模型。尤其是，在一条路线中，具有拓扑结构特征恒的路段串在状态数据上具有关联关系，因此采用循环神经网络算法训练模型会更符合实际情况，更能学习到前后路段在不同时间点的状态数据转换情况，从而提高了模型的准确度。
由于训练过程可以获取到大量的样本路线，而每一条样本路线由多个路段组成，这样可以得到大量在位置上相邻的两个路段，由于路段组合过多，直接根据大量路段组合的样本数据进行训练，会导致计算量过大，超出内存空间和运算效率的限制。而通过观察发现，位置上相邻的两个路段的状态数据，之间的差异仅在于路段的实时通行速度和剩余路段的实时通行速度，因此，除这两项状态数 据之外，下一个路段的其他项状态数据均可通过上一个路段的相应状态数据计算得到，无需通过模型进行预测。为此，可以针对路段的实时通行速度和剩余路段的实时通行速度这两项状态数据训练模型。
在一种实现方式中，任一路段的状态数据包括该路段的实时数据，该实时数据是指该路段的实时通行速度。则状态数据预测模型包括第一速度预测模型，第一速度预测模型用于根据在路段起点时的时间点该路段的实时通行速度以及该路段的通行时间，确定按照该通行时间通过该路段之后在该路段终点时的时间点下一个路段的实时通行速度。
相应地，步骤103可以包括：根据获取到的历史通行数据，构造多组样本数据，每组样本数据中包括一个路段的实时通行速度、该路段的通行时间和下一个路段的实时通行速度，根据每组样本数据进行训练，可以得到第一速度预测模型。
参见图2，通过在状态数据预测模型中设置第一速度预测模型，可以保证基于状态数据预测模型进行预测时，根据任一路段的状态数据和下一路段除实时通行速度之外的其他项状态数据，即可预测出下一路段的实时通行速度，从而将下一路段的其他项状态数据与实时通行速度整合，得到下一路段完整的状态数据。
在另一种实现方式中，任一路段的状态数据包括剩余路段的实时数据，该剩余路段的实时数据是指该路段之后的每个路段的实时通行速度，即在该路段的起点时，该路段之后的每个路段的实时通行速度。则状态数据预测模型包括第二速度预测模型，第二速度预测模型用于根据在路段起点时的时间点该路段之后的每个路段的实时通行速度以及该路段的通行时间，确定按照该通行时间通过该路段之后在该路段终点时的时间点下一个路段之后的每个路段的实时通行速度。
相应地，步骤103可以包括：根据获取到的历史通行数据，构造多组样本数据，每组样本数据中包括一个路段之后的每个路段的实时通行速度、该路段的通行时间以及该路段的下一个路段之后的每个路段的实时通行速度，根据每组样本数据进行训练，可以得到第二速度预测模型。
参见图3，通过在状态数据预测模型中设置第二速度预测模型，可以保证基于状态数据预测模型进行预测时，根据任一路段的状态数据和下一路段除之后每个路段的实时通行速度之外的其他项状态数据，即可预测出下一路段之后每个路段的实时通行速度，从而将下一路段的其他项状态数据与之后每个路段的 实时通行速度整合，得到下一路段完整的状态数据。
对于除上述路段的实时数据和剩余路段的实时数据之外的其他项状态数据，也可以不训练模型，后续预测过程中可以采用其他方式来预测这些状态数据。
104、获取样本路线的全局激励数值和样本路线中每个路段的局部激励数值。
其中，全局激励数值用于衡量样本路线的通行时间的准确程度，全局激励数值越高，表示为样本路线预测出的通行时间越准确，越符合实际情况。局部激励数值用于衡量对应路段的通行时间的准确程度，局部激励数值越高，表示为路段预测出的通行时间越准确，越符合实际情况。
在一种实现方式中，该步骤104可以包括以下步骤1041-1043：
1041、基于当前训练的通行时间选取模型和状态数据预测模型，确定样本路线中每个路段的预测通行时间。
通过上述步骤102和103可以训练出通行时间选取模型和状态数据预测模型，则对于样本路线中的每个路段，可以基于通行时间选取模型，根据路段的状态数据确定路段的预测通行时间，基于状态数据预测模型又可以根据路段的通行时间确定下一个路段的状态数据，以此类推，即可确定样本路线中每个路段的预测通行时间。
在一种实现方式中，考虑到如果仅按照价值最优的方向制定通行策略，很可能会陷入局部最优而失去其他获得最大收益的机会。因此，为了防止局部最优的问题，在为每个路段预测通行时间时，先基于通行时间选取模型确定路段的最优通行时间，另外再结合路段的其他因素引入噪声，确定路段其他的通行时间，如次优的通行时间、其他可能的通行时间等。这样可以扩大搜索范围，保证搜索更加全面，使最终制定的通行策略更为合理。例如，可以加入蒙特卡洛树搜索策略来进行搜索，并采用随机采样的思想尽可能地降低搜索空间，提高搜索效率。
1042、根据每个路段的预测通行时间，获取样本路线的预测总通行时间，根据样本路线的预测总通行时间与样本路线的实际总通行时间之间的第一误差，确定样本路线的全局激励数值，全局激励数值与第一误差呈反比关系。
样本路线中每个路段的预测通行时间之和即为样本路线的预测总通行时间。该样本路线的历史通行数据中包括该样本路线的实际总通行时间，预测总通行时间与实际总通行时间之间的第一误差越大，表示样本路线预测的总通行时间越不准确，因此全局激励数值与第一误差呈反比关系，可以根据第一误差确定全局激励数值。
例如，采用以下公式，根据样本路线的预测总通行时间和实际总通行时间确定第一误差：
Figure PCTCN2019091311-appb-000001
采用以下公式，根据第一误差确定全局激励数值：
Figure PCTCN2019091311-appb-000002
其中，mape Traj表示第一误差，T表示预测总通行时间，Traj表示实际总通行时间，abs表示取整后求绝对值的函数，R finish表示全局激励数值，α表示权重系数。
通过将第一误差与全局激励数值关联起来，使两者呈反比关系可以保证误差越小全局激励数值越大，因此根据全局激励数值计算得到收益数值，可以保证模型尽量在路段级别上实现收益数值的准确预测。
1043、根据每个路段的通行时间与每个路段的实际通行时间之间的第二误差，确定每个路段的局部激励数值，局部激励数值与第二误差呈反比关系。
该样本路线的历史通行数据中包括每个路段的实际通行时间，对于该样本路线中的每个路段来说，预测通行时间与实际通行时间之间的差值即为第二误差，该第二误差越大表示路段预测的通行时间越不准确，因此局部激励数值与第二误差呈反比关系，可以根据第二误差确定局部激励数值。
例如，采用以下公式，根据第二误差确定局部激励数值：
Figure PCTCN2019091311-appb-000003
其中，R each表示局部激励数值，β表示权重系数，mape link表示第二误差。
通过将第二误差与局部激励数值关联起来，使两者呈反比关系可以保证误差越小局部激励数值越大，因此根据局部激励数值计算得到收益数值，可以保证模型尽量在路段级别上实现收益数值的准确预测。
105、对于样本路线中的任一路段的第一样本状态数据，根据第一样本状态数据在每条样本路线中的下一样本状态数据的收益数值、从第一样本状态数据转换到下一样本状态数据的概率和从第一样本状态数据转换到下一样本状态数据的条件下该下一样本状态数据的收益数值，获取第一样本状态数据的收益数值，直至获取到样本路线中每个状态数据的收益数值。
其中，任一样本路线中最后一个状态数据的收益数值等于样本路线的全局激励数值与样本路线中每个路段的局部激励数值的总和。
那么，对于每条样本路线来说，可以先根据该样本路线的全局激励数值和每 个路段的局部激励数值的总和，确定最后一个状态数据的收益数值。
对于样本路线中任两个位置上相邻的第一路段和第二路段来说，由于基于通行时间选取模型并根据第一路段的状态数据可以确定第一路段的通行时间的概率，基于状态数据预测模型并根据第一路段的状态数据和通行时间可以确定第二路段的状态数据。也即是，从第一路段的状态数据转换到第二路段的状态数据的概率等于该第一路段的通行时间的概率，而从第一路段的状态数据转换到第二路段的状态数据的条件下第二路段的状态数据的收益数值等于该第一路段的局部激励数值。
因此，从最后一个状态数据开始，对于任一路段的第一样本状态数据，根据下一样本状态数据的收益数值、从第一样本状态数据转换到下一样本状态数据的概率和从第一样本状态数据转换到下一样本状态数据的条件下该下一样本状态数据的收益数值，即可获取到第一样本状态数据的收益数值，从而得到样本路线中每个状态数据的收益数值。
在一种实现方式中，采用以下公式，获取第一样本状态数据的收益数值：
Figure PCTCN2019091311-appb-000004
其中，s i表示第一样本状态数据，V π(s i)表示第一样本状态数据的收益数值，s i+1表示下一样本状态数据，V π(s i+1)表示下一样本状态数据的收益数值，S表示第一样本状态数据在至少一条样本路线中所有的下一样本状态数据构成的集合；π表示由多个路段的通行时间构成的通行策略；
P(s i+1|s i,a i)表示从第一样本状态数据转换到下一样本状态数据的概率，且等于从第一样本状态数据转换到下一样本状态数据时按照的通行时间的概率，R(s i+1|s i,a i)表示从第一样本状态数据转换到下一样本状态数据的条件下该下一样本状态数据的收益数值，且等于从第一样本状态数据转换到下一样本状态数据时所通过路段的局部激励数值，γ表示折扣因子。
通过贝尔曼方程(Bellman Equation)可知，从当前的决策时间点(可以认为是出发时间点)至整条路线的决策结束时间点，每个路段的状态数据的收益数值为：
Figure PCTCN2019091311-appb-000005
即，假设在π策略下状态数据s的收益数值为后续每一步的收益数值与γ j的 乘积的累积和的期望，以第一个状态数据s0为例，该公式可以写为：
V π(s)＝E π[V 0+γV 1+γ 2V 2+γ 3V 3+…|s＝s 0]
＝E π[V 0+γE π[V 1+γ 1V 2+γ 2V 3+…]|s＝s 0]
＝E π[R(s 1|s 0,a 0)+γV π(s 1)|s＝s 0]
对于任一状态数据s i，
Figure PCTCN2019091311-appb-000006
举例来说，多条样本路段中路段1和路段2的历史通行数据如图4所示，在状态s0时按照通行时间a1通过路段1，之后出现了6种通行时间来通过路段2，每种通行时间的概率各不相同，且总和为100％。基于图4所示的历史通行数据可以得到如图5所示的状态转换示意图。
参见图5，由于路段2出现了6种通行时间，因此构成了6条样本路线，每条样本路线中包括路段1和路段2，每条样本路线中，可以计算出激励数值为全局激励数值R finish和两个路段的局部激励数值R each的总和，且最后一个状态数据的收益数值与激励数值相等，因此可以计算得到状态数据s11至s16的收益数值。之后，状态数据s1的收益数值可以采用以下公式计算得到：
V(s1)＝P(s11|s1,a21)[R(s11|s1,a21)+γV(s11)+…
+P(s16|s1,a26)[R(s11|s1,a26)+γV(s11)
其中，P(s11|s1,a21)为状态数据s1按照通行时间a21通过路段1，转换为状态数据s11的概率，即通行时间a21的概率，R(s11|s1,a21)为状态数据s1按照通行时间a21通过路段1，转换为状态数据s11的条件下，状态数据s11的收益数值，即s11的收益数值。其他路段类似，在此不再赘述。
之后，采用类似的方式也可以计算出状态数据s0的收益数值，进而得到每个状态数据的收益数值。
106、根据每个状态数据以及每个状态数据的收益数值进行训练，得到收益数值预测模型。
通过上述步骤106可以确定状态数据与收益数值的对应关系，将该对应关系作为一组样本数据，则根据每组样本数据进行训练，可以得到收益数值预测模型，该收益数值预测模型用于根据任一状态数据获取该状态数据的收益数值，以收益数值来表示在当前所处状态下未来期望获得的收益。
在一种实现方式中，可以采用深度神经网络算法来训练收益数值预测模型。该收益数值预测模型最初采用随机值预设，通过不断地尝试学习，该收益数值预测模型能够学习到状态数据对应的收益数值的规则，并更新收益数值预测模型 中的模型参数，保持模型随着试错学习而不断进行更新，使得模型的准确度不断提高，进而从一个随机初始化的模型不断优化直至收敛，此时基于收益数值预测模型做出的决策才会更加趋于最优。
本申请实施例提供的方法，获取至少一条样本路线中每个路段的历史通行数据，能够从局部的角度来描述样本路线，根据获取到的历史通行数据进行训练得到针对路段的通行时间选取模型和状态数据预测模型，基于通行时间选取模型和状态数据预测模型，能够以路段为单位来预测通行时间和状态数据，充分考虑了路线的局部信息，提高了预测准确度。
并且，根据获取到的历史通行数据以及训练出的通行时间选取模型和状态数据预测模型，可以训练收益数值预测模型，基于收益数值预测模型能够以路段为单位来预测每个状态数据的收益数值，以收益数值来衡量状态数据的准确程度，从而根据状态数据确定更为合理的通行时间，提高了准确度。
并且，相关技术采用的预测模型对路线的交通状况基本没有预测能力，仅根据出发时间点的交通状况来进行预测，对到达终点的通行时间与实际通行时间会有很大的出入。而本申请实施例通过训练第一速度预测模型和第二速度预测模型中的至少一个，可以预测出路段的实时通行速度，实时通行速度能够表示路段在到达该路段时的交通状况，因此在预测通行时间时能够考虑到实时交通状况的影响，预测更为准确。
图6是本申请实施例提供的一种通行时间确定方法的流程图，该通行时间确定方法的执行主体为预测装置，对预测目标路线的通行时间的过程进行说明。其中，该预测装置可以为具备导航功能的手机、计算机等终端或者服务器等，且预测装置与上述实施例中的训练装置可以为相同装置，或者也可以为不同装置，当预测装置与训练装置为不同装置时，训练装置可以将训练完成的模型提供给预测装置，供预测装置使用。参见图6，该方法包括：
601、获取待通行的目标路线。
其中，目标路线包括按照顺序排列的多个路段，任意两个在位置上相邻的路段相连，上一个路段的终点即为下一个路段的起点。
该目标路线可以由用户选择，或者由预测装置根据起点和终点在电子地图中选择。例如，当用户要出发去往某一目的地时，预测装置可以将用户当前所在的位置作为路线起点，将目的地作为路线终点，将电子地图中从起点到终点的一 条或多条路线均作为目标路线，以预测每条目标路线的总通行时间。
602、从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，将第一路段的第一状态数据输入至通行时间选取模型中，基于通行时间选取模型确定多个通行时间的概率，根据多个通行时间的概率从多个通行时间中选取第一路段的多个备选通行时间。
本申请实施例中，通行时间选取模型用于根据路段的状态数据确定多个通行时间的概率，则将第一路段的第一状态数据输入至通行时间选取模型中，基于通行时间选取模型确定多个通行时间的概率，此时可以根据多个通行时间的概率，从多个通行时间中选取多个备选通行时间。如按照概率从大到小的顺序选取预设数量的通行时间，作为备选通行时间，从而得到多个备选通行时间，该预设数量可以根据准确度需求确定，或者根据多个通行时间的数量和固定的选取比例确定。
例如，将路段i的状态数据s i,t输入至通行时间选取模型中可以得到路段i的备选通行时间a i。
603、对于每个备选通行时间，将第一状态数据和备选通行时间输入至状态数据预测模型中，基于状态数据预测模型确定在第一状态数据下按照备选通行时间通过第一路段后第二路段的备选状态数据。
例如，将路段i的状态数据s i,t和备选通行时间a i输入至状态数据预测模型中，可以得到在时间点t之后按照备选通行时间a i通过路段i，在时间点t+1时到达路段i+1时，路段i+1的备选状态数据s i+1,t+1。
在一种实现方式中，任一路段的状态数据包括该路段的实时数据，该实时数据是指该路段的实时通行速度，即在该路段起点时该路段的实时通行速度。则将在第一路段起点时的时间点该第一路段的实时通行速度以及该第一路段的备选通行时间输入至第一速度预测模型，基于第一速度预测模型确定按照备选通行时间通过第一路段后在该第一路段终点时的时间点第二路段的实时通行速度。
例如，以i表示上一个路段，以i+1表示下一个路段，i为整数，基于第一速度预测模型，可以根据在路段i起点的时间点t时，路段i的实时通行速度V i,t和路段i的通行时间a i，预测出按照该通行时间a i通过路段i后，在时间点t+1时到达路段i+1的起点时，路段i+1的实时通行速度V i+1,t+1。
可选地，为了提高准确度，还可以将在第一路段起点时的时间点该第一路段的实时通行速度和该第二路段的实时通行速度，以及该第一路段的备选通行时 间输入至第一速度预测模型，基于第一速度预测模型确定按照备选通行时间通过第一路段后在该第一路段终点时的时间点第二路段的实时通行速度。
例如，基于第一速度预测模型，可以根据在路段i起点的时间点t时，路段i的实时通行速度V i,t、路段i+1的实时通行速度V i+1,t和路段i的通行时间a i，预测出按照该通行时间a i通过路段i后，在时间点t+1时到达路段i+1的起点时，路段i+1的实时通行速度V i+1,t+1。
在另一种实现方式中，当任一路段的状态数据包括剩余路段的实时数据，该剩余路段的实时数据是指该路段之后的每个路段的实时通行速度，即在该路段的起点时，该路段之后的每个路段的实时通行速度。则将在第一路段起点时的时间点该第一路段之后每个路段的实时通行速度以及该第一路段的备选通行时间输入至第二速度预测模型，基于第二速度预测模型确定按照备选通行时间通过第一路段后在该第一路段终点时的时间点第二路段之后每个路段的实时通行速度。
例如，以i表示上一个路段，以i+1表示下一个路段，i为整数，基于第二速度预测模型，可以根据在路段i起点的时间点t时，路段i之后每个路段的实时通行速度V left,i,t和路段i的通行时间a i，预测出按照该通行时间a i通过路段i后，在时间点t+1时到达路段i+1的起点时，路段i+1之后每个路段的实时通行速度V left,i+1,t+1。
可选地，为了提高准确度，还可以将在第一路段起点时的时间点该第一路段之后每个路段的实时通行速度和该第二路段之后每个路段的实时通行速度，以及该第一路段的备选通行时间输入至第二速度预测模型，基于第二速度预测模型确定按照备选通行时间通过第一路段后在该第一路段终点时的时间点第二路段之后每个路段的实时通行速度。
例如，基于第二速度预测模型，可以根据在路段i起点的时间点t时，路段i之后每个路段的实时通行速度V left,i,t、路段i+1之后每个路段的实时通行速度V left,i+1,t和路段i的通行时间a i，预测出按照该通行时间a i通过路段i后，在时间点t+1时到达路段i+1的起点时，路段i+1之后每个路段的实时通行速度V left,i+1,t+1。
另外，第一状态数据转换到第二状态数据时，对于除上述路段的实时数据和剩余路段的实时数据之外的其他项状态数据，第二状态数据中的初始速度与第一状态数据的初始速度相等，因此初始速度不变；由于路段和时间点均发生变化，因此历史统计速度更换为在该第一路段终点时的时间点第二路段的历史统计速 度，累积通行时间需要在原有累积通行时间的基础上增加该第一路段的通行时间。
例如在时间点t时路段i的状态数据为s i,t＝<V 0，V hts,i,t，T sum,i,t，L i,t，L left,i,t>，当按照通行时间通过路段i到达路段i+1时，在时间点t+1时路段i+1的状态数据为S i+1,t+1＝<V 0，V hts,i+1,t+1，T sum,i+1,t+1，L i+1,t+1，L left,i+1,t+1>。
604、根据多个备选通行时间对应的备选状态数据，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据。
在一种实现方式中，将每个备选通行时间对应的备选状态数据输入至收益数值预测模型中，基于收益数值预测模型，获取每个备选状态数据的收益数值，收益数值用于衡量对应状态数据的预测准确程度，则从多个备选通行时间对应的备选状态数据中，选取收益数值最大的备选状态数据，确定为第二状态数据，将第二状态数据对应的备选通行时间确定为第一路段的通行时间。
需要说明的是，预测装置获取到目标路线后，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据。而上述步骤602-604为可选步骤，预测装置也可以采用其他方式来确定第一路段的通行时间和第二状态数据。
在另一实施例中，该通行时间选取模型用于根据任一路段的状态数据确定该任一路段的通行时间，因此，将第一状态数据输入至通行时间选取模型中，基于通行时间选取模型确定第一路段的通行时间，将第一状态数据和通行时间输入至状态数据预测模型中，基于状态数据预测模型确定第二状态数据。
605、根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间。
获取到第二状态数据之后，可以对第二路段和第二路段的下一个路段，继续执行上述步骤602-604确定第二路段的通行时间，以此类推，即可确定目标路线中每个路段的通行时间。
606、根据多个路段的通行时间，确定目标路线的总通行时间，根据当前时间点和总通行时间，确定到达时间点。
将目标路线中多个路段的通行时间的总和确定为目标路线的总通行时间，将当前时间点作为用户的出发时间点，则出发时间点经过该总通行时间后达到 的时间点即为到达时间点。
本申请实施例提供的方法，通过训练得到针对路段的通行时间选取模型和状态数据预测模型，基于通行时间选取模型和状态数据预测模型，能够以路段为单位来预测通行时间和状态数据，充分考虑了路线的局部信息，提高了预测准确度，能够弥补传统预测模型的缺点。
并且，通过训练得到针对路段的收益数值预测模型，基于收益数值预测模型能够以路段为单位来预测每个状态数据的收益数值，以收益数值来衡量状态数据的准确程度，从而根据状态数据确定更为合理的通行时间，提高了准确度。
并且，相关技术采用的预测模型对路线的交通状况基本没有预测能力，仅根据出发时间点的交通状况来进行预测，对到达终点的通行时间与实际通行时间会有很大的出入。而本申请实施例通过训练第一速度预测模型和第二速度预测模型中的至少一个，可以预测出路段的实时通行速度，实时通行速度能够表示路段在到达该路段时的交通状况，因此在预测通行时间时能够考虑到实时交通状况的影响，预测更为准确。
总结来说，本申请实施例中将通行时间选取模型和状态数据预测模型共同组成了预测方案中的策略网络，通行时间选取模型用于判断在当前状态下，为当前路段预估多少通行时间最为恰当，而状态数据预测模型基于通行时间选取模型的输出，对下一状态数据进行预测，预测下一状态最大可能是什么样的，具体的状态数据是什么。
本申请实施例中将收益数值预测模型组成了估值网络，为每个通行时间达到的状态数据的收益进行估算，得到估值网络与最佳策略之间的关联关系。
通过根据大量的历史通行数据进行训练，使模型学习到基于时间、路段及实时通行速度的变化与道路通行能力变化之间的关联关系，进而基于模型为每个路段的通行时间进行预测时，将学习到的变化考虑进来，综合策略网络和估值网络给出的结果确定预测结果。
本申请实施例通过通行时间选取模型、状态数据预测模型和收益数值预测模型，提供了一种基于强化学习框架的预测方案，通过样本数据训练得到最佳决策过程，最终可以输出预测结果。该预测方案无需进行详细的规则设计，通过路段级别的建模，能够保留路线的局部信息，而且具有一定的推理能力，理论上能够解决对交通状况变化的预测问题。而且利用了强化学习框架可以在线更新的 特性，对样本数据具有很好的敏感性，可以对用户的通行情况、用户分布变化、交通状况等数据做到实时更新。
图7是本申请实施例提供的一种通行时间确定装置的结构示意图。参见图7，该装置包括：
路线获取模块701，用于执行上述实施例中获取目标路线的步骤；
策略确定模块702，用于执行上述实施例中确定第一路段的通行时间，以及第二状态数据的步骤；
策略确定模块702，还用于执行上述实施例中根据第二状态数据，继续确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间的步骤。
可选地，策略确定模块702，包括：
时间确定单元，用于执行上述实施例中基于通行时间选取模型确定第一路段的通行时间的步骤；
状态确定单元，用于执行上述实施例中基于状态数据预测模型确定第二状态数据的步骤。
可选地，策略确定模块702，包括：
备选时间确定单元，用于执行上述实施例中基于通行时间选取模型确定多个通行时间的概率，选取第一路段的多个备选通行时间的步骤；
备选状态确定单元，用于执行上述实施例中对于每个备选通行时间，基于状态数据预测模型确定在第一状态数据下按照备选通行时间通过第一路段后第二路段的备选状态数据的步骤；
策略确定单元，用于执行上述实施例中根据多个备选通行时间对应的备选状态数据，确定第一路段的通行时间，以及通行时间对应的第二状态数据的步骤。
可选地，策略确定单元，还用于执行上述实施例中基于收益数值预测模型获取备选状态数据的收益数值，选取收益数值最大的备选状态数据，确定为第二状态数据，并确定第一路段的通行时间的步骤。
可选地，策略确定单元，还用于执行上述实施例中获取第一路段在第一时间点的实时通行速度，基于第一速度预测模拟性确定第二路段在第二时间点的实时通行速度的步骤。
可选地，策略确定单元，还用于执行上述实施例中获取第一路段之后的每个路段在第一时间点的实时通行速度，基于第二速度预测模型确定第二路段之后 的每个路段在第二时间点的实时通行速度的步骤。
可选地，装置还包括：
样本获取模块，用于执行上述实施例中获取至少一条样本路线中每个路段的历史通行数据的步骤；
第一训练模块，用于执行上述实施例中根据获取到的历史通行数据，构造多组第一样本数据，进行训练得到通行时间选取模型的步骤；
第二训练模块，用于执行上述实施例中根据获取到的历史通行数据，构造多组第二样本数据，进行训练得到状态数据预测模型的步骤。
可选地，装置还包括：
激励获取模块，用于执行上述实施例中对于每条样本路线，获取样本路线的全局激励数值和样本路线中每个路段的局部激励数值的步骤；
收益获取模块，用于执行上述实施例中对于至少一条样本路线中的任一路段的第一样本状态数据，获取第一样本状态数据的收益数值，直至获取到至少一条样本路线中每个状态数据的收益数值的步骤；
第三训练模块，用于执行上述实施例中根据每个状态数据以及每个状态数据的收益数值进行训练，得到收益数值预测模型的步骤。
可选地，激励获取模块，包括：
时间预测单元，用于执行上述实施例中基于当前训练的通行时间选取模型和状态数据预测模型，确定样本路线中每个路段的预测通行时间的步骤；
全局激励获取单元，用于执行上述实施例中根据每个路段的预测通行时间，获取样本路线的预测总通行时间，根据样本路线的预测总通行时间与样本路线的实际总通行时间之间的第一误差，确定样本路线的全局激励数值的步骤；
局部激励获取单元，用于执行上述实施例中根据每个路段的预测通行时间与每个路段的实际通行时间之间的第二误差，确定每个路段的局部激励数值的步骤。
可选地，收益获取模块，用于执行上述实施例中采用以下公式获取第一样本状态数据的收益数值：
Figure PCTCN2019091311-appb-000007
需要说明的是：上述实施例提供的通行时间确定装置在确定通行时间时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成，即将训练装置或预测装置的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。另外，上述实施例提供的 通行时间确定装置与通行时间确定方法实施例属于同一构思，其具体实现过程详见方法实施例，这里不再赘述。
图8示出了本申请一个示例性实施例提供的终端800的结构框图。该终端800可以是便携式移动终端，比如：智能手机、平板电脑、MP3播放器(Moving Picture Experts Group Audio Layer III，动态影像专家压缩标准音频层面3)、MP4(Moving Picture Experts Group Audio Layer IV，动态影像专家压缩标准音频层面4)播放器、笔记本电脑、台式电脑、头戴式设备，或其他任意智能终端。终端800还可能被称为用户设备、便携式终端、膝上型终端、台式终端等其他名称。
通常，终端800包括有：处理器801和存储器802。
处理器801可以包括一个或多个处理核心，比如4核心处理器、8核心处理器等。处理器801可以采用DSP(Digital Signal Processing，数字信号处理)、FPGA(Field－Programmable Gate Array，现场可编程门阵列)、PLA(Programmable Logic Array，可编程逻辑阵列)中的至少一种硬件形式来实现。处理器801也可以包括主处理器和协处理器，主处理器是用于对在唤醒状态下的数据进行处理的处理器，也称CPU(Central Processing Unit，中央处理器)；协处理器是用于对在待机状态下的数据进行处理的低功耗处理器。在一些实施例中，处理器801可以在集成有GPU(Graphics Processing Unit，图像处理器)，GPU用于负责显示屏所需要显示的内容的渲染和绘制。一些实施例中，处理器801还可以包括AI(Artificial Intelligence，人工智能)处理器，该AI处理器用于处理有关机器学习的计算操作。
存储器802可以包括一个或多个计算机可读存储介质，该计算机可读存储介质可以是非暂态的。存储器802还可包括高速随机存取存储器，以及非易失性存储器，比如一个或多个磁盘存储设备、闪存存储设备。在一些实施例中，存储器802中的非暂态的计算机可读存储介质用于存储至少一个指令，该至少一个指令用于被处理器801所具有以实现本申请中方法实施例提供的通行时间确定方法。
在一些实施例中，终端800还可选包括有：外围设备接口803和至少一个外围设备。处理器801、存储器802和外围设备接口803之间可以通过总线或信号线相连。各个外围设备可以通过总线、信号线或电路板与外围设备接口803相 连。具体地，外围设备包括：射频电路804、触摸显示屏805、摄像头806、音频电路807、定位组件808和电源809中的至少一种。
外围设备接口803可被用于将I/O(Input/Output，输入/输出)相关的至少一个外围设备连接到处理器801和存储器802。在一些实施例中，处理器801、存储器802和外围设备接口803被集成在同一芯片或电路板上；在一些其他实施例中，处理器801、存储器802和外围设备接口803中的任意一个或两个可以在单独的芯片或电路板上实现，本实施例对此不加以限定。
射频电路804用于接收和发射RF(Radio Frequency，射频)信号，也称电磁信号。射频电路804通过电磁信号与通信网络以及其他通信设备进行通信。射频电路804将电信号转换为电磁信号进行发送，或者，将接收到的电磁信号转换为电信号。可选地，射频电路804包括：天线系统、RF收发器、一个或多个放大器、调谐器、振荡器、数字信号处理器、编解码芯片组、用户身份模块卡等等。射频电路804可以通过至少一种无线通信协议来与其它终端进行通信。该无线通信协议包括但不限于：城域网、各代移动通信网络(2G、3G、4G及8G)、无线局域网和/或WiFi(Wireless Fidelity，无线保真)网络。在一些实施例中，射频电路804还可以包括NFC(Near Field Communication，近距离无线通信)有关的电路，本申请对此不加以限定。
显示屏805用于显示UI(User Interface，用户界面)。该UI可以包括图形、文本、图标、视频及其它们的任意组合。当显示屏805是触摸显示屏时，显示屏805还具有采集在显示屏805的表面或表面上方的触摸信号的能力。该触摸信号可以作为控制信号输入至处理器801进行处理。此时，显示屏805还可以用于提供虚拟按钮和/或虚拟键盘，也称软按钮和/或软键盘。在一些实施例中，显示屏805可以为一个，设置终端800的前面板；在另一些实施例中，显示屏805可以为至少两个，分别设置在终端800的不同表面或呈折叠设计；在再一些实施例中，显示屏805可以是柔性显示屏，设置在终端800的弯曲表面上或折叠面上。甚至，显示屏805还可以设置成非矩形的不规则图形，也即异形屏。显示屏805可以采用LCD(Liquid Crystal Display，液晶显示屏)、OLED(Organic Light-Emitting Diode,有机发光二极管)等材质制备。
摄像头组件808用于采集图像或视频。可选地，摄像头组件808包括前置摄像头和后置摄像头。通常，前置摄像头设置在终端的前面板，后置摄像头设置在终端的背面。在一些实施例中，后置摄像头为至少两个，分别为主摄像头、景 深摄像头、广角摄像头、长焦摄像头中的任意一种，以实现主摄像头和景深摄像头融合实现背景虚化功能、主摄像头和广角摄像头融合实现全景拍摄以及VR(Virtual Reality，虚拟现实)拍摄功能或者其它融合拍摄功能。在一些实施例中，摄像头组件808还可以包括闪光灯。闪光灯可以是单色温闪光灯，也可以是双色温闪光灯。双色温闪光灯是指暖光闪光灯和冷光闪光灯的组合，可以用于不同色温下的光线补偿。
音频电路807可以包括麦克风和扬声器。麦克风用于采集用户及环境的声波，并将声波转换为电信号输入至处理器801进行处理，或者输入至射频电路804以实现语音通信。出于立体声采集或降噪的目的，麦克风可以为多个，分别设置在终端800的不同部位。麦克风还可以是阵列麦克风或全向采集型麦克风。扬声器则用于将来自处理器801或射频电路804的电信号转换为声波。扬声器可以是传统的薄膜扬声器，也可以是压电陶瓷扬声器。当扬声器是压电陶瓷扬声器时，不仅可以将电信号转换为人类可听见的声波，也可以将电信号转换为人类听不见的声波以进行测距等用途。在一些实施例中，音频电路807还可以包括耳机插孔。
定位组件808用于定位终端800的当前地理位置，以实现导航或LBS(Location Based Service，基于位置的服务)。定位组件808可以是基于美国的GPS(Global Positioning System，全球定位系统)、中国的北斗系统、俄罗斯的格雷纳斯系统或欧盟的伽利略系统的定位组件。
电源809用于为终端800中的各个组件进行供电。电源809可以是交流电、直流电、一次性电池或可充电电池。当电源809包括可充电电池时，该可充电电池可以支持有线充电或无线充电。该可充电电池还可以用于支持快充技术。
在一些实施例中，终端800还包括有一个或多个传感器810。该一个或多个传感器810包括但不限于：加速度传感器811、陀螺仪传感器812、压力传感器813、指纹传感器814、光学传感器815以及接近传感器816。
加速度传感器811可以检测以终端800建立的坐标系的三个坐标轴上的加速度大小。比如，加速度传感器811可以用于检测重力加速度在三个坐标轴上的分量。处理器801可以根据加速度传感器811采集的重力加速度信号，控制触摸显示屏805以横向视图或纵向视图进行用户界面的显示。加速度传感器811还可以用于游戏或者用户的运动数据的采集。
陀螺仪传感器812可以检测终端800的机体方向及转动角度，陀螺仪传感 器812可以与加速度传感器811协同采集用户对终端800的3D动作。处理器801根据陀螺仪传感器812采集的数据，可以实现如下功能：动作感应(比如根据用户的倾斜操作来改变UI)、拍摄时的图像稳定、游戏控制以及惯性导航。
压力传感器813可以设置在终端800的侧边框和/或触摸显示屏805的下层。当压力传感器813设置在终端800的侧边框时，可以检测用户对终端800的握持信号，由处理器801根据压力传感器813采集的握持信号进行左右手识别或快捷操作。当压力传感器813设置在触摸显示屏805的下层时，由处理器801根据用户对触摸显示屏805的压力操作，实现对UI界面上的可操作性控件进行控制。可操作性控件包括按钮控件、滚动条控件、图标控件、菜单控件中的至少一种。
指纹传感器814用于采集用户的指纹，由处理器801根据指纹传感器814采集到的指纹识别用户的身份，或者，由指纹传感器814根据采集到的指纹识别用户的身份。在识别出用户的身份为可信身份时，由处理器801授权该用户具有相关的敏感操作，该敏感操作包括解锁屏幕、查看加密信息、下载软件、支付及更改设置等。指纹传感器814可以被设置终端800的正面、背面或侧面。当终端800上设置有物理按键或厂商Logo时，指纹传感器814可以与物理按键或厂商标志集成在一起。
光学传感器815用于采集环境光强度。在一个实施例中，处理器801可以根据光学传感器815采集的环境光强度，控制触摸显示屏805的显示亮度。具体地，当环境光强度较高时，调高触摸显示屏805的显示亮度；当环境光强度较低时，调低触摸显示屏805的显示亮度。在另一个实施例中，处理器801还可以根据光学传感器815采集的环境光强度，动态调整摄像头组件808的拍摄参数。
接近传感器816，也称距离传感器，通常设置在终端800的前面板。接近传感器816用于采集用户与终端800的正面之间的距离。在一个实施例中，当接近传感器816检测到用户与终端800的正面之间的距离逐渐变小时，由处理器801控制触摸显示屏805从亮屏状态切换为息屏状态；当接近传感器816检测到用户与终端800的正面之间的距离逐渐变大时，由处理器801控制触摸显示屏805从息屏状态切换为亮屏状态。
本领域技术人员可以理解，图8中示出的结构并不构成对终端800的限定，可以包括比图示更多或更少的组件，或者组合某些组件，或者采用不同的组件布置。
图9是本申请实施例提供的一种服务器的结构示意图，该服务器900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处理器(central processing units，CPU)901和一个或一个以上的存储器902，其中，所述存储器902中存储有至少一条指令，所述至少一条指令由所述处理器901加载并执行以实现上述各个方法实施例提供的方法。当然，该服务器还可以具有有线或无线网络接口、键盘以及输入输出接口等部件，以便进行输入输出，该服务器还可以包括其他用于实现设备功能的部件，在此不做赘述。
服务器900可以用于执行上述通行时间确定中预测装置所执行的步骤。
图10示出了一个实施例中计算机设备的内部结构图。该计算机设备具体可以是图1A中的终端110或服务器120。如图10所示，该计算机设备包括该计算机设备包括通过系统总线连接的处理器、存储器、网络接口、输入装置和显示屏。其中，存储器包括非易失性存储介质和内存储器。该计算机设备的非易失性存储介质存储有操作系统，还可存储有计算机可读指令，该计算机可读指令被处理器执行时，可使得处理器实现通行时间确定方法。该内存储器中也可储存有计算机可读指令，该计算机可读指令被处理器执行时，可使得处理器执行通行时间确定方法。计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图10中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，本申请提供的通行时间确定装置可以实现为一种计算机可读指令的形式，计算机可读指令可在如图10所示的计算机设备上运行。计算机设备的存储器中可存储组成该通行时间确定装置的各个程序模块，比如，图7所示的路线获取模块和策略确定模块。各个程序模块构成的计算机可读指令使得处理器执行本说明书中描述的本申请各个实施例的通行时间确定方法中的步骤。
例如，图10所示的计算机设备可以通过如图7所示的通行时间确定装置中 的路线获取模块执行获取待通行的目标路线，目标路线包括按照顺序排列的多个路段。计算机设备可通过策略确定模块执行从多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定第一路段的通行时间，以及在第一状态数据下按照通行时间通过第一路段后第二路段的第二状态数据。计算机设备还可通过策略确定模块还执行根据第二状态数据，继续基于通行时间选取模型和状态数据预测模型确定第二路段的通行时间，直至确定多个路段中每个路段的通行时间。
在一个实施例中，提供了一种计算机设备，包括存储器和处理器，存储器存储有计算机可读指令，计算机可读指令被处理器执行时，使得处理器执行上述通行时间确定方法的步骤。此处通行时间确定方法的步骤可以是上述各个实施例的通行时间确定方法中的步骤。
在一个实施例中，提供了一种计算机可读存储介质，存储有计算机可读指令，计算机可读指令被处理器执行时，使得处理器执行上述通行时间确定方法的步骤。此处通行时间确定方法的步骤可以是上述各个实施例的通行时间确定方法中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的程序可存储于一非易失性计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对本申请专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种通行时间确定方法，其特征在于，所述方法包括：
    计算机设备获取待通行的目标路线，所述目标路线包括按照顺序排列的多个路段；
    所述计算机设备从所述多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据；
    所述计算机设备根据所述第二状态数据，继续基于所述通行时间选取模型和所述状态数据预测模型确定所述第二路段的通行时间，直至确定所述多个路段中每个路段的通行时间；
    其中，所述通行时间选取模型用于根据任一路段的状态数据确定所述任一路段的通行时间，所述状态数据预测模型用于根据所述任一路段的状态数据和通行时间，确定所述任一路段的下一路段的状态数据。
  2. 根据权利要求1所述的方法，其特征在于，所述计算机设备根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    所述计算机设备将所述第一状态数据输入至所述通行时间选取模型中，基于所述通行时间选取模型确定所述第一路段的通行时间；
    所述计算机设备将所述第一状态数据和所述通行时间输入至所述状态数据预测模型中，基于所述状态数据预测模型确定所述第二状态数据。
  3. 根据权利要求1所述的方法，其特征在于，所述计算机设备根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    所述计算机设备将所述第一状态数据输入至所述通行时间选取模型中，基 于所述通行时间选取模型确定多个通行时间的概率，根据所述多个通行时间的概率从所述多个通行时间中选取所述第一路段的多个备选通行时间；
    所述计算机设备对于每个备选通行时间，将所述第一状态数据和所述备选通行时间输入至所述状态数据预测模型中，基于所述状态数据预测模型确定在所述第一状态数据下按照所述备选通行时间通过所述第一路段后所述第二路段的备选状态数据；
    所述计算机设备根据所述多个备选通行时间对应的备选状态数据，确定所述第一路段的通行时间，以及所述通行时间对应的所述第二状态数据。
  4. 根据权利要求3所述的方法，其特征在于，所述计算机设备根据所述多个备选通行时间对应的备选状态数据，确定所述第一路段的通行时间，以及所述通行时间对应的所述第二状态数据，包括：
    所述计算机设备将每个备选通行时间对应的备选状态数据输入至收益数值预测模型中，基于所述收益数值预测模型，获取每个备选状态数据的收益数值，所述收益数值用于衡量对应状态数据的预测准确程度；
    所述计算机设备从所述多个备选通行时间对应的备选状态数据中，选取收益数值最大的备选状态数据，确定为所述第二状态数据；
    所述计算机设备将所述第二状态数据对应的备选通行时间确定为所述第一路段的通行时间。
  5. 根据权利要求1-4任一项所述的方法，其特征在于，所述任一路段的状态数据包括所述任一路段的实时通行速度，所述状态数据预测模型包括第一速度预测模型；
    所述计算机设备根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    所述计算机设备根据所述第一状态数据，基于所述通行时间选取模型和所述状态数据预测模型，确定所述第一路段的通行时间；
    所述计算机设备根据当前时间点以及所述第一路段之前的每个路段的通行 时间，确定到达所述第一路段起点时的第一时间点，获取所述第一路段在所述第一时间点的实时通行速度；
    所述计算机设备将所述第一路段在所述第一时间点的实时通行速度以及所述第一路段的通行时间输入至所述第一速度预测模型中，基于所述第一速度预测模型确定所述第二路段在第二时间点的实时通行速度，所述第二时间点根据所述第一时间点和所述通行时间确定。
  6. 根据权利要求1-4任一项所述的方法，其特征在于，所述任一路段的状态数据包括所述任一路段之后的每个路段的实时通行速度，所述状态数据预测模型包括第二速度预测模型；
    所述计算机设备根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    所述计算机设备根据所述第一状态数据，基于所述通行时间选取模型和所述状态数据预测模型，确定所述第一路段的通行时间；
    所述计算机设备根据当前时间点以及所述第一路段之前的每个路段的通行时间，确定到达所述第一路段起点时的第一时间点，获取所述第一路段之后的每个路段在所述第一时间点的实时通行速度；
    所述计算机设备将所述第一路段之后的每个路段在所述第一时间点的实时通行速度以及所述第一路段的通行时间输入至所述第二速度预测模型中，基于所述第二速度预测模型确定所述第二路段之后的每个路段在第二时间点的实时通行速度，所述第二时间点根据所述第一时间点和所述通行时间确定。
  7. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    所述计算机设备获取至少一条样本路线中每个路段的历史通行数据，所述路段的历史通行数据包括所述路段的通行时间和所述路段的状态数据；
    所述计算机设备根据获取到的历史通行数据，构造多组第一样本数据，每组第一样本数据中包括一组状态数据以及与所述一组状态数据对应的路段的通行时间；
    所述计算机设备根据所述多组第一样本数据进行训练，得到所述通行时间选取模型；
    所述计算机设备根据获取到的历史通行数据，构造多组第二样本数据，每组第二样本数据中包括一组状态数据、与所述一组状态数据对应的路段的通行时间以及所述一组状态数据的下一组状态数据；根据所述多组第二样本数据进行训练，得到所述状态数据预测模型。
  8. 根据权利要求7所述的方法，其特征在于，所述方法还包括：
    所述计算机设备对于每条样本路线，获取所述样本路线的全局激励数值和所述样本路线中每个路段的局部激励数值，所述全局激励数值用于衡量所述样本路线的通行时间的准确程度，所述局部激励数值用于衡量对应路段的通行时间的准确程度；
    所述计算机设备对于所述至少一条样本路线中任一路段的第一样本状态数据，根据所述第一样本状态数据在每条样本路线中的下一样本状态数据的收益数值、从所述第一样本状态数据转换到所述下一样本状态数据的概率和从所述第一样本状态数据转换到所述下一样本状态数据的条件下所述下一样本状态数据的收益数值，获取所述第一样本状态数据的收益数值，直至获取到所述至少一条样本路线中每个状态数据的收益数值；
    其中，任一样本路线中最后一个状态数据的收益数值等于所述样本路线的全局激励数值与所述样本路线中每个路段的局部激励数值的总和；
    所述计算机设备根据所述每个状态数据以及所述每个状态数据的收益数值进行训练，得到收益数值预测模型，所述收益数值预测模型用于根据任一状态数据获取所述任一状态数据的收益数值。
  9. 根据权利要求8所述的方法，其特征在于，所述计算机设备获取所述样本路线的全局激励数值和所述样本路线中每个路段的局部激励数值，包括：
    所述计算机设备基于当前训练的所述通行时间选取模型和所述状态数据预测模型，确定所述样本路线中每个路段的预测通行时间；
    所述计算机设备根据所述每个路段的预测通行时间，获取所述样本路线的预测总通行时间，根据所述样本路线的预测总通行时间与所述样本路线的实际 总通行时间之间的第一误差，确定所述样本路线的全局激励数值，所述全局激励数值与所述第一误差呈反比关系；
    所述计算机设备根据所述每个路段的预测通行时间与所述每个路段的实际通行时间之间的第二误差，确定所述每个路段的局部激励数值，所述局部激励数值与所述第二误差呈反比关系。
  10. 根据权利要求8所述的方法，其特征在于，所述计算机设备根据所述第一样本状态数据在每条样本路线中的下一样本状态数据的收益数值、从所述第一样本状态数据转换到所述下一样本状态数据的概率和从所述第一样本状态数据转换到所述下一样本状态数据的条件下所述下一样本状态数据的收益数值，获取所述第一样本状态数据的收益数值，包括：
    所述计算机设备采用以下公式，获取所述第一样本状态数据的收益数值：
    Figure PCTCN2019091311-appb-100001
    其中，s i表示所述第一样本状态数据，V π(s i)表示所述第一样本状态数据的收益数值，s i+1表示所述下一样本状态数据，V π(s i+1)表示所述下一样本状态数据的收益数值，S表示所述第一样本状态数据在所述至少一条样本路线中所有的下一样本状态数据构成的集合，π表示由多个路段的通行时间构成的通行策略；；
    P(s i+1|s i,a i)表示从所述第一样本状态数据转换到所述下一样本状态数据的概率，且等于从所述第一样本状态数据转换到所述下一样本状态数据时按照的通行时间的概率，R(s i+1|s i,a i)表示从所述第一样本状态数据转换到所述下一样本状态数据的条件下所述下一样本状态数据的收益数值，且等于从所述第一样本状态数据转换到所述下一样本状态数据时所通过路段的局部激励数值，γ表示折扣因子。
  11. 一种通行时间确定装置，其特征在于，所述装置包括：
    路线获取模块，用于获取待通行的目标路线，所述目标路线包括按照顺序排列的多个路段；
    策略确定模块，用于从所述多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述 第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据；
    所述策略确定模块，还用于根据所述第二状态数据，继续基于所述通行时间选取模型和所述状态数据预测模型确定所述第二路段的通行时间，直至确定所述多个路段中每个路段的通行时间；
    其中，所述通行时间选取模型用于根据任一路段的状态数据确定所述任一路段的通行时间，所述状态数据预测模型用于根据所述任一路段的状态数据和通行时间，确定所述任一路段的下一路段的状态数据。
  12. 一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行如下步骤：
    获取待通行的目标路线，所述目标路线包括按照顺序排列的多个路段；
    从所述多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据；
    根据所述第二状态数据，继续基于所述通行时间选取模型和所述状态数据预测模型确定所述第二路段的通行时间，直至确定所述多个路段中每个路段的通行时间；
    其中，所述通行时间选取模型用于根据任一路段的状态数据确定所述任一路段的通行时间，所述状态数据预测模型用于根据所述任一路段的状态数据和通行时间，确定所述任一路段的下一路段的状态数据。
  13. 根据权利要求12所述的计算机设备，其特征在于，所述根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    将所述第一状态数据输入至所述通行时间选取模型中，基于所述通行时间选取模型确定所述第一路段的通行时间；
    将所述第一状态数据和所述通行时间输入至所述状态数据预测模型中，基于所述状态数据预测模型确定所述第二状态数据。
  14. 根据权利要求12所述的计算机设备，其特征在于，所述根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    将所述第一状态数据输入至所述通行时间选取模型中，基于所述通行时间选取模型确定多个通行时间的概率，根据所述多个通行时间的概率从所述多个通行时间中选取所述第一路段的多个备选通行时间；
    对于每个备选通行时间，将所述第一状态数据和所述备选通行时间输入至所述状态数据预测模型中，基于所述状态数据预测模型确定在所述第一状态数据下按照所述备选通行时间通过所述第一路段后所述第二路段的备选状态数据；
    根据所述多个备选通行时间对应的备选状态数据，确定所述第一路段的通行时间，以及所述通行时间对应的所述第二状态数据。
  15. 根据权利要求12-14任一项所述的计算机设备，其特征在于，所述任一路段的状态数据包括所述任一路段的实时通行速度，所述状态数据预测模型包括第一速度预测模型；
    所述根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    根据所述第一状态数据，基于所述通行时间选取模型和所述状态数据预测模型，确定所述第一路段的通行时间；
    根据当前时间点以及所述第一路段之前的每个路段的通行时间，确定到达所述第一路段起点时的第一时间点，获取所述第一路段在所述第一时间点的实时通行速度；
    将所述第一路段在所述第一时间点的实时通行速度以及所述第一路段的通行时间输入至所述第一速度预测模型中，基于所述第一速度预测模型确定所述第二路段在第二时间点的实时通行速度，所述第二时间点根据所述第一时间点 和所述通行时间确定。
  16. 一个或多个存储有计算机可读指令的非易失性存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行如下步骤：
    获取待通行的目标路线，所述目标路线包括按照顺序排列的多个路段；
    从所述多个路段中的第一个路段开始，对于任意两个位置上相邻的第一路段和第二路段，根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据；
    根据所述第二状态数据，继续基于所述通行时间选取模型和所述状态数据预测模型确定所述第二路段的通行时间，直至确定所述多个路段中每个路段的通行时间；
    其中，所述通行时间选取模型用于根据任一路段的状态数据确定所述任一路段的通行时间，所述状态数据预测模型用于根据所述任一路段的状态数据和通行时间，确定所述任一路段的下一路段的状态数据。
  17. 根据权利要求16所述的存储介质，其特征在于，所述根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    将所述第一状态数据输入至所述通行时间选取模型中，基于所述通行时间选取模型确定所述第一路段的通行时间；
    将所述第一状态数据和所述通行时间输入至所述状态数据预测模型中，基于所述状态数据预测模型确定所述第二状态数据。
  18. 根据权利要求16所述的存储介质，其特征在于，所述根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    将所述第一状态数据输入至所述通行时间选取模型中，基于所述通行时间 选取模型确定多个通行时间的概率，根据所述多个通行时间的概率从所述多个通行时间中选取所述第一路段的多个备选通行时间；
    对于每个备选通行时间，将所述第一状态数据和所述备选通行时间输入至所述状态数据预测模型中，基于所述状态数据预测模型确定在所述第一状态数据下按照所述备选通行时间通过所述第一路段后所述第二路段的备选状态数据；
    根据所述多个备选通行时间对应的备选状态数据，确定所述第一路段的通行时间，以及所述通行时间对应的所述第二状态数据。
  19. 根据权利要求18所述的存储介质，其特征在于，所述根据所述多个备选通行时间对应的备选状态数据，确定所述第一路段的通行时间，以及所述通行时间对应的所述第二状态数据，包括：
    将每个备选通行时间对应的备选状态数据输入至收益数值预测模型中，基于所述收益数值预测模型，获取每个备选状态数据的收益数值，所述收益数值用于衡量对应状态数据的预测准确程度；
    从所述多个备选通行时间对应的备选状态数据中，选取收益数值最大的备选状态数据，确定为所述第二状态数据；
    将所述第二状态数据对应的备选通行时间确定为所述第一路段的通行时间。
  20. 根据权利要求16-19任一项所述的存储介质，其特征在于，所述任一路段的状态数据包括所述任一路段的实时通行速度，所述状态数据预测模型包括第一速度预测模型；
    所述根据所述第一路段的第一状态数据，基于通行时间选取模型和状态数据预测模型，确定所述第一路段的通行时间，以及在所述第一状态数据下按照所述通行时间通过所述第一路段后所述第二路段的第二状态数据，包括：
    根据所述第一状态数据，基于所述通行时间选取模型和所述状态数据预测模型，确定所述第一路段的通行时间；
    根据当前时间点以及所述第一路段之前的每个路段的通行时间，确定到达所述第一路段起点时的第一时间点，获取所述第一路段在所述第一时间点的实时通行速度；
    将所述第一路段在所述第一时间点的实时通行速度以及所述第一路段的通 行时间输入至所述第一速度预测模型中，基于所述第一速度预测模型确定所述第二路段在第二时间点的实时通行速度，所述第二时间点根据所述第一时间点和所述通行时间确定。
PCT/CN2019/091311 2018-07-23 2019-06-14 通行时间确定方法、装置、计算机设备及存储介质 WO2020019901A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/004,442 US11846519B2 (en) 2018-07-23 2020-08-27 Travel time determining method and apparatus, computer device, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810814290.1A CN110163405B (zh) 2018-07-23 2018-07-23 通行时间确定方法、装置、终端及存储介质
CN201810814290.1 2018-07-23

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/004,442 Continuation US11846519B2 (en) 2018-07-23 2020-08-27 Travel time determining method and apparatus, computer device, and storage medium

Publications (1)

Publication Number Publication Date
WO2020019901A1 true WO2020019901A1 (zh) 2020-01-30

Family

ID=67645123

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/091311 WO2020019901A1 (zh) 2018-07-23 2019-06-14 通行时间确定方法、装置、计算机设备及存储介质

Country Status (3)

Country Link
US (1) US11846519B2 (zh)
CN (1) CN110163405B (zh)
WO (1) WO2020019901A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111292549A (zh) * 2020-02-03 2020-06-16 百度在线网络技术（北京）有限公司 建立路线耗时预估模型的方法、预估路线耗时的方法及对应装置
US11719550B2 (en) 2020-02-03 2023-08-08 Baidu Online Network Teghnology (Beiiing) Co., Ltd. Method and apparatus for building route time consumption estimation model, and method and apparatus for estimating route time consumption

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112991798B (zh) * 2019-12-12 2022-05-20 沈阳美行科技股份有限公司 基于通行速度数据的道路线段行驶时间确定方法及装置
KR20210094915A (ko) * 2020-01-22 2021-07-30 삼성전자주식회사 스토리지 컨트롤러, 이를 포함하는 스토리지 장치 및 스토리지 컨트롤러의동작 방법
CN111582543A (zh) * 2020-04-01 2020-08-25 北京百度网讯科技有限公司 预测模型的生成方法、估计到达时间确定方法及装置
CN111489553B (zh) * 2020-04-26 2022-02-25 百度在线网络技术（北京）有限公司 一种路线规划方法、装置、设备和计算机存储介质
CN111833600B (zh) * 2020-06-10 2022-07-08 北京嘀嘀无限科技发展有限公司 通行时间预测方法、装置及数据处理设备
US20220034667A1 (en) * 2020-07-30 2022-02-03 Here Global B.V. Method, apparatus, and computer program product for estimating a time-of-arrival at a destination
US11933615B2 (en) * 2021-01-26 2024-03-19 Ubkang (Qingdao) Technology Co., Ltd. Estimated time of arrival calculating method and system and mobile machine using the same
CN113902427B (zh) * 2021-12-09 2022-10-04 腾讯科技（深圳）有限公司 一种预估到达时间的确定方法、装置、设备及存储介质
CN114822042B (zh) * 2022-06-28 2022-09-02 深圳市华耀商品检验有限公司 一种用于车载终端检测的信息安全测试管理系统及方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120089322A1 (en) * 2006-06-30 2012-04-12 Microsoft Corporation Computation of travel routes, durations, and plans over multiple contexts
CN103632542A (zh) * 2012-08-27 2014-03-12 国际商业机器公司 交通信息处理方法、装置和相应设备
CN103745611A (zh) * 2013-12-23 2014-04-23 北京掌城科技有限公司 一种交通路径规划方法、装置及系统
CN104613974A (zh) * 2014-05-08 2015-05-13 腾讯科技（深圳）有限公司 导航时间获取方法、系统、导航方法和装置
CN105894121A (zh) * 2016-04-18 2016-08-24 腾讯科技（深圳）有限公司 一种线路规划方法及系统
CN106940829A (zh) * 2017-04-28 2017-07-11 兰州交通大学 一种车联网环境下的个性化路径推荐方法

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7355528B2 (en) * 2003-10-16 2008-04-08 Hitachi, Ltd. Traffic information providing system and car navigation system
CA2602879A1 (en) * 2006-09-14 2008-03-14 University Of South Florida System and method for real-time travel path prediction and automatic incident alerts
US8478642B2 (en) * 2008-10-20 2013-07-02 Carnegie Mellon University System, method and device for predicting navigational decision-making behavior
US9109895B2 (en) * 2009-04-30 2015-08-18 The Boeing Company Estimating probabilities of arrival times for voyages
US8738289B2 (en) * 2011-01-04 2014-05-27 International Business Machines Corporation Advanced routing of vehicle fleets
CN102110362B (zh) * 2011-02-01 2013-05-29 北京汉风和科技发展有限公司 一种规划出行路线的处理方法及系统
US9188985B1 (en) * 2012-09-28 2015-11-17 Google Inc. Suggesting a route based on desired amount of driver interaction
US20160334235A1 (en) * 2013-03-19 2016-11-17 The Florida International University Board Of Trustees Itpa informed traveler program and application
DE102013005362A1 (de) * 2013-03-28 2013-10-10 Daimler Ag Verfahren zur Analyse einer Verkehrssituation
US9200910B2 (en) * 2013-12-11 2015-12-01 Here Global B.V. Ranking of path segments based on incident probability
US20150345967A1 (en) * 2014-06-03 2015-12-03 Nissan North America, Inc. Probabilistic autonomous vehicle routing and navigation
CN105674994B (zh) * 2014-11-17 2020-04-24 深圳市腾讯计算机系统有限公司 获取行车路线的方法、装置及导航设备
US9792821B1 (en) * 2016-03-25 2017-10-17 Toyota Jidosha Kabushiki Kaisha Understanding road scene situation and semantic representation of road scene situation for reliable sharing
WO2018087550A2 (en) * 2016-11-09 2018-05-17 Inventive Cogs (Campbell) Limited Vehicle route guidance
US11747162B2 (en) * 2021-09-01 2023-09-05 Delphi Technologies Ip Limited Systems and methods for vehicle cruise speed recommendation
US20230137058A1 (en) * 2021-11-02 2023-05-04 Tusimple, Inc. Optimized routing application for providing service to an autonomous vehicle

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120089322A1 (en) * 2006-06-30 2012-04-12 Microsoft Corporation Computation of travel routes, durations, and plans over multiple contexts
CN103632542A (zh) * 2012-08-27 2014-03-12 国际商业机器公司 交通信息处理方法、装置和相应设备
CN103745611A (zh) * 2013-12-23 2014-04-23 北京掌城科技有限公司 一种交通路径规划方法、装置及系统
CN104613974A (zh) * 2014-05-08 2015-05-13 腾讯科技（深圳）有限公司 导航时间获取方法、系统、导航方法和装置
CN105894121A (zh) * 2016-04-18 2016-08-24 腾讯科技（深圳）有限公司 一种线路规划方法及系统
CN106940829A (zh) * 2017-04-28 2017-07-11 兰州交通大学 一种车联网环境下的个性化路径推荐方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111292549A (zh) * 2020-02-03 2020-06-16 百度在线网络技术（北京）有限公司 建立路线耗时预估模型的方法、预估路线耗时的方法及对应装置
US11719550B2 (en) 2020-02-03 2023-08-08 Baidu Online Network Teghnology (Beiiing) Co., Ltd. Method and apparatus for building route time consumption estimation model, and method and apparatus for estimating route time consumption

Also Published As

Publication number Publication date
US20200393262A1 (en) 2020-12-17
CN110163405A (zh) 2019-08-23
US11846519B2 (en) 2023-12-19
CN110163405B (zh) 2022-03-25

Similar Documents

Publication Publication Date Title
WO2020019901A1 (zh) 通行时间确定方法、装置、计算机设备及存储介质
CN111182453B (zh) 定位方法、装置、电子设备及存储介质
CN109556621B (zh) 一种路线规划的方法及相关设备
US20140336926A1 (en) Adaptive clustering of locations
CN109962939B (zh) 位置推荐方法、装置、服务器、终端及存储介质
CN111080207A (zh) 订单处理方法、装置、设备及存储介质
CN112732858B (zh) 路径规划方法、装置、计算机设备及存储介质
CN112163717B (zh) 人口数据预测方法、装置、计算机设备及介质
CN109887268B (zh) 车辆调度方法、装置及存储介质
CN111813532A (zh) 一种基于多任务机器学习模型的图像管理方法及装置
CN115199240B (zh) 一种页岩气井产量预测方法、装置及存储介质
CN114331492A (zh) 媒体资源的推荐方法、装置、设备及存储介质
WO2022213733A1 (zh) 获取飞行航线的方法、装置、计算机设备及可读存储介质
CN111429235A (zh) 获取订单热力信息的方法、装置、设备及存储介质
CN111428937A (zh) 用户流失的预测方法、装置、电子设备及可读存储介质
CN112414420A (zh) 一种基于交通流量的导航方法以及相关装置
CN114879879B (zh) 一种显示健康码的方法、电子设备及存储介质
CN111695981A (zh) 资源转移方法、装置及存储介质
CN106331310A (zh) 移动终端及其控制方法
CN115966098A (zh) 一种预测公交车到站时间的方法及设备
CN111080033B (zh) 预测服务容量的方法、装置、计算机设备及存储介质
CN111275300A (zh) 路网数据处理方法、装置、设备及存储介质
CN115222316A (zh) 预计送达时间获取方法、装置、电子设备及存储介质
CN115311851B (zh) 路况信息确定方法、装置、电子设备及存储介质
US20220237639A1 (en) System and method for data prediction using heat maps

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: 19840900

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19840900

Country of ref document: EP

Kind code of ref document: A1