WO2022142574A1 - 流量预测模型的训练方法、装置和电子设备 - Google Patents

流量预测模型的训练方法、装置和电子设备 Download PDF

Info

Publication number
WO2022142574A1
WO2022142574A1 PCT/CN2021/123025 CN2021123025W WO2022142574A1 WO 2022142574 A1 WO2022142574 A1 WO 2022142574A1 CN 2021123025 W CN2021123025 W CN 2021123025W WO 2022142574 A1 WO2022142574 A1 WO 2022142574A1
Authority
WO
WIPO (PCT)
Prior art keywords
traffic
prediction model
area
characteristic data
traffic prediction
Prior art date
Application number
PCT/CN2021/123025
Other languages
English (en)
French (fr)
Inventor
宋礼
张钧波
郑宇�
Original Assignee
京东城市(北京)数字科技有限公司
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
Priority claimed from CN202011584094.3A external-priority patent/CN113762578B/zh
Application filed by 京东城市(北京)数字科技有限公司 filed Critical 京东城市(北京)数字科技有限公司
Publication of WO2022142574A1 publication Critical patent/WO2022142574A1/zh

Links

Images

Classifications

    • 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
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • 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
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Definitions

  • the present application relates to the technical field of deep learning, and in particular, to a training method, apparatus, electronic device and storage medium for a traffic prediction model.
  • traffic forecasting technology is widely used in traffic planning, data statistics and other fields.
  • traffic forecasting technology can be used to predict the flow of people in multiple areas of the city, and public transportation facilities, such as public transportation, can be set up in areas with high traffic flow. Stations, subway stations, public bicycles, etc.
  • Most of the traffic prediction methods in the related art use models to predict the traffic flow.
  • the traffic prediction models in the related art cannot accurately reflect the influence of the characteristic data of the influencing factors on the predicted traffic flow, which is not conducive to the performance of the traffic prediction model.
  • the present application aims to solve one of the technical problems in the related art at least to a certain extent.
  • the embodiment of the first aspect of the present application proposes a training method for a traffic prediction model, which inputs the historical traffic characteristics of each area and the characteristic data of the influencing factors that affect the regional traffic into the traffic prediction model.
  • the input layer of the model begins to introduce the influence of the characteristic data of the influencing factors on the predicted traffic of each area, so as to introduce the characteristic data of the influencing factors in the whole process of the traffic prediction model, and fuse the historical traffic characteristics and characteristic data of each area.
  • the embodiment of the second aspect of the present application provides an apparatus for training a traffic prediction model.
  • An embodiment of the third aspect of the present application provides an electronic device.
  • Embodiments of the fourth aspect of the present application provide a computer-readable storage medium.
  • An embodiment of the first aspect of the present application proposes a method for training a traffic prediction model, including: acquiring historical traffic characteristics of each region and feature data of influencing factors affecting regional traffic, and inputting them into the traffic prediction model; In the traffic prediction model, for each region, feature fusion is performed on the historical traffic characteristics of each region and the feature data to obtain the predicted traffic of the traffic prediction model corresponding to each region; According to the predicted traffic and tag traffic corresponding to the area, the traffic prediction model is adjusted to obtain the target traffic prediction model.
  • the historical traffic characteristics of each region and the characteristic data of the influencing factors affecting the regional traffic are input into the traffic prediction model, and the influence can be introduced from the input layer of the traffic prediction model.
  • the predicted flow of the flow forecasting model can effectively solve the problem of lag in the fusion of the characteristic data of the influencing factors and the historical flow characteristics in related technologies, and can accurately reflect the influence of the characteristic data of the influencing factors on the forecast flow, which is conducive to improving the flow forecasting model. performance.
  • training method of the traffic prediction model according to the above-mentioned embodiment of the present application may also have the following additional technical features:
  • the traffic prediction model includes N feature extraction layers, and the Nth layer outputs the predicted traffic in each area, then in the traffic prediction model, for each area, Perform feature fusion on the historical traffic characteristics of each area and the characteristic data, including: for each area in the i-th layer, for each area in the previous layer of traffic characteristics, and the influence factor in the previous layer.
  • the feature data is fused to obtain the traffic characteristics of each region in the i-th layer, where N ⁇ i ⁇ 1.
  • the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer are fused to obtain the traffic characteristics of each area in the i-th layer , including: obtaining the weight set of the i-th layer output after the last training in the traffic prediction model, wherein the weight set includes the weights corresponding to the respective regions on the i-th layer and the influencing factors; Based on the weight set, weighted fusion is performed on the traffic characteristics of each region in the previous layer and the feature data of the influencing factors in the previous layer to obtain the traffic characteristics of each region in the i-th layer.
  • the method further includes: separately acquiring each region A vector representation of the historical traffic characteristics of the area and the characteristic data of the influencing factors.
  • the obtaining the vector representation of the historical traffic characteristics of each area and the characteristic data of the influencing factors respectively includes: based on a nonlinear function in the traffic prediction model, and the The non-linear change parameters of the previous layer outputted by the traffic prediction model after the last training, perform non-linear transformation processing on the historical traffic characteristics and the characteristic data of the influencing factors to generate the vector representation.
  • adjusting the traffic prediction model according to the predicted traffic and label traffic corresponding to each area includes: acquiring the predicted traffic and the label corresponding to each area The error between the flows, based on the error, obtain the loss function of the flow prediction model; based on the loss function, obtain the gradient information of the model parameters of the predicted flow model; update the predicted flow based on the gradient information Model parameters of the model, wherein the model parameters include the weight set of each layer in the N feature extraction layers and the nonlinear change parameter.
  • the method before adjusting the traffic prediction model according to the predicted traffic and tag traffic corresponding to each area, the method further includes: determining the respective all traffic based on the historical traffic corresponding to each area. described tag traffic.
  • the feature data of the influence factor is historical feature data of the influence factor or predicted feature data of the influence factor.
  • the embodiment of the second aspect of the present application proposes a training device for a traffic prediction model, including: an acquisition module for acquiring historical traffic characteristics of each area and characteristic data of influencing factors affecting regional traffic, and inputting them into the In the traffic prediction model; the prediction module is used for, in the traffic prediction model, for each region, feature fusion of the historical traffic characteristics and the feature data of each region to obtain the The predicted traffic of the traffic prediction model; the adjustment module is used to adjust the traffic prediction model according to the predicted traffic and tag traffic corresponding to each area, so as to obtain the target traffic prediction model.
  • the training device for the traffic prediction model inputs the historical traffic characteristics of each area and the characteristic data of the influencing factors affecting the regional traffic into the traffic prediction model, and the influencing factors can be introduced from the input layer of the traffic prediction model.
  • the influence of the characteristic data on the forecast traffic of each area so that the characteristic data of the influencing factors are introduced in the whole process of the traffic forecast model, and the historical traffic characteristics and characteristic data of each area are fused to obtain the corresponding
  • the predicted flow of the flow prediction model can effectively solve the problem that the fusion of the characteristic data of the influencing factors and the historical flow characteristics is relatively lagging in related technologies, and can accurately reflect the influence of the characteristic data of the influencing factors on the predicted flow, which is conducive to improving the performance of the flow prediction model. performance.
  • training device for the traffic prediction model may also have the following additional technical features:
  • the traffic prediction model includes N feature extraction layers, and the Nth layer outputs the predicted traffic of each area, then the prediction module includes: a fusion unit for each The area is in the i-th layer, and the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer are fused to obtain the traffic characteristics of each area in the i-th layer, where N ⁇ i ⁇ 1.
  • the fusion unit is specifically configured to: obtain the weight set of the i-th layer output after the last training in the traffic prediction model, wherein the weight set includes the The weights corresponding to each area on the i-th layer and the influencing factors; based on the weight set, the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer are weighted and fused to obtain Traffic characteristics of each region at the i-th layer.
  • the prediction module includes: a first obtaining unit, configured to obtain a vector representation of the historical traffic characteristics of each area and the characteristic data of the influencing factors, respectively.
  • the first obtaining unit is specifically configured to: based on the nonlinear function in the traffic prediction model and the nonlinear change of the previous layer output after the last training of the traffic prediction model parameters, and perform nonlinear transformation processing on the historical traffic characteristics and the characteristic data of the influencing factors to generate the vector representation.
  • the adjustment module includes: a second acquisition unit, configured to acquire an error between the predicted traffic corresponding to each area and the tag traffic, and based on the error, acquire the The loss function of the traffic prediction model; the third obtaining unit is used to obtain the gradient information of the model parameters of the predicted traffic model based on the loss function; the updating unit is used to update the predicted traffic model based on the gradient information
  • the model parameters wherein the model parameters include the weight set of each layer in the N feature extraction layers and the nonlinear change parameter.
  • the adjustment module is further configured to: determine the respective label traffic based on the historical traffic corresponding to each area.
  • the feature data of the influence factor is historical feature data of the influence factor or predicted feature data of the influence factor.
  • the embodiment of the third aspect of the present application proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and running on the processor, when the processor executes the program, the above-mentioned first
  • the method for training a traffic prediction model is described in the embodiment.
  • the processor executes the computer program stored on the memory, and inputs the historical flow characteristics of each area and the characteristic data of the influencing factors affecting the regional flow into the flow prediction model, which can be predicted from the flow rate.
  • the input layer of the model begins to introduce the influence of the characteristic data of the influencing factors on the predicted traffic of each area, so as to introduce the characteristic data of the influencing factors in the whole process of the traffic prediction model, and fuse the historical traffic characteristics and characteristic data of each area.
  • Embodiments of the fourth aspect of the present application provide a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the method for training a traffic prediction model as described in the foregoing first aspect embodiment.
  • the computer-readable storage medium of the embodiment of the present application by storing a computer program and being executed by the processor, inputs the historical traffic characteristics of each area and the characteristic data of the influencing factors that affect the regional traffic into the traffic prediction model, and the traffic can be obtained from the traffic
  • the input layer of the prediction model begins to introduce the influence of the characteristic data of the influencing factors on the predicted traffic of each area, so as to introduce the characteristic data of the influencing factors in the whole process of the traffic prediction model, and analyze the historical traffic characteristics and characteristic data of each area.
  • Fusion to obtain the predicted flow of the corresponding flow forecasting model for each area, which can effectively solve the problem of lag in the fusion of the characteristic data of the influencing factors and the historical flow characteristics in related technologies, and can accurately reflect the influence of the characteristic data of the influencing factors on the predicted flow. It is beneficial to improve the performance of the traffic prediction model.
  • FIG. 1 is a schematic flowchart of a training method for a traffic prediction model according to an embodiment of the present application
  • FIG. 2 is a schematic flowchart of obtaining the traffic characteristics of each area at the i-th layer in a training method for a traffic prediction model according to an embodiment of the present application;
  • FIG. 3 is a schematic flowchart of adjusting a traffic prediction model in a training method for a traffic prediction model according to an embodiment of the present application
  • FIG. 4 is a schematic structural diagram of a traffic prediction model in a training method for a traffic prediction model according to an embodiment of the present application
  • FIG. 5 is a schematic structural diagram of a training device for a traffic prediction model according to an embodiment of the present application.
  • FIG. 6 is a schematic structural diagram of a training device for a traffic prediction model according to another embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of a training method for a traffic prediction model according to an embodiment of the present application.
  • the training method of the traffic prediction model includes:
  • the execution body of the method for training the traffic prediction model in the embodiment of the present application may be a training device for the traffic prediction model
  • the training device for the traffic prediction model in the embodiment of the present application may be configured in any electronic device, so that the The electronic device may execute the training method of the traffic prediction model according to the embodiment of the present application.
  • the electronic device can be a personal computer (Personal Computer, PC for short), a cloud device, a mobile device, etc.
  • the mobile device can be, for example, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, a vehicle-mounted device, etc. with various operating systems, Hardware devices for touch screens and/or display screens.
  • region division may be performed according to actual conditions. For example, it can be divided according to the functional area of the city, for example, it can be divided into administrative area, entertainment area, shopping area, residential area and so on.
  • regional division may also be performed according to administrative regions at all levels and traffic route information, where the traffic route information includes but is not limited to urban arterial roads, traffic intersections, and the like.
  • historical traffic characteristics include but are not limited to human traffic characteristics, vehicle traffic characteristics, and the like.
  • different regions may correspond to different historical traffic characteristics.
  • the divided areas include Company A and Park B, and the historical traffic characteristics are the characteristics of human traffic
  • the characteristics of the traffic of Company A during the working hours are aggregation characteristics
  • the company The characteristics of the flow of people in the off-duty time period of weekdays (such as 6:00 pm to 7:00 pm) are scattered characteristics
  • the park B has no aggregation characteristics and scattered characteristics during the on-duty and off-duty time periods on weekdays
  • company A and park B can correspond to different historical traffic characteristics.
  • the types of influencing factors affecting regional traffic are not limited, and the types of characteristic data of the influencing factors are also not limited.
  • the influencing factors include but are not limited to weather, date, location, transportation facilities, etc.
  • the corresponding characteristic data may include temperature, wind direction, wind force, humidity, etc.
  • the corresponding The characteristic data of the data can include working days and holidays.
  • the influencing factor is transportation facilities
  • the corresponding characteristic data can include bus lines, subway lines, high-speed lines, etc.
  • the characteristic data of the influencing factors may be historical characteristic data of the influencing factors or predicted characteristic data of the influencing factors.
  • the corresponding feature data may be historical weather feature data, such as historical temperature, historical wind direction, historical humidity, etc., or the corresponding feature data may also be weather forecast feature data, such as forecasting Temperature, predicted wind direction, predicted humidity, etc.
  • different regions may correspond to different influencing factors and feature data corresponding to different influencing factors.
  • the feature data of different influencing factors is used for model training, which helps to improve the accuracy of the model.
  • the historical traffic characteristics of each region and the characteristic data of the influencing factors affecting the regional traffic can be input into the traffic prediction model.
  • the traffic prediction model can be set according to the actual situation, for example, it can be a deep learning model, a traffic prediction model can be constructed based on the TensorFlow platform, and it can also be an STMP (Spatio-Temporal Multiple Points) prediction model.
  • the historical traffic characteristics of each area and the characteristic data of the influencing factors affecting the regional traffic can be input into the traffic prediction model, and the traffic prediction model can be specific to each area. , and perform feature fusion on the historical traffic features and feature data of each region to obtain the predicted traffic of the traffic prediction model corresponding to each region. That is to say, the influence of the characteristic data of the influencing factors on the predicted flow of each region can be introduced from the input layer of the flow prediction model, so that the characteristic data of the influencing factors can be introduced in the whole process of the flow prediction model, and the characteristic data of each region can be introduced.
  • the historical traffic features and feature data are fused to obtain the predicted traffic of the traffic forecasting model corresponding to each region, which can effectively solve the problem of lag in the fusion of the feature data of the influencing factors and the historical traffic features in related technologies, and can accurately reflect the influencing factors.
  • the influence of the characteristic data on the predicted traffic is beneficial to improve the performance of the traffic prediction model.
  • the tag traffic corresponding to each area is the actual reference traffic of the predicted traffic.
  • the respective tagged traffic may be determined based on the historical traffic corresponding to each area. For example, the most recent N times of historical traffic corresponding to each area can be obtained, and the historical traffic with the most occurrences among the most recent N times of historical traffic or the average of the most recent N times of historical traffic can be used as the label traffic corresponding to each area. Alternatively, the most recent historical traffic corresponding to each area can be obtained and used as the label traffic corresponding to each area.
  • N is a positive integer, which can be set according to the actual situation, for example, N can be 3.
  • the model parameters of the traffic prediction model can be adjusted according to the above error until the number of iterations of the model reaches the preset number of times threshold, or reaches the predetermined number of times.
  • the training of the model can be stopped when the preset stopping conditions are met, or the preset model evaluation index is reached, and the model obtained at the end of the last adjustment is used as the target traffic prediction model.
  • the historical traffic characteristics of each area and the characteristic data of the influencing factors that affect the regional traffic are input into the traffic prediction model, which can be obtained from the input layer of the traffic prediction model.
  • the traffic prediction model which can be obtained from the input layer of the traffic prediction model.
  • the predicted traffic of the traffic prediction model corresponding to each region can effectively solve the problem of lag in the fusion of the characteristic data of the influencing factors and the historical traffic characteristics in related technologies, and can accurately reflect the influence of the characteristic data of the influencing factors on the predicted traffic, which is conducive to improving the Performance of Traffic Prediction Models.
  • acquiring the historical traffic characteristics of each area in step S101 may include determining the respective historical traffic characteristics based on the historical traffic corresponding to each area.
  • feature extraction may be performed on the historical traffic corresponding to each area according to a priori rule, and the extracted features may be used as the historical traffic characteristics corresponding to each area.
  • the historical traffic characteristics of the last day corresponding to each area can be extracted, or the historical traffic characteristics of Mondays in the last three weeks of each area can be extracted, or the historical traffic characteristics of each area in July in the last three years can be extracted.
  • determining the respective historical traffic characteristics based on the historical traffic corresponding to each area may include extracting the adjacent characteristics and periodic characteristics of the historical traffic corresponding to each area according to a preset frequency, and extracting the adjacent characteristics of the extracted adjacent characteristics. and periodic features as the historical traffic features corresponding to each region.
  • the proximity feature refers to the historical traffic feature closest to the historical traffic time at the current moment
  • the periodic feature refers to the historical traffic feature corresponding to the same moment in multiple cycles.
  • the proximity feature may include the history corresponding to time t-1, t-2, t-3, t-4, t-5, t-6, and t-7 Traffic characteristics
  • the preset frequency is 7
  • the periodic characteristics with the number of time slices being 2 may include historical traffic characteristics corresponding to time t-7 and t-14.
  • the traffic prediction model may include N feature extraction layers, and the Nth layer outputs the predicted traffic of each area. It can be understood that the traffic prediction model can achieve different levels of traffic feature extraction through N feature extraction layers, and the i-th layer outputs the traffic characteristics of each region in the i-th layer and the characteristic data of the influencing factors in the i-th layer. Among them, N ⁇ i ⁇ 1, N is a positive integer, which can be set according to the actual situation.
  • the N feature extraction layers may adopt a fully connected neural network (Fully Connected Neural Network, FCNN) structure.
  • FCNN Fully Connected Neural Network
  • step S102 in the traffic prediction model, for each region, feature fusion is performed on the historical traffic characteristics and feature data of each region, which may include the i-th layer for each region, and the previous layer for each region.
  • the traffic characteristics and the feature data of the influencing factors in the previous layer are fused to obtain the traffic characteristics of each area in the i-th layer, where N ⁇ i ⁇ 1.
  • the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer can be fused to obtain the traffic characteristics of each area in the i-th layer.
  • the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer are fused to obtain the traffic characteristics of each area in the i-th layer, which may include:
  • S201 Obtain a weight set of the ith layer output after the last training in the traffic prediction model, wherein the weight set includes the weights corresponding to each area on the ith layer and the influencing factors.
  • the traffic prediction model can output the weight set of the ith layer after the last training, and the weight set includes the weights corresponding to each area on the ith layer and the influencing factors.
  • each area and influencing factor on the i-th layer may correspond to different weights respectively.
  • weighted fusion is performed on the traffic characteristics of each region in the previous layer and the feature data of the influencing factors in the previous layer, so as to obtain the traffic characteristics of each region in the i-th layer.
  • the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer are weighted and fused to obtain the traffic characteristics of each area in the i-th layer, which may include obtaining The first product sum of the traffic characteristics of each region in the previous layer and the corresponding weights, and the second product sum of the characteristic data of the influencing factors in the previous layer and the corresponding weights, the first product sum and the second product sum. and value, as the traffic characteristics of each region in the i-th layer.
  • the method can perform weighted fusion of the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer according to the weight set composed of the weights corresponding to each area and the influencing factors on the i-th layer, so as to The traffic characteristics of each area at the i-th layer can be obtained, and the influence of different areas on the traffic characteristics can be considered.
  • step S102 in the traffic prediction model, for each region, before performing feature fusion on the historical traffic characteristics and feature data of each region, it also includes separately acquiring the historical traffic characteristics of each region. and the vector representation of the characteristic data of the influencing factors.
  • the historical traffic characteristics and the characteristic data of the influencing factors of each area can be in the form of vectors, and the vector representation of the historical traffic characteristics and the characteristic data of the influencing factors of each area can be obtained respectively.
  • the vector representation of the historical traffic characteristics and the characteristic data of the influencing factors of each area which may include the nonlinear function based on the traffic prediction model and the nonlinearity of the previous layer output after the last training of the traffic prediction model.
  • the traffic prediction model can output the nonlinear variation parameters of the i-th layer after the last training, where N ⁇ i ⁇ 1.
  • the same nonlinear variation parameters can be shared between different regions, so as to reduce model parameter data and reduce the complexity of model optimization.
  • the method can perform nonlinear transformation processing on the characteristic data of historical traffic characteristics and influencing factors to generate a vector representation, which can improve the nonlinear ability of the model.
  • step S103 according to the predicted traffic and tag traffic corresponding to each area, the traffic prediction model is adjusted, which may include:
  • a forward propagation (Forward Propagation) algorithm the above error, and the loss function of the computer traffic prediction model can be used.
  • the number of model parameters is at least one, and the gradient information of each model parameter of the predicted traffic model can be obtained based on the loss function.
  • At least one model parameter optimizer may be pre-built, and based on the loss function, the gradient information of each model parameter corresponding to each model parameter optimizer is obtained.
  • model parameter optimizers A and B For example, assuming that two model parameter optimizers are preset, namely model parameter optimizers A and B, model parameter optimizer A is used to optimize nonlinear parameters, and model parameter optimizer B is used to optimize model structure parameters. Loss function to obtain the gradient information of each model parameter corresponding to model parameter optimizers A and B.
  • the model parameters may include weight sets and nonlinear change parameters of each of the N feature extraction layers, and may also include model structure parameters, etc., which are not limited here.
  • the model structure parameters are used to determine the impact of different characteristics on traffic forecasting.
  • the method can obtain the error between the predicted traffic and tag traffic corresponding to each area, obtain the loss function of the traffic prediction model based on the error, obtain the gradient information of the model parameters of the predicted traffic model based on the loss function, and The gradient information updates the model parameters of the predicted traffic model.
  • the traffic prediction model includes a representation learning layer and feature extraction layers 1 to 3.
  • the historical traffic characteristics of each area and the characteristic data of the influencing factors that affect the regional traffic can be input into the representation learning layer, and the vector representation of the historical traffic characteristics of each area and the characteristic data of the influencing factors can be obtained through the representation learning layer.
  • Each feature extraction layer performs feature fusion on the historical traffic features and feature data of each region for each region to obtain the traffic features of each region in each feature extraction layer, and feature extraction layer 3 outputs the predicted traffic of each region.
  • the present disclosure also provides a training device for the traffic prediction model. It corresponds to the training method of the traffic prediction model provided in the embodiment of FIG. 4 . Therefore, the implementation of the training method of the traffic prediction model is also applicable to the training device of the traffic prediction model provided by the embodiment of the present disclosure, which is not repeated in the embodiment of the present disclosure. Detailed Description.
  • FIG. 5 is a schematic structural diagram of a training apparatus for a traffic prediction model according to an embodiment of the present application.
  • the apparatus 100 for training a traffic prediction model in this embodiment of the present application may include: an acquisition module 110 , a prediction module 120 and an adjustment module 130 .
  • the acquisition module 110 is used to acquire the historical traffic characteristics of each region and the characteristic data of the influencing factors affecting the regional traffic, and input them into the traffic prediction model;
  • the prediction module 120 is configured to, in the traffic prediction model, perform feature fusion on the historical traffic features and the feature data of each region for each region, so as to obtain the prediction of the traffic prediction model corresponding to each region flow;
  • the adjustment module 130 is configured to adjust the traffic prediction model according to the predicted traffic and tag traffic corresponding to each area to obtain a target traffic prediction model.
  • the traffic prediction model includes N feature extraction layers, and the Nth layer outputs the predicted traffic of each area.
  • the prediction module 120 includes: fusion The unit 1201 is used to fuse the traffic characteristics of each area in the previous layer and the feature data of the influencing factors in the previous layer for each area in the i-th layer, so as to obtain each area in the i-th layer , where N ⁇ i ⁇ 1.
  • the fusion unit 1201 is specifically configured to: obtain the weight set of the i-th layer output after the last training in the traffic prediction model, wherein the weight set includes all weights corresponding to each area on the i-th layer and the influencing factors; based on the weight set, the traffic characteristics of each area in the previous layer and the characteristic data of the influencing factors in the previous layer are weighted and fused to Obtain the traffic characteristics of each area in the i-th layer.
  • the prediction module 120 includes: a first obtaining unit 1202, configured to obtain the historical traffic characteristics of each area and the characteristic data of the influencing factors respectively vector representation of .
  • the first obtaining unit 1202 is specifically configured to: based on the nonlinear function in the traffic prediction model and the nonlinearity of the previous layer output after the last training of the traffic prediction model changing parameters, and performing nonlinear transformation processing on the historical traffic characteristics and the characteristic data of the influencing factors to generate the vector representation.
  • the adjustment module 130 includes: a second acquiring unit 1301, configured to acquire the error between the predicted traffic corresponding to each area and the tag traffic , based on the error, the loss function of the traffic prediction model is acquired; the third acquisition unit 1302 is used for acquiring the gradient information of the model parameters of the predictive traffic model based on the loss function; the updating unit 1303 is used for The gradient information updates model parameters of the predicted traffic model, wherein the model parameters include a weight set of each of the N feature extraction layers and the nonlinear change parameter.
  • the adjustment module 130 is further configured to: determine the respective label traffic based on the historical traffic corresponding to each area.
  • the feature data of the influence factor is historical feature data of the influence factor or predicted feature data of the influence factor.
  • the training device for the traffic prediction model inputs the historical traffic characteristics of each area and the characteristic data of the influencing factors affecting the regional traffic into the traffic prediction model, and the influencing factors can be introduced from the input layer of the traffic prediction model.
  • the influence of the characteristic data on the forecast traffic of each area so that the characteristic data of the influencing factors are introduced in the whole process of the traffic forecast model, and the historical traffic characteristics and characteristic data of each area are fused to obtain the corresponding
  • the predicted flow of the flow prediction model can effectively solve the problem that the fusion of the characteristic data of the influencing factors and the historical flow characteristics is relatively lagging in related technologies, and can accurately reflect the influence of the characteristic data of the influencing factors on the predicted flow, which is conducive to improving the performance of the flow prediction model. performance.
  • the present application further proposes an electronic device 200, including: a memory 210, a processor 220, and a computer program stored in the memory 210 and running on the processor 220, the processor 220 When the program is executed, the training method for the traffic prediction model proposed in the foregoing embodiments of the present application is implemented.
  • the processor executes the computer program stored on the memory, and inputs the historical flow characteristics of each area and the characteristic data of the influencing factors affecting the regional flow into the flow prediction model, which can be predicted from the flow rate.
  • the input layer of the model begins to introduce the influence of the characteristic data of the influencing factors on the predicted traffic of each area, so as to introduce the characteristic data of the influencing factors in the whole process of the traffic prediction model, and fuse the historical traffic characteristics and characteristic data of each area.
  • the present application further provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the training method of the traffic prediction model proposed in the foregoing embodiments of the present application.
  • the computer-readable storage medium of the embodiment of the present application by storing a computer program and being executed by the processor, inputs the historical traffic characteristics of each area and the characteristic data of the influencing factors that affect the regional traffic into the traffic prediction model, and the traffic can be obtained from the traffic
  • the input layer of the prediction model begins to introduce the influence of the characteristic data of the influencing factors on the predicted traffic of each area, so as to introduce the characteristic data of the influencing factors in the whole process of the traffic prediction model, and analyze the historical traffic characteristics and characteristic data of each area.
  • Fusion in order to obtain the predicted flow of the corresponding flow prediction model for each area, can effectively solve the problem that the fusion of the characteristic data of the influencing factors and the historical flow characteristics is relatively lagging in the related technology, and can accurately reflect the characteristic data of the influencing factors on the predicted flow. It is beneficial to improve the performance of the traffic prediction model.
  • first and second are only used for descriptive purposes, and should not be construed as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature delimited with “first”, “second” may expressly or implicitly include at least one of that feature.
  • plurality means at least two, such as two, three, etc., unless expressly and specifically defined otherwise.
  • a "computer-readable medium” can be any device that can contain, store, communicate, propagate, or transport the program for use by or in conjunction with an instruction execution system, apparatus, or apparatus.
  • computer readable media include the following: electrical connections with one or more wiring (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), Read Only Memory (ROM), Erasable Editable Read Only Memory (EPROM or Flash Memory), Fiber Optic Devices, and Portable Compact Disc Read Only Memory (CDROM).
  • the computer readable medium may even be paper or other suitable medium on which the program may be printed, as the paper or other medium may be optically scanned, for example, followed by editing, interpretation, or other suitable medium as necessary process to obtain the program electronically and then store it in computer memory.
  • each functional unit in each embodiment of the present application may be integrated into one processing module, or each unit may exist physically alone, or two or more units may be integrated into one module.
  • the above-mentioned integrated modules can be implemented in the form of hardware, and can also be implemented in the form of software function modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.
  • the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, and the like.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Health & Medical Sciences (AREA)
  • Educational Administration (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Traffic Control Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请提出一种流量预测模型的训练方法、装置和电子设备，其中训练方法包括：获取每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，并输入到流量预测模型中；在流量预测模型中，针对每个区域，对各个区域的历史流量特征和特征数据进行特征融合，以获取每个区域对应的预测流量；根据每个区域对应的预测流量和标签流量，对流量预测模型进行调整，以得到目标流量预测模型。由此，该方法可从模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，可准确反映特征数据对预测流量的影响，有利于提高模型的性能。

Description

流量预测模型的训练方法、装置和电子设备
相关申请的交叉引用
本申请要求京东城市(北京)数字科技有限公司于2020年12月28日提交的、发明名称为“流量预测模型的训练方法、装置和电子设备”的、中国专利申请号“202011584094.3”的优先权。
技术领域
本申请涉及深度学习技术领域，尤其涉及一种流量预测模型的训练方法、装置、电子设备和存储介质。
背景技术
目前，流量预测技术被广泛应用于交通规划、数据统计等领域中，例如，可采用流量预测技术对城市的多个区域进行人流量预测，在人流量较多的区域设置公共交通设施，例如公交站、地铁站、公共自行车等。相关技术中的流量预测方法，大多采用模型来进行流量预测，然而相关技术中的流量预测模型，无法准确反映影响因素的特征数据对预测流量的影响，不利于流量预测模型的性能。
发明内容
本申请旨在至少在一定程度上解决相关技术中的技术问题之一。
为此，本申请第一方面实施例提出一种流量预测模型的训练方法，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
本申请第二方面实施例提出一种流量预测模型的训练装置。
本申请第三方面实施例提出一种电子设备。
本申请第四方面实施例提出一种计算机可读存储介质。
本申请第一方面实施例提出了一种流量预测模型的训练方法，包括：获取每个区域的 历史流量特征，以及影响区域流量的影响因素的特征数据，并输入到所述流量预测模型中；在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合，以获取每个区域对应的所述流量预测模型的预测流量；根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整，以得到目标流量预测模型。
根据本申请实施例的流量预测模型的训练方法，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
另外，根据本申请上述实施例的流量预测模型的训练方法还可以具有如下附加的技术特征：
在本申请的一个实施例中，所述流量预测模型包括N个特征提取层，第N层输出每个区域的所述预测流量，则所述在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合，包括：针对每个区域在第i层，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行融合，得到每个区域在所述第i层的流量特征，其中，N≥i≥1。
在本申请的一个实施例中，所述对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行融合，得到每个区域在所述第i层的流量特征，包括：获取所述流量预测模型中上一次训练后输出的所述第i层的权重集合，其中，所述权重集合中包括所述第i层上各个区域和所述影响因素对应的权重；基于所述权重集合，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行加权融合，以得到每个区域在所述第i层的流量特征。
在本申请的一个实施例中，所述在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合之前，还包括：分别获取每个区域的所述历史流量特征和所述影响因素的特征数据的向量表征。
在本申请的一个实施例中，所述分别获取每个区域的所述历史流量特征和所述影响因素的特征数据的向量表征，包括：基于所述流量预测模型中非线性函数，以及所述流量预测模型上一次训练后输出的前一层的非线性变化参数，对所述历史流量特征和所述影响因素的特征数据进行非线性转换处理，以生成所述向量表征。
在本申请的一个实施例中，所述根据每个区域对应的所述预测流量和标签流量，对所 述流量预测模型进行调整，包括：获取每个区域对应的所述预测流量和所述标签流量之间的误差，基于所述误差，获取所述流量预测模型的损失函数；基于所述损失函数，获取所述预测流量模型的模型参数的梯度信息；基于所述梯度信息更新所述预测流量模型的模型参数，其中，所述模型参数包括所述N个特征提取层中每层的权重集合以及所述非线性变化参数。
在本申请的一个实施例中，所述根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整之前，还包括：基于每个区域对应的历史流量确定各自的所述标签流量。
在本申请的一个实施例中，所述影响因素的特征数据为所述影响因数的历史特征数据或者所述影响因数的预测特征数据。
本申请第二方面实施例提出了一种流量预测模型的训练装置，包括：获取模块，用于获取每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，并输入到所述流量预测模型中；预测模块，用于在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合，以获取每个区域对应的所述流量预测模型的预测流量；调整模块，用于根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整，以得到目标流量预测模型。
本申请实施例的流量预测模型的训练装置，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
另外，根据本申请上述实施例的流量预测模型的训练装置还可以具有如下附加的技术特征：
在本申请的一个实施例中，所述流量预测模型包括N个特征提取层，第N层输出每个区域的所述预测流量，则所述预测模块，包括：融合单元，用于针对每个区域在第i层，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行融合，得到每个区域在所述第i层的流量特征，其中，N≥i≥1。
在本申请的一个实施例中，所述融合单元，具体用于：获取所述流量预测模型中上一次训练后输出的所述第i层的权重集合，其中，所述权重集合中包括所述第i层上各个区域和所述影响因素对应的权重；基于所述权重集合，对各个区域在前一层的流量特征，以及 所述影响因素在前一层的特征数据进行加权融合，以得到每个区域在所述第i层的流量特征。
在本申请的一个实施例中，所述预测模块，包括：第一获取单元，用于分别获取每个区域的所述历史流量特征和所述影响因素的特征数据的向量表征。
在本申请的一个实施例中，所述第一获取单元，具体用于：基于所述流量预测模型中非线性函数，以及所述流量预测模型上一次训练后输出的前一层的非线性变化参数，对所述历史流量特征和所述影响因素的特征数据进行非线性转换处理，以生成所述向量表征。
在本申请的一个实施例中，所述调整模块，包括：第二获取单元，用于获取每个区域对应的所述预测流量和所述标签流量之间的误差，基于所述误差，获取所述流量预测模型的损失函数；第三获取单元，用于基于所述损失函数，获取所述预测流量模型的模型参数的梯度信息；更新单元，用于基于所述梯度信息更新所述预测流量模型的模型参数，其中，所述模型参数包括所述N个特征提取层中每层的权重集合以及所述非线性变化参数。
在本申请的一个实施例中，所述调整模块，还用于：基于每个区域对应的历史流量确定各自的所述标签流量。
在本申请的一个实施例中，所述影响因素的特征数据为所述影响因数的历史特征数据或者所述影响因数的预测特征数据。
本申请第三方面实施例提出了一种电子设备，包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时，实现如前述第一方面实施例所述的流量预测模型的训练方法。
本申请实施例的电子设备，通过处理器执行存储在存储器上的计算机程序，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
本申请第四方面实施例提出了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时，实现如前述第一方面实施例所述的流量预测模型的训练方法。
本申请实施例的计算机可读存储介质，通过存储计算机程序并被处理器执行，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特 征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
本申请附加的方面和优点将在下面的描述中部分给出，部分将从下面的描述中变得明显，或通过本发明的实践了解到。
附图说明
本申请上述的和/或附加的方面和优点从下面结合附图对实施例的描述中将变得明显和容易理解，其中：
图1为根据本申请一个实施例的流量预测模型的训练方法的流程示意图；
图2为根据本申请一个实施例的流量预测模型的训练方法中得到每个区域在第i层的流量特征的流程示意图；
图3为根据本申请一个实施例的流量预测模型的训练方法中对流量预测模型进行调整的流程示意图；
图4为根据本申请一个实施例的流量预测模型的训练方法中流量预测模型的结构示意图；
图5为根据本申请一个实施例的流量预测模型的训练装置的结构示意图；
图6为根据本申请另一个实施例的流量预测模型的训练装置的结构示意图；
图7为根据本申请一个实施例的电子设备的结构示意图。
具体实施方式
下面详细描述本申请的实施例，所述实施例的示例在附图中示出，其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的，旨在用于解释本申请，而不能理解为对本申请的限制。
下面参照附图描述本申请实施例的流量预测模型的训练方法、装置、电子设备和存储介质。
图1为根据本申请一个实施例的流量预测模型的训练方法的流程示意图。
如图1所示，本申请实施例的流量预测模型的训练方法，包括：
S101，获取每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，并输入到流量预测模型中。
需要说明的是，本申请实施例的流量预测模型的训练方法的执行主体可为流量预测模型的训练装置，本申请实施例的流量预测模型的训练装置可以配置在任意电子设备中，以 使该电子设备可以执行本申请实施例的流量预测模型的训练方法。其中，电子设备可以为个人电脑(Personal Computer，简称PC)、云端设备、移动设备等，移动设备例如可以为手机、平板电脑、个人数字助理、穿戴式设备、车载设备等具有各种操作系统、触摸屏和/或显示屏的硬件设备。
本申请的实施例中，可根据实际情况进行区域划分。例如，可根据城市功能区域进行划分，比如，可划分为行政区域、娱乐区域、购物区域、居住区域等。或者，还可根据各级行政区域、交通路线信息进行区域划分，其中，交通路线信息包括但不限于城市主干道、交通路口等。
本申请的实施例中，对每个区域的历史流量特征的类型不做限定，例如，历史流量特征包括但不限于人流量特征、车流量特征等。
可以理解的是，不同区域可对应不同的历史流量特征。例如，假设划分的区域包括公司A、公园B，历史流量特征为人流量特征，则公司A在工作日的上班时间段内(例如早上8点至早上9点)的人流量特征为聚集特征，公司A在工作日的下班时间段内(例如下午6点至下午7点)的人流量特征为分散特征，而公园B在工作日的上班时间段和下班时间段内不具备聚集特征和分散特征，可知公司A与公园B可对应不同的历史流量特征。
本申请的实施例中，对影响区域流量的影响因素的类型不做限定，对影响因素的特征数据的类型也不做限定。可选的，影响因素包括但不限于天气、日期、位置、交通设施等，比如，影响因素为天气时，对应的特征数据可包括温度、风向、风力、湿度等，影响因素为日期时，对应的特征数据可包括工作日、节假日，影响因素为交通设施时，对应的特征数据可包括公交线路、地铁线路、高速线路等。
可选的，影响因素的特征数据可为影响因数的历史特征数据或者影响因数的预测特征数据。例如，影响因素为天气时，对应的特征数据可为天气的历史特征数据，比如，历史温度、历史风向、历史湿度等，或者，对应的特征数据还可为天气的预测特征数据，比如，预测温度、预测风向、预测湿度等。
本申请的实施例中，不同的区域可对应不同的影响因素，以及对应不同的影响因素的特征数据，相较于相关技术中不同区域之间采用相同的影响因素的特征数据进行模型训练，可以针对不同的区域采用不同的影响因素的特征数据进行模型训练，有助于提高模型的精准性。
本申请的实施例中，可将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，输入到流量预测模型中。其中，流量预测模型可根据实际情况进行设置，例如，可为深度学习模型，可基于TensorFlow平台进行流量预测模型的构建，还可为STMP(Spatio-Temporal Multiple Points)预测模型。
S102，在流量预测模型中，针对每个区域，对各个区域的历史流量特征和特征数据进行特征融合，以获取每个区域对应的流量预测模型的预测流量。
相关技术中的流量预测模型，大多仅在模型的最后阶段才引入影响因素的特征数据，影响因素的特征数据与历史流量特征的融合较为滞后，从而无法准确反映影响因素的特征数据对预测流量的影响，不利于流量预测模型的性能。
为了解决这一问题，本申请的实施例中，可将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，输入到流量预测模型中，且流量预测模型可针对每个区域，对各个区域的历史流量特征和特征数据进行特征融合，以获取每个区域对应的流量预测模型的预测流量。也就是说，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
S103，根据每个区域对应的预测流量和标签流量，对流量预测模型进行调整，以得到目标流量预测模型。
本申请的实施例中，每个区域对应的标签流量为预测流量的实际参考流量。可选的，可基于每个区域对应的历史流量确定各自的标签流量。例如，可获取每个区域对应的最近N次的历史流量，将最近N次的历史流量中出现次数最多的历史流量或者最近N次的历史流量的平均值，作为每个区域对应的标签流量。或者，可获取每个区域对应的最近1次的历史流量，将其作为每个区域对应的标签流量。其中，N为正整数，可根据实际情况进行设置，例如，N可为3。
可以理解的是，每个区域对应的预测流量和标签流量之间可能存在误差，则可根据上述误差对流量预测模型的模型参数进行调整，直至模型的迭代次数达到预设次数阈值，或者达到预设的停止条件，或者达到预设的模型评价指标时，可停止模型的训练，将最后一次调整结束时获取的模型作为目标流量预测模型。
综上，根据本申请实施例的流量预测模型的训练方法，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测 流量的影响，有利于提高流量预测模型的性能。
在上述任一实施例的基础上，步骤S101中获取每个区域的历史流量特征，可包括基于每个区域对应的历史流量确定各自的历史流量特征。
例如，可按照先验规则对每个区域对应的历史流量进行特征提取，将提取出的特征作为每个区域对应的历史流量特征。比如，可提取出每个区域对应的最近一天的历史流量特征，或者提取出每个区域最近三周中周一的历史流量特征，或者提取出每个区域最近三年中7月的历史流量特征。
可选的，基于每个区域对应的历史流量确定各自的历史流量特征，可包括按照预设频率对每个区域对应的历史流量进行近邻性特征和周期性特征提取，将提取出的近邻性特征和周期性特征作为每个区域对应的历史流量特征。
其中，近邻性特征指的是距离当前时刻的历史流量时间最近的历史流量特征，周期性特征指的是多个周期内相同时刻对应的历史流量特征。例如，假设预设频率为1，当前时刻为t，近邻性特征可包括t-1、t-2、t-3、t-4、t-5、t-6、t-7时刻对应的历史流量特征，预设频率为7，时间片的数量为2的周期性特征可包括t-7、t-14时刻对应的历史流量特征。
在上述任一实施例的基础上，流量预测模型可包括N个特征提取层，第N层输出每个区域的预测流量。可以理解的是，流量预测模型可通过N个特征提取层实现不同层次的流量特征提取，第i层输出每个区域在第i层的流量特征和影响因素在第i层的特征数据，其中，N≥i≥1，N为正整数，可根据实际情况进行设置。
可选的，N个特征提取层可采用全连接神经网络(Fully Connected Neural Network，FCNN)结构。
进一步地，步骤S102中在流量预测模型中，针对每个区域，对各个区域的历史流量特征和特征数据进行特征融合，可包括针对每个区域在第i层，对各个区域在前一层的流量特征，以及影响因素在前一层的特征数据进行融合，得到每个区域在第i层的流量特征，其中，N≥i≥1。
由此，可对各个区域在前一层的流量特征，以及影响因素在前一层的特征数据进行融合，以得到每个区域在第i层的流量特征。
可选的，如图2所示，对各个区域在前一层的流量特征，以及影响因素在前一层的特征数据进行融合，得到每个区域在第i层的流量特征，可包括：
S201，获取流量预测模型中上一次训练后输出的第i层的权重集合，其中，权重集合中包括第i层上各个区域和影响因素对应的权重。
本申请的实施例中，流量预测模型在上一次训练后可输出第i层的权重集合，权重集合中包括第i层上各个区域和影响因素对应的权重。
可以理解的是，第i层上各个区域和影响因素可能分别对应不同的权重。
S202，基于权重集合，对各个区域在前一层的流量特征，以及影响因素在前一层的特征数据进行加权融合，以得到每个区域在第i层的流量特征。
可以理解的是，基于权重集合，对各个区域在前一层的流量特征，以及影响因素在前一层的特征数据进行加权融合，以得到每个区域在第i层的流量特征，可包括获取各个区域在前一层的流量特征与对应的权重的第一乘积和，以及影响因素在前一层的特征数据与对应的权重的第二乘积和，将第一乘积和与第二乘积和的和值，作为每个区域在第i层的流量特征。
例如，若共有3个区域，分别为区域A、区域B、区域C，区域A的流量等于区域B的流量和区域C的流量之和。则区域A在第i层的流量特征可通过下述公式计算得到：
Figure PCTCN2021123025-appb-000001
其中，
Figure PCTCN2021123025-appb-000002
为区域A在第i层的流量特征，
Figure PCTCN2021123025-appb-000003
为区域A在第i-1层的流量特征，
Figure PCTCN2021123025-appb-000004
为区域B在第i-1层的流量特征，
Figure PCTCN2021123025-appb-000005
为区域C在第i-1层的流量特征，Y i-1为影响因素在第i-1层的特征数据，k 0、k 1、k 2、k 3分别为第i层上区域A、区域B、区域C、影响因素对应的权重。
由此，该方法可根据第i层上各个区域和影响因素对应的权重组成的权重集合，对各个区域在前一层的流量特征，以及影响因素在前一层的特征数据进行加权融合，以得到每个区域在第i层的流量特征，能够考虑到不同区域之间对流量特征的影响。
在上述任一实施例的基础上，步骤S102中在流量预测模型中，针对每个区域，对各个区域的历史流量特征和特征数据进行特征融合之前，还包括分别获取每个区域的历史流量特征和影响因素的特征数据的向量表征。
可以理解的是，每个区域的历史流量特征和影响因素的特征数据可为向量形式，可分别获取每个区域的历史流量特征和影响因素的特征数据的向量表征。
可选的，分别获取每个区域的历史流量特征和影响因素的特征数据的向量表征，可包括基于流量预测模型中非线性函数，以及流量预测模型上一次训练后输出的前一层的非线性变化参数，对历史流量特征和影响因素的特征数据进行非线性转换处理，以生成向量表征。
本申请的实施例中，流量预测模型在上一次训练后可输出第i层的非线性变化参数，其中，N≥i≥1。
可选的，非线性函数可根据实际情况进行设置，例如，可设置为σ(x)＝max(0，x)。
本申请的实施例中，不同区域之间可共享相同的非线性变化参数，以减少模型参数数 据，降低模型优化的复杂度。
由此，该方法可对历史流量特征和影响因素的特征数据进行非线性转换处理，以生成向量表征，从而可提高模型的非线性能力。
在上述任一实施例的基础上，如图3所示，步骤S103中根据每个区域对应的预测流量和标签流量，对流量预测模型进行调整，可包括：
S301，获取每个区域对应的预测流量和标签流量之间的误差，基于误差，获取流量预测模型的损失函数。
可选的，可采用前向传播(Forward Propagation)算法以及上述误差，计算机流量预测模型的损失函数。
S302，基于损失函数，获取预测流量模型的模型参数的梯度信息。
可以理解的是，模型参数的数量为至少一个，可基于损失函数，获取预测流量模型的每个模型参数的梯度信息。
可选的，可预先构建至少一个模型参数优化器，基于损失函数，获取每个模型参数优化器对应的每个模型参数的梯度信息。
例如，假设预先设置两个模型参数优化器，分别为模型参数优化器A、B，模型参数优化器A用于优化非线性变化参数，模型参数优化器B用于优化模型结构参数，则可基于损失函数，获取模型参数优化器A、B对应的每个模型参数的梯度信息。
S303，基于梯度信息更新预测流量模型的模型参数，其中，模型参数包括N个特征提取层中每层的权重集合以及非线性变化参数。
本申请的实施例中，模型参数可包括N个特征提取层中每层的权重集合以及非线性变化参数，还可包括模型结构参数等，这里不做过多限定。其中，模型结构参数用于确定不同特征对流量预测的影响。
由此，该方法可获取每个区域对应的预测流量和标签流量之间的误差，基于误差，获取流量预测模型的损失函数，基于损失函数，获取预测流量模型的模型参数的梯度信息，并基于梯度信息更新预测流量模型的模型参数。
如图4所示，流量预测模型包括表征学习层、特征提取层1至3。可将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到表征学习层，通过表征学习层分别获取每个区域的历史流量特征和影响因素的特征数据的向量表征，通过每个特征提取层针对每个区域，对各个区域的历史流量特征和特征数据进行特征融合，得到每个区域在每个特征提取层的流量特征，特征提取层3输出每个区域的预测流量。
与上述图1至图4实施例提供的流量预测模型的训练方法相对应，本公开还提供一种流量预测模型的训练装置，由于本公开实施例提供的流量预测模型的训练装置与上述图1 至图4实施例提供的流量预测模型的训练方法相对应，因此流量预测模型的训练方法的实施方式也适用于本公开实施例提供的流量预测模型的训练装置，在本公开实施例中不再详细描述。
图5为根据本申请一个实施例的流量预测模型的训练装置的结构示意图。
如图5所示，本申请实施例的流量预测模型的训练装置100可以包括：获取模块110、预测模块120和调整模块130。
获取模块110用于获取每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，并输入到所述流量预测模型中；
预测模块120用于在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合，以获取每个区域对应的所述流量预测模型的预测流量；
调整模块130用于根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整，以得到目标流量预测模型。
在本申请的一个实施例中，所述流量预测模型包括N个特征提取层，第N层输出每个区域的所述预测流量，则如图6所示，所述预测模块120，包括：融合单元1201，用于针对每个区域在第i层，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行融合，得到每个区域在所述第i层的流量特征，其中，N≥i≥1。
在本申请的一个实施例中，所述融合单元1201，具体用于：获取所述流量预测模型中上一次训练后输出的所述第i层的权重集合，其中，所述权重集合中包括所述第i层上各个区域和所述影响因素对应的权重；基于所述权重集合，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行加权融合，以得到每个区域在所述第i层的流量特征。
在本申请的一个实施例中，如图6所示，所述预测模块120，包括：第一获取单元1202，用于分别获取每个区域的所述历史流量特征和所述影响因素的特征数据的向量表征。
在本申请的一个实施例中，所述第一获取单元1202，具体用于：基于所述流量预测模型中非线性函数，以及所述流量预测模型上一次训练后输出的前一层的非线性变化参数，对所述历史流量特征和所述影响因素的特征数据进行非线性转换处理，以生成所述向量表征。
在本申请的一个实施例中，如图6所示，所述调整模块130，包括：第二获取单元1301，用于获取每个区域对应的所述预测流量和所述标签流量之间的误差，基于所述误差，获取所述流量预测模型的损失函数；第三获取单元1302，用于基于所述损失函数，获取所述预测流量模型的模型参数的梯度信息；更新单元1303，用于基于所述梯度信息更新所述预测 流量模型的模型参数，其中，所述模型参数包括所述N个特征提取层中每层的权重集合以及所述非线性变化参数。
在本申请的一个实施例中，所述调整模块130，还用于：基于每个区域对应的历史流量确定各自的所述标签流量。
在本申请的一个实施例中，所述影响因素的特征数据为所述影响因数的历史特征数据或者所述影响因数的预测特征数据。
本申请实施例的流量预测模型的训练装置，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
为了实现上述实施例，如图7所示，本申请还提出一种电子设备200，包括：存储器210、处理器220及存储在存储器210上并可在处理器220上运行的计算机程序，处理器220执行程序时，实现如本申请前述实施例提出的流量预测模型的训练方法。
本申请实施例的电子设备，通过处理器执行存储在存储器上的计算机程序，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
为了实现上述实施例，本申请还提出一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时，实现如本申请前述实施例提出的流量预测模型的训练方法。
本申请实施例的计算机可读存储介质，通过存储计算机程序并被处理器执行，将每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据输入到流量预测模型中，可从流量预测模型的输入层开始引入影响因素的特征数据对每个区域的预测流量的影响，从而在流量预测模型的整个过程中引入影响因素的特征数据，并将各个区域的历史流量特征和特征数据进行融合，以获取每个区域对应的流量预测模型的预测流量，能够有效解决相关技术中影响因素的特征数据与历史流量特征的融合较为滞后的问题，能够准确反映影响因素的特征数据对预测流量的影响，有利于提高流量预测模型的性能。
在本说明书的描述中，参考术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本申请的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不必须针对的是相同的实施例或示例。而且，描述的具体特征、结构、材料或者特点可以在任一个或多个实施例或示例中以合适的方式结合。此外，在不相互矛盾的情况下，本领域的技术人员可以将本说明书中描述的不同实施例或示例以及不同实施例或示例的特征进行结合和组合。
此外，术语“第一”、“第二”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。在本申请的描述中，“多个”的含义是至少两个，例如两个，三个等，除非另有明确具体的限定。
流程图中或在此以其他方式描述的任何过程或方法描述可以被理解为，表示包括一个或更多个用于实现定制逻辑功能或过程的步骤的可执行指令的代码的模块、片段或部分，并且本申请的优选实施方式的范围包括另外的实现，其中可以不按所示出或讨论的顺序，包括根据所涉及的功能按基本同时的方式或按相反的顺序，来执行功能，这应被本申请的实施例所属技术领域的技术人员所理解。
在流程图中表示或在此以其他方式描述的逻辑和/或步骤，例如，可以被认为是用于实现逻辑功能的可执行指令的定序列表，可以具体实现在任何计算机可读介质中，以供指令执行系统、装置或设备(如基于计算机的系统、包括处理器的系统或其他可以从指令执行系统、装置或设备取指令并执行指令的系统)使用，或结合这些指令执行系统、装置或设备而使用。就本说明书而言，"计算机可读介质"可以是任何可以包含、存储、通信、传播或传输程序以供指令执行系统、装置或设备或结合这些指令执行系统、装置或设备而使用的装置。计算机可读介质的更具体的示例(非穷尽性列表)包括以下：具有一个或多个布线的电连接部(电子装置)，便携式计算机盘盒(磁装置)，随机存取存储器(RAM)，只读存储器(ROM)，可擦除可编辑只读存储器(EPROM或闪速存储器)，光纤装置，以及便携式光盘只读存储器(CDROM)。另外，计算机可读介质甚至可以是可在其上打印所述程序的纸或其他合适的介质，因为可以例如通过对纸或其他介质进行光学扫描，接着进行编辑、解译或必要时以其他合适方式进行处理来以电子方式获得所述程序，然后将其存储在计算机存储器中。
应当理解，本申请的各部分可以用硬件、软件、固件或它们的组合来实现。在上述实施方式中，多个步骤或方法可以用存储在存储器中且由合适的指令执行系统执行的软件或固件来实现。如，如果用硬件来实现和在另一实施方式中一样，可用本领域公知的下列技 术中的任一项或他们的组合来实现：具有用于对数据信号实现逻辑功能的逻辑门电路的离散逻辑电路，具有合适的组合逻辑门电路的专用集成电路，可编程门阵列(PGA)，现场可编程门阵列(FPGA)等。
本技术领域的普通技术人员可以理解实现上述实施例方法携带的全部或部分步骤是可以通过程序来指令相关的硬件完成，所述的程序可以存储于一种计算机可读存储介质中，该程序在执行时，包括方法实施例的步骤之一或其组合。
此外，在本申请各个实施例中的各功能单元可以集成在一个处理模块中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个模块中。上述集成的模块既可以采用硬件的形式实现，也可以采用软件功能模块的形式实现。所述集成的模块如果以软件功能模块的形式实现并作为独立的产品销售或使用时，也可以存储在一个计算机可读取存储介质中。
上述提到的存储介质可以是只读存储器，磁盘或光盘等。尽管上面已经示出和描述了本申请的实施例，可以理解的是，上述实施例是示例性的，不能理解为对本申请的限制，本领域的普通技术人员在本申请的范围内可以对上述实施例进行变化、修改、替换和变型。

Claims (18)

  1. 一种流量预测模型的训练方法，其特征在于，包括：
    获取每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，并输入到所述流量预测模型中；
    在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合，以获取每个区域对应的所述流量预测模型的预测流量；
    根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整，以得到目标流量预测模型。
  2. 根据权利要求1所述的流量预测模型的训练方法，其特征在于，所述流量预测模型包括N个特征提取层，第N层输出每个区域的所述预测流量，则所述在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合，包括：
    针对每个区域在第i层，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行融合，得到每个区域在所述第i层的流量特征，其中，N≥i≥1。
  3. 根据权利要求2所述的流量预测模型的训练方法，其特征在于，所述对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行融合，得到每个区域在所述第i层的流量特征，包括：
    获取所述流量预测模型中上一次训练后输出的所述第i层的权重集合，其中，所述权重集合中包括所述第i层上各个区域和所述影响因素对应的权重；
    基于所述权重集合，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行加权融合，以得到每个区域在所述第i层的流量特征。
  4. 根据权利要求1-3任一项所述的流量预测模型的训练方法，其特征在于，所述在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合之前，还包括：
    分别获取每个区域的所述历史流量特征和所述影响因素的特征数据的向量表征。
  5. 根据权利要求4所述的流量预测模型的训练方法，其特征在于，所述分别获取每个区域的所述历史流量特征和所述影响因素的特征数据的向量表征，包括：
    基于所述流量预测模型中非线性函数，以及所述流量预测模型上一次训练后输出的前 一层的非线性变化参数，对所述历史流量特征和所述影响因素的特征数据进行非线性转换处理，以生成所述向量表征。
  6. 根据权利要求5所述的流量预测模型的训练方法，其特征在于，所述根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整，包括：
    获取每个区域对应的所述预测流量和所述标签流量之间的误差，基于所述误差，获取所述流量预测模型的损失函数；
    基于所述损失函数，获取所述预测流量模型的模型参数的梯度信息；
    基于所述梯度信息更新所述预测流量模型的模型参数，其中，所述模型参数包括所述N个特征提取层中每层的权重集合以及所述非线性变化参数。
  7. 根据权利要求1所述的流量预测模型的训练方法，其特征在于，所述根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整之前，还包括：
    基于每个区域对应的历史流量确定各自的所述标签流量。
  8. 根据权利要求1所述的流量预测模型的训练方法，其特征在于，所述影响因素的特征数据为所述影响因数的历史特征数据或者所述影响因数的预测特征数据。
  9. 一种流量预测模型的训练装置，其特征在于，包括：
    获取模块，用于获取每个区域的历史流量特征，以及影响区域流量的影响因素的特征数据，并输入到所述流量预测模型中；
    预测模块，用于在所述流量预测模型中，针对每个区域，对各个区域的所述历史流量特征和所述特征数据进行特征融合，以获取每个区域对应的所述流量预测模型的预测流量；
    调整模块，用于根据每个区域对应的所述预测流量和标签流量，对所述流量预测模型进行调整，以得到目标流量预测模型。
  10. 根据权利要求9所述的流量预测模型的训练装置，其特征在于，所述流量预测模型包括N个特征提取层，第N层输出每个区域的所述预测流量，则所述预测模块，包括：
    融合单元，用于针对每个区域在第i层，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行融合，得到每个区域在所述第i层的流量特征，其中，N≥i≥1。
  11. 根据权利要求10所述的流量预测模型的训练装置，其特征在于，所述融合单元，具体用于：
    获取所述流量预测模型中上一次训练后输出的所述第i层的权重集合，其中，所述权重集合中包括所述第i层上各个区域和所述影响因素对应的权重；
    基于所述权重集合，对各个区域在前一层的流量特征，以及所述影响因素在前一层的特征数据进行加权融合，以得到每个区域在所述第i层的流量特征。
  12. 根据权利要求9-11任一项所述的流量预测模型的训练装置，其特征在于，所述预测模块，包括：
    第一获取单元，用于分别获取每个区域的所述历史流量特征和所述影响因素的特征数据的向量表征。
  13. 根据权利要求12所述的流量预测模型的训练装置，其特征在于，所述第一获取单元，具体用于：
    基于所述流量预测模型中非线性函数，以及所述流量预测模型上一次训练后输出的前一层的非线性变化参数，对所述历史流量特征和所述影响因素的特征数据进行非线性转换处理，以生成所述向量表征。
  14. 根据权利要求13所述的流量预测模型的训练装置，其特征在于，所述调整模块，包括：
    第二获取单元，用于获取每个区域对应的所述预测流量和所述标签流量之间的误差，基于所述误差，获取所述流量预测模型的损失函数；
    第三获取单元，用于基于所述损失函数，获取所述预测流量模型的模型参数的梯度信息；
    更新单元，用于基于所述梯度信息更新所述预测流量模型的模型参数，其中，所述模型参数包括所述N个特征提取层中每层的权重集合以及所述非线性变化参数。
  15. 根据权利要求9所述的流量预测模型的训练装置，其特征在于，所述调整模块，还用于：
    基于每个区域对应的历史流量确定各自的所述标签流量。
  16. 根据权利要求9所述的流量预测模型的训练装置，其特征在于，所述影响因素的 特征数据为所述影响因数的历史特征数据或者所述影响因数的预测特征数据。
  17. 一种电子设备，其特征在于，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时，实现如权利要求1-8中任一项所述的流量预测模型的训练方法。
  18. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，该程序被处理器执行时实现如权利要求1-8中任一项所述的流量预测模型的训练方法。
PCT/CN2021/123025 2020-12-28 2021-10-11 流量预测模型的训练方法、装置和电子设备 WO2022142574A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011584094.3A CN113762578B (zh) 2020-12-28 流量预测模型的训练方法、装置和电子设备
CN202011584094.3 2020-12-28

Publications (1)

Publication Number Publication Date
WO2022142574A1 true WO2022142574A1 (zh) 2022-07-07

Family

ID=78786221

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/123025 WO2022142574A1 (zh) 2020-12-28 2021-10-11 流量预测模型的训练方法、装置和电子设备

Country Status (1)

Country Link
WO (1) WO2022142574A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115936264A (zh) * 2023-02-22 2023-04-07 北京师范大学 单日工程量计算方法、阶段性工程量预测方法及预测装置
CN116370819A (zh) * 2023-04-18 2023-07-04 安徽通灵仿生科技有限公司 一种心室辅助装置的泵血流量估测方法及装置
CN117273199A (zh) * 2023-08-29 2023-12-22 长江水上交通监测与应急处置中心 一种航运信息动态监测及超前预报方法及系统
CN118277066A (zh) * 2024-05-30 2024-07-02 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） 实时任务的任务流量预测方法、装置、设备、介质和产品

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109559512A (zh) * 2018-12-05 2019-04-02 北京掌行通信息技术有限公司 一种区域交通流量预测方法及装置
US20190122144A1 (en) * 2017-10-25 2019-04-25 International Business Machines Corporation Regression for metric dataset
CN110969285A (zh) * 2019-10-29 2020-04-07 京东方科技集团股份有限公司 预测模型训练方法、预测方法、装置、设备及介质
CN111091196A (zh) * 2019-11-15 2020-05-01 佳都新太科技股份有限公司 客流数据确定方法、装置、计算机设备和存储介质
CN111461384A (zh) * 2019-12-10 2020-07-28 阿里巴巴集团控股有限公司 对象流量预测方法、装置及设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190122144A1 (en) * 2017-10-25 2019-04-25 International Business Machines Corporation Regression for metric dataset
CN109559512A (zh) * 2018-12-05 2019-04-02 北京掌行通信息技术有限公司 一种区域交通流量预测方法及装置
CN110969285A (zh) * 2019-10-29 2020-04-07 京东方科技集团股份有限公司 预测模型训练方法、预测方法、装置、设备及介质
CN111091196A (zh) * 2019-11-15 2020-05-01 佳都新太科技股份有限公司 客流数据确定方法、装置、计算机设备和存储介质
CN111461384A (zh) * 2019-12-10 2020-07-28 阿里巴巴集团控股有限公司 对象流量预测方法、装置及设备

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115936264A (zh) * 2023-02-22 2023-04-07 北京师范大学 单日工程量计算方法、阶段性工程量预测方法及预测装置
CN116370819A (zh) * 2023-04-18 2023-07-04 安徽通灵仿生科技有限公司 一种心室辅助装置的泵血流量估测方法及装置
CN116370819B (zh) * 2023-04-18 2024-03-12 安徽通灵仿生科技有限公司 一种心室辅助装置的泵血流量估测方法及装置
CN117273199A (zh) * 2023-08-29 2023-12-22 长江水上交通监测与应急处置中心 一种航运信息动态监测及超前预报方法及系统
CN117273199B (zh) * 2023-08-29 2024-04-02 长江水上交通监测与应急处置中心 一种航运信息动态监测及超前预报方法及系统
CN118277066A (zh) * 2024-05-30 2024-07-02 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） 实时任务的任务流量预测方法、装置、设备、介质和产品

Also Published As

Publication number Publication date
CN113762578A (zh) 2021-12-07

Similar Documents

Publication Publication Date Title
WO2022142574A1 (zh) 流量预测模型的训练方法、装置和电子设备
CN108052937B (zh) 基于弱监督的字符检测器训练方法、装置、系统及介质
US11715001B2 (en) Water quality prediction
WO2022121510A1 (zh) 基于随机策略梯度的交通信号控制方法、系统及电子设备
Dunne et al. Regime-based short-term multivariate traffic condition forecasting algorithm
EP3035314A1 (en) A traffic data fusion system and the related method for providing a traffic state for a network of roads
Shih et al. Grey number prediction using the grey modification model with progression technique
Huang et al. A bimodal Gaussian inhomogeneous Poisson algorithm for bike number prediction in a bike-sharing system
KR101828215B1 (ko) Long Short Term Memory 기반 순환형 상태 전이 모델의 학습 방법 및 장치
Tu et al. Deep traffic congestion prediction model based on road segment grouping
CN111047078B (zh) 交通特征预测方法、系统及存储介质
CN116029395B (zh) 商业区域的人流预警方法、装置、电子设备及存储介质
Zhang et al. Traffic signal optimization for partially observable traffic system and low penetration rate of connected vehicles
CN117590766B (zh) 通道入口导流栏杆角度调整的控制方法和装置
CN115311860B (zh) 一种交通流量预测模型的在线联邦学习方法
Lai et al. Bus arrival time prediction using wavelet neural network trained by improved particle swarm optimization
CN115146844A (zh) 一种基于多任务学习的多模式交通短时客流协同预测方法
Yu et al. Speed‐Density Model of Interrupted Traffic Flow Based on Coil Data
Zou et al. A novel spatio-temporal generative inference network for predicting the long-term highway traffic speed
Han et al. An evacuation guidance model for pedestrians with limited vision
US9412267B2 (en) Auto-calibration for road traffic prediction
CN116992253A (zh) 与目标业务关联的目标预测模型中超参数的取值确定方法
CN113762578B (zh) 流量预测模型的训练方法、装置和电子设备
Bi et al. Particle filter for estimating freeway traffic state in Beijing
Rahman et al. MDLpark: available parking prediction for smart parking through mobile deep learning

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: 21913319

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: 21913319

Country of ref document: EP

Kind code of ref document: A1