Traffic flow data completion method, model training method, device and equipment

Abstract

The present disclosure provides a traffic flow data completion method, a model training method, an apparatus and a device, and relates to the fields of artificial intelligence and intelligent transportation. The specific implementation scheme is: obtaining road network data and vehicle driving data in a traffic area, and historical flow data monitored at a non-target intersection in the traffic area; obtaining historical flow completion data of the non-target intersection and preliminary flow completion data of a target intersection based on trajectory mapping features between the road network data and the vehicle driving data; performing fusion processing on the historical flow data, the historical flow completion data and the preliminary flow completion data to obtain corrected flow data of the non-target intersection; and obtaining target flow completion data of the target intersection based on spatial correlation features between the trajectory mapping features, the corrected flow data and the preliminary flow completion data. The present disclosure solves the problem of real-time traffic flow missing caused by high equipment cost, and provides more complete data support for optimization of a traffic signal control system.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to the fields of artificial intelligence and intelligent transportation. Background Technology

[0002] Traffic Signal Control Systems (TSCS) analyze traffic flow data to detect traffic conditions. Traffic flow data plays a crucial role in assessing road traffic conditions and developing strategies to optimize traffic efficiency. However, due to the high cost of deploying urban traffic equipment and the inability to cover all intersections, some intersections lack real-time traffic flow data, which in turn affects the effective operation and strategy optimization of TSCS. Summary of the Invention

[0003] This disclosure provides a method for completing traffic flow data, a model training method, an apparatus, an electronic device, and a storage medium.

[0004] According to a first aspect of this disclosure, a method for completing traffic flow data is provided, comprising:

[0005] Acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area;

[0006] Based on the trajectory mapping characteristics between the road network data and the vehicle driving data, historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections are obtained.

[0007] The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection.

[0008] The historical traffic data, the preliminary traffic data, and the historical traffic data are fused together to obtain corrected traffic data for non-target intersections;

[0009] Based on the spatial correlation features between the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data, the target traffic completion data for the target intersection is obtained.

[0010] According to a second aspect of this disclosure, a training method for a traffic flow data completion model is provided. The traffic flow data completion model includes a data processing module, a first fusion module, and a second fusion module. The training method includes:

[0011] Acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area;

[0012] The trajectory mapping features between the road network data and the vehicle driving data are input into the data processing module to obtain the historical traffic flow completion data of the non-target intersection and the preliminary traffic flow completion data of the target intersection;

[0013] The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection.

[0014] The historical traffic data, the historical traffic completion data, and the preliminary traffic completion data are input into the first fusion module to obtain the corrected traffic data for the non-target intersection;

[0015] The corrected traffic flow data of the non-target intersection, the preliminary traffic flow completion data, and the trajectory mapping features are input into the second fusion module to obtain the target traffic flow completion data of the target intersection;

[0016] The model loss value is calculated based on the historical traffic completion data, the preliminary traffic completion data, and the target traffic completion data.

[0017] Based on the model loss value, a completion model for the traffic flow data is trained.

[0018] According to a third aspect of this disclosure, a traffic flow data completion device is provided, comprising:

[0019] The acquisition module is used to acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area.

[0020] The generation module is used to obtain historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections based on the trajectory mapping features between the road network data and the vehicle driving data.

[0021] The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection.

[0022] The correction module is used to fuse the historical traffic completion data, the preliminary traffic completion data, and the historical traffic data to obtain corrected traffic data for non-target intersections.

[0023] The fusion module is used to obtain the target traffic completion data of the target intersection based on the spatial correlation features between the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data.

[0024] According to a fourth aspect of this disclosure, a training apparatus for a traffic flow data completion model is provided. The traffic flow data completion model includes a data processing module, a first fusion module, and a second fusion module. The training apparatus includes:

[0025] The acquisition module is used to acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area.

[0026] The generation module is used to input the trajectory mapping features between the road network data and the vehicle driving data into the data processing module to obtain the historical traffic flow completion data of the non-target intersection and the preliminary traffic flow completion data of the target intersection;

[0027] The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection.

[0028] The correction module is used to input the historical traffic data, the historical traffic completion data, and the preliminary traffic completion data into the first fusion module to obtain the corrected traffic data for non-target intersections;

[0029] The completion module is used to input the corrected traffic flow data of the non-target intersection, the preliminary traffic flow completion data, and the trajectory mapping features into the second fusion module to obtain the target traffic flow completion data of the target intersection;

[0030] The calculation module is used to calculate the model loss value based on the historical traffic completion data, the preliminary traffic completion data, and the target traffic completion data;

[0031] An optimization module is used to train a completion model for the traffic flow data based on the model loss value.

[0032] According to a fifth aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the traffic flow data completion method of the first aspect of this disclosure, or to perform the training method of the traffic flow data completion model of the second aspect of this disclosure.

[0033] According to a sixth aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are configured to cause the computer to execute a traffic flow data completion method according to a first aspect of this disclosure, or to execute a training method for a traffic flow data completion model according to a second aspect of this disclosure.

[0034] According to a seventh aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of the traffic flow data completion method according to a first aspect of this disclosure, or the training method of the traffic flow data completion model according to a second aspect of this disclosure.

[0035] The technology disclosed herein solves the problem of missing real-time traffic flow due to high equipment costs, and provides more complete data support for the optimization of traffic control systems.

[0036] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0037] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0038] Figure 1 This is a flowchart of a traffic flow data completion method provided in the first embodiment of this disclosure;

[0039] Figure 2 This is a flowchart of a traffic flow data completion method provided in the second embodiment of this disclosure;

[0040] Figure 3 This is a flowchart of another method for completing traffic flow data provided in the second embodiment of this disclosure;

[0041] Figure 4 This is a flowchart of a traffic flow data completion method provided in the third embodiment of this disclosure;

[0042] Figure 5 This is a flowchart of another method for completing traffic flow data provided in the third embodiment of this disclosure;

[0043] Figure 6 This is a flowchart of a traffic flow data completion model training method provided in the fourth embodiment of this disclosure;

[0044] Figure 7 This is a network structure block diagram of a traffic flow data completion model provided in the fourth embodiment of this disclosure;

[0045] Figure 8 This is a network structure block diagram of a data processing module provided in the fourth embodiment of this disclosure;

[0046] Figure 9 This is a flowchart of a traffic flow data completion model training method provided in the fifth embodiment of this disclosure;

[0047] Figure 10 This is a structural block diagram of a traffic flow data completion device provided in the sixth embodiment of this disclosure;

[0048] Figure 11 This is a structural block diagram of a training device for a traffic flow data completion model provided in the seventh embodiment of this disclosure;

[0049] Figure 12 This is a block diagram of an electronic device provided in the eighth embodiment of the present disclosure for implementing one of the data completion boxes and model training methods of the present disclosure. Detailed Implementation

[0050] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0051] Traffic control systems typically collect traffic flow data through devices deployed at intersections, such as video surveillance and radar equipment. This data is then used to assess road conditions and develop strategies for optimizing traffic flow. However, the high cost of deploying urban traffic equipment means that not all intersections can be covered, resulting in some intersections lacking real-time traffic flow data. This impacts the effective operation and optimization of traffic control systems.

[0052] To address the aforementioned issues, this disclosure proposes a traffic flow data completion method based on deep learning. First, road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area, are acquired. These three types of sample data facilitate comprehensive analysis of traffic space, vehicle trajectories, and traffic flow. Next, based on the trajectory mapping characteristics between the road network data and the vehicle driving data, historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections are obtained. The historical traffic flow completion data represents historical traffic flow data not acquired at the non-target intersections. In this step, the one-to-one spatial-temporal trajectory mapping relationship between traffic space and vehicle trajectories is utilized to generate historical traffic flow data not acquired at non-target intersections and preliminary traffic flow completion data for target intersections. Then, the historical traffic flow completion data, the preliminary traffic flow completion data, and the historical traffic flow data are fused to obtain corrected traffic flow data for non-target intersections. In this step, the spatiotemporal correlation between traffic flow and intersections is used to optimize and correct the traffic flow at non-target intersections. Finally, based on the spatial correlation features between the trajectory mapping features, the corrected traffic flow data, and the preliminary traffic flow completion data, the target traffic flow completion data for the target intersection is obtained. In this step, the spatial correlation between the upstream and downstream non-target intersection traffic flow data and the target intersection traffic flow data in the traffic space is used to generate the target traffic flow completion data for the target intersection. This solves the problem of missing real-time traffic flow data caused by high equipment costs, providing more complete data support for the optimization of traffic control systems.

[0053] The following description, with reference to the accompanying drawings, outlines the methods for completing traffic flow data, model training methods, apparatus and devices, and storage media provided in this disclosure.

[0054] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0055] First Embodiment

[0056] Figure 1 This is a flowchart illustrating a traffic flow data completion method provided in an embodiment of this disclosure. It should be noted that the traffic flow data completion method in this embodiment can be executed by an electronic device. Figure 1 As shown, the method for completing the traffic flow data may include, but is not limited to, the following steps.

[0057] In step 101, road network data and vehicle driving data within the traffic area are acquired, as well as historical traffic flow data monitored at non-target intersections within the traffic area.

[0058] The above three types of sample data are helpful for comprehensive analysis of traffic space, vehicle trajectory and traffic flow.

[0059] Road network data:

[0060] A road network refers to a network of roads composed of connected points and lines. Road network data includes the geometric characteristics, topological relationships, and attribute information of roads. Specifically, it includes information such as the starting and ending points of roads, their length, width, type, and grade, as well as the connections between roads. It provides fundamental data support for applications such as traffic planning, traffic management, and route planning.

[0061] Road network data can be obtained through, but is not limited to, the following methods:

[0062] Public data sources: Publicly available regional road network data provided by government agencies, research institutions, data provider associations, and open data sharing platforms.

[0063] Surveys and monitoring: For specific traffic areas or road networks, detailed data can be obtained through on-site surveys and monitoring.

[0064] Vehicle driving data:

[0065] Vehicle driving data records information such as the vehicle's driving route, time, and vehicle identification information (hereinafter referred to as ID information). The ID information is used for data identification and association.

[0066] Vehicle driving data can be obtained, but is not limited to, from trajectory samples provided by intelligent traffic management systems and vehicle-to-everything (V2X) networks.

[0067] The vehicle driving data used in this embodiment comes from trajectory samples provided by the Internet of Vehicles (IoV). IoV data can be understood as a partial sample of all passing vehicles. Due to the complex and diverse sources of IoV data, and the poor quality of some sources, the sample size of intersection flow trajectories obtained based on trajectory extraction may contain anomalies. To select highly reliable data for extracting intersection trajectory samples, this embodiment performs statistical analysis on a large amount of data, performs anomaly detection and screening on the original trajectory data, eliminates the influence of low-reliability data sources, and uses sample values ​​aggregated from other highly reliable data sources as vehicle driving data.

[0068] Historical traffic data:

[0069] Traffic flow data is statistical data describing the volume of traffic in each direction at an intersection. In embodiments of this disclosure, traffic flow data obtained through monitoring at non-target intersections within a traffic area is referred to as historical flow data.

[0070] Historical traffic data in this embodiment can be obtained from data monitored by devices such as video surveillance systems, radar, vehicle sensors, and mobile device locators located at intersections in traffic areas.

[0071] Due to unstable equipment operation or fluctuating service quality, the collected historical traffic data often contains some outliers. To improve the quality of historical traffic data in the training data, this embodiment uses statistical methods to detect outliers in the original historical traffic data. Based on the 3sigma method, outliers are filtered out and set to -1 to obtain historical traffic data usable in this embodiment.

[0072] In step 102, based on the trajectory mapping features between the road network data and the vehicle driving data, historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections are obtained; wherein, the historical traffic flow completion data is used to represent historical traffic flow data that was not obtained at the non-target intersections.

[0073] Due to equipment malfunctions at some non-target intersections, historical traffic data for some non-target intersections could not be obtained. Therefore, it is necessary to first complete the historical traffic data for the non-target intersections.

[0074] In some embodiments of this disclosure, in this step, the historical traffic flow data not obtained at non-target intersections and the preliminary traffic flow data for the target intersection are generated by utilizing the one-to-one spatial and temporal trajectory mapping relationship between traffic space and vehicle trajectory.

[0075] In step 103, the historical traffic data, the preliminary traffic data, and the historical traffic data are fused to obtain the corrected traffic data for the non-target intersection.

[0076] In some embodiments of this disclosure, the learning and adaptive capabilities of neural networks are used to learn and fuse historical traffic data, historical traffic completion data, and preliminary traffic completion data. By utilizing the spatiotemporal correlation between traffic flow and intersections, traffic flow at non-target intersections is optimized and corrected, and corrected traffic flow data for non-target intersections is generated.

[0077] In step 104, target traffic completion data for the target intersection is obtained based on the spatial correlation features between the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data.

[0078] In some embodiments of this disclosure, based on a deep learning system, the corrected traffic flow data of non-target intersections, the preliminary traffic flow completion data, and the trajectory mapping features described in step 102 are input as sample data into a preset training model. The target traffic flow completion data of the target intersection is generated by using the traffic flow data of upstream and downstream non-target intersections to the traffic flow data of the target intersection in the traffic space.

[0079] The method provided in this embodiment solves the problem of missing real-time traffic flow data caused by high equipment costs, providing more complete data support for the optimization of traffic control systems. It offers a more accurate and reliable data foundation for transportation departments and urban traffic managers, helping to formulate more effective traffic management strategies and decisions.

[0080] Second Embodiment

[0081] To address the issue of missing historical traffic data at non-target intersections and incomplete preliminary traffic data at target intersections, the trajectory mapping features between road network data and vehicle driving data can be integrated to fully utilize the vehicle trajectory mapping relationship between intersections and vehicle trajectories within the traffic area.

[0082] In another embodiment proposed in this disclosure, such as Figure 2 As shown, the traffic flow data completion method may include, but is not limited to, the following steps.

[0083] In step 201, road network data and vehicle driving data within the traffic area are acquired, as well as historical traffic flow data measured at non-target intersections within the traffic area.

[0084] Optionally, step 201 can be implemented using any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0085] In step 202, the trajectory mapping features between the road network data and the vehicle driving data are extracted.

[0086] Optionally, in this embodiment, the vehicle driving data's identity and time features, as well as the road network data's channelization features, are obtained through embedding learning; and the vehicle driving data's trajectory features are obtained through residual learning.

[0087] Embedding learning can map road network data and vehicle driving data into a low-dimensional space, reducing data dimensionality and computational complexity, thereby improving the efficiency of extracting trajectory mapping features and reducing computation time and cost. However, the trajectory features of vehicle driving data are complex and the samples are diverse. Using embedding learning may cause the loss of some important feature information, affecting the accuracy of the results. Therefore, when obtaining the trajectory features of the vehicle driving data, residual learning, which has a higher network generalization ability, is used to obtain vehicle trajectory features more comprehensively and accurately.

[0088] Increased computational complexity: Introducing residual blocks increases the computational complexity of the model, requiring more computational resources.

[0089] Based on each time window, extract the trajectory sample size S, time feature data T, and ID information data I from the vehicle driving data.

[0090] The channelization information flow direction channelization data L for each lane in each direction at each intersection in the road network is extracted and processed into flow-level channelization features. In addition, based on the road network, the spatial connectivity relationships between intersections need to be extracted for subsequent spatial correlation learning in the model.

[0091] Let N be the target intersection containing the target flow direction, P be the non-target intersection containing the non-target flow direction, and M be the sum of the number of all flow directions at all intersections. That is, the set of N and P is M.

[0092] Then, the data is processed to obtain the trajectory mapping features between the road network data and the vehicle driving data: flow-level vehicle driving data S for all time periods. M Time feature data T M ID information data I M Channelization feature data L M .

[0093] By extracting the trajectory mapping features between road network data and vehicle driving data, it is beneficial to make full use of the vehicle trajectory mapping relationship between intersections and vehicle trajectories within the traffic area to complete the traffic flow data of the target intersection.

[0094] In step 203, the corresponding trajectory mapping features are encoded based on the pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data.

[0095] Encoding the corresponding trajectory mapping features based on a pre-trained autoencoder model is beneficial for extracting effective features from the input trajectory mapping features, reducing computational complexity, and improving robustness and scalability.

[0096] Specifically, such as Figure 3 As shown, including but not limited to the following steps.

[0097] In step 301, the corresponding trajectory mapping features are input into the autoencoder learning model; the autoencoder learning model includes an encoder and a decoder.

[0098] The encoder used in this embodiment can be a convolutional encoder or a cyclic encoder, and the decoder can also be a convolutional decoder or a cyclic decoder. This embodiment does not limit the specific encoder used in this embodiment.

[0099] Using autoencoder learning models to process sample data can convert the sample data into a fixed-length numerical representation, which facilitates processing, storage, and transmission, and improves the efficiency and accuracy of sample processing.

[0100] In step 302, the encoder outputs the corresponding encoded information after it has sequentially encoded the corresponding trajectory mapping features multiple times.

[0101] By encoding the trajectory mapping features multiple times, the trajectory mapping features can be abstracted and extracted layer by layer, capturing more effective feature representations from the input trajectory mapping features.

[0102] In step 303, the corresponding encoded information is input to the decoder to obtain the historical traffic completion data and the preliminary traffic completion data output by the decoder after the corresponding encoded information is decoded multiple times in sequence.

[0103] In step 303, the number of decoding operations is the same as the number of encoding operations in step 302.

[0104] By obtaining decoding information through multiple decodings, the original features and structure of the input data can be gradually restored, resulting in a more accurate and complete decoding result for the trajectory mapping features.

[0105] For example, regarding the acquisition of historical traffic completion data: In this embodiment, the vehicle driving data S in the trajectory mapping features is first... M Time feature data T M ID information data I M Channelization feature data L M The data is input into the autoencoder learning model. Then the encoder processes the vehicle driving data S... M Time feature data T M ID information data I M Channelization feature data L M The corresponding encoded information is output after multiple encoding processes. Finally, the corresponding encoded information is input to the decoder, which decodes the corresponding encoded information multiple times and outputs historical traffic completion data.

[0106] For example, regarding the acquisition of preliminary traffic flow completion data for a target intersection: since there are no traffic flow monitoring devices at the target intersection, the traffic flow at the target intersection is not visible; therefore, the ID information data I... M It cannot be used to supplement traffic flow at the target intersection.

[0107] Therefore, in this embodiment, the vehicle driving data S in the trajectory mapping features is first... M Time feature data T M Channelization feature data L M The data is input into the autoencoder learning model. Then the encoder processes the vehicle driving data S... M Time feature data T M Channelization feature data L M The corresponding encoded information is output after multiple encoding processes. Finally, the corresponding encoded information is input to the decoder, which decodes the corresponding encoded information multiple times and outputs preliminary traffic completion data.

[0108] In step 204, the historical traffic data, the historical traffic completion data, and the preliminary traffic completion data are fused to obtain the corrected traffic data for non-target intersections.

[0109] Optionally, step 204 can be implemented in any of the embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0110] In step 205, target traffic completion data for the target intersection is obtained based on the spatial correlation features between the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data.

[0111] Optionally, step 205 can be implemented using any of the implementation methods in the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this, nor will they be described in detail.

[0112] The method provided in this embodiment integrates the trajectory mapping features between road network data and vehicle driving data, and makes full use of the vehicle trajectory mapping relationship between intersections and vehicle trajectories within the traffic area, thereby specifically solving the problem of missing historical traffic data for some non-target intersections and preliminary traffic data completion data for target intersections.

[0113] Third Embodiment

[0114] In order to specifically address the problem of missing target traffic flow data at the target intersection, the corrected traffic flow data from non-target intersections, the preliminary traffic flow data, and the trajectory mapping features in step 202 can be integrated to fully utilize the spatial relationship between the upstream and downstream non-target intersection traffic flow data and the target intersection.

[0115] In yet another embodiment proposed in this disclosure, such as Figure 4 As shown, the traffic flow data completion method may include, but is not limited to, the following steps.

[0116] In step 401, road network data and vehicle driving data within the traffic area are acquired, as well as historical traffic flow data measured at non-target intersections within the traffic area.

[0117] Optionally, step 401 can be implemented using any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0118] In step 402, based on the trajectory mapping features between the road network data and the vehicle driving data, historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections are obtained; wherein, the historical traffic flow completion data is used to represent historical traffic flow data that was not obtained at the non-target intersections.

[0119] Optionally, step 402 can be implemented in any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this, nor will they be described in detail.

[0120] In step 403, the historical traffic data, the historical traffic completion data, and the preliminary traffic completion data are fused to obtain the corrected traffic data for the non-target intersection.

[0121] Optionally, step 403 can be implemented using any of the implementation methods in the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this, nor will they be described in detail.

[0122] In step 404, a comprehensive spatial representation of the target traffic flow at the target intersection is generated based on the trajectory mapping features, the corrected traffic flow data, and the preliminary traffic flow completion data, so as to obtain the target traffic flow completion data.

[0123] The target traffic integrated spatial representation can integrate multiple spatial data sets and feature information to provide comprehensive spatial analysis and decision support. Obtaining the target traffic completion data through the integrated spatial representation generated from the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data improves data completeness, visualization, scalability, flexibility, and efficiency.

[0124] Specifically, such as Figure 5 As shown, including but not limited to the following steps.

[0125] In step 501, a spatial attention mechanism is used to generate an intersection traffic spatial relationship matrix based on the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data.

[0126] Optionally, in step 501, the preliminary traffic data and its corresponding spatiotemporal characteristics are supplemented: vehicle driving data S M Time feature data T M Channelization feature data L M The input is fed into the autoencoder model described in step 303, and after encoding, a preliminary restored and fused feature representation is generated.

[0127] The corrected traffic flow data for non-target intersections and their corresponding spatiotemporal characteristics: vehicle driving data S M Time feature data T M Channelization feature data L M ID information data I M The input is fed into the autoencoder model described in step 303, and the modified fusion feature representation is generated through encoding.

[0128] Next, in the attention network model, the initial restored fusion feature representation is used as the query vector to query and calculate the relationship or similarity with other elements, and the corrected fusion feature representation is used as the key vector to calculate the similarity between two elements, thus constructing a traffic space relationship matrix.

[0129] By employing a spatial attention mechanism, a spatial relationship matrix of intersection traffic flow is generated based on the corrected traffic flow data of non-target intersections, the preliminary traffic flow completion data, and the trajectory mapping features. This is beneficial for describing in detail the changes in traffic flow at each intersection over time and space, thereby discovering the spatiotemporal patterns of traffic flow, such as traffic distribution patterns, and thus more accurately predicting the target completion traffic flow data.

[0130] In step 502, the comprehensive spatial representation of intersection traffic flow is generated based on the intersection traffic flow spatial relationship matrix.

[0131] Optionally, in step 502, the weight features of the intersection traffic flow spatial representation are obtained through a normalization function, and the weight features are combined with the target value features for feature training to obtain the intersection traffic flow spatial representation.

[0132] Generating a comprehensive spatial representation of intersection traffic flow based on the intersection traffic flow spatial relationship matrix can more comprehensively reflect the traffic flow relationship between various intersections within the traffic area, providing a more accurate spatial reference for traffic control and optimization.

[0133] In step 503, the target complete traffic data is obtained based on the intersection traffic integrated spatial representation.

[0134] The target complete flow data is obtained by using the comprehensive spatial representation of traffic flow at intersections. By making full use of historical data and traffic flow patterns from other intersections, more accurate and comprehensive supplements are provided for missing flow data, thereby improving the effectiveness of traffic control and optimization.

[0135] Optionally, in step 503, the intersection traffic flow spatial representation is input into the autoencoder model described in step 303, and the target completed traffic flow data is obtained after decoding.

[0136] This embodiment fully utilizes the spatial relationship between upstream and downstream non-target intersection traffic data and target intersection traffic data by integrating the corrected traffic data of non-target intersections, the preliminary traffic data completion data, and the trajectory mapping features in step 202. This effectively solves the problem of missing target traffic data completion data at target intersections due to high equipment costs.

[0137] Figure 6 This is a flowchart illustrating a training method for a traffic flow data completion model provided in this embodiment of the disclosure. The network structure of the training model is as follows: Figure 7 As shown. It should be noted that the training method for the traffic flow data completion model in this embodiment can be executed by an electronic device. The traffic flow data completion model includes a data processing module, a first fusion module, and a second fusion module, as shown... Figure 6 As shown, the training method includes the following steps.

[0138] In step 601, road network data and vehicle driving data within the traffic area are acquired, as well as historical traffic flow data monitored at non-target intersections within the traffic area.

[0139] Optionally, step 601 can be implemented in any of the embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0140] In step 602, the trajectory mapping features between the road network data and the vehicle driving data are input to the data processing module to obtain the historical traffic flow completion data of the non-target intersection and the preliminary traffic flow completion data of the target intersection; wherein, the historical traffic flow completion data is used to represent the historical traffic flow data that was not obtained at the non-target intersection.

[0141] Optionally, in this embodiment, flow direction ID information features, time features, and road network channelization features are extracted first. Since traffic flow at the target intersection is invisible during training, the flow direction ID information features can only be used to reconstruct traffic flow at non-target intersections, and to learn the changing trends of different flow directions at various times. Time features mainly include information such as hour, date, and weekday type. Channelization features mainly include information such as the number of lanes.

[0142]

[0143]

[0144]

[0145] This indicates the embedded information leading to the ID, where Embedded information representing temporal features. This indicates the embedded information of the lane.

[0146] Then, for the trajectory sample information, a data processing module based on a residual network structure is designed to extract trajectory sample features from the vehicle driving data. This data processing module is used to implement skip links, enhancing the fluidity of the gradient. In this embodiment, the data processing module used is an embedded Residual Block.

[0147] like Figure 8 As shown, an embedded block specifically consists of the following modules:

[0148]

[0149]

[0150]

[0151] f(x) = max(0,x)

[0152] in, This represents the input information for the data processing module. `Linear` indicates a linear layer, and `f()` represents the activation function. A ReLU function is used for nonlinear transformation. This is the output information of the data processing module.

[0153] Next, the historical traffic data of non-target intersections is completed: combining ID information, time feature information, lane embedding information and trajectory information, through an Encoder layer and a Decoder layer, the prediction result of the historical traffic data is finally obtained.

[0154] The corresponding network structure is as follows:

[0155]

[0156]

[0157]

[0158] Both the Encoder and Decoder layers consist of multiple data processing modules.

[0159] Simultaneously, the preliminary traffic flow data for the target intersection is supplemented. Considering that the labels in the training data do not contain traffic flow data for the target intersection, the ID information cannot be fully trained. Therefore, the ID information cannot be used when supplementing the preliminary traffic flow data. Therefore, by combining time feature information, lane embedding information, and trajectory information, an Encoder layer and a Decoder layer are used to finally obtain the prediction result of the preliminary traffic flow data for the target intersection.

[0160] The corresponding network structure is shown below:

[0161]

[0162]

[0163]

[0164] In step 603, the historical traffic data, the historical traffic completion data, and the preliminary traffic completion data are input into the first fusion module to obtain the corrected traffic data for non-target intersections.

[0165] Optionally, in step 603, the historical traffic data, the historical traffic completion data, and the preliminary traffic completion data are input to the first fusion module to extract the corrected traffic map. The corrected traffic map is used to extract historical traffic data from non-target intersections and correct any missing historical traffic data. In step 603, the corrected traffic map is the corrected traffic data for non-target intersections.

[0166] For non-target intersections, historical traffic flow data is used when there is no missing traffic flow data; when traffic flow is perceived to be missing, historical traffic flow data is used. The restoration results; for the target intersection, the following method is used. The restoration result.

[0167] Extract the preliminary completed traffic flow map of the target intersection. The preliminary traffic flow completion data in step 603 is the preliminary completed traffic flow map of the target intersection. The preliminary completed traffic flow map of the target intersection refers to the preliminary reconstruction of the traffic flow of all target intersections. Specifically, the traffic flow of all target intersections is... The restoration result.

[0168] In step 604, the corrected traffic flow data of the non-target intersection, the preliminary traffic flow completion data, and the trajectory mapping features are input to the second fusion module to obtain the target traffic flow completion data of the target intersection.

[0169] Optionally, in step 604, the obtained corrected flow map and the target preliminary completed flow map are combined with trajectory information, traffic flow information, time feature information, and lane embedding information and input into the second fusion module for encoding to obtain corrected fusion feature representations. Preliminary reconstruction and fusion feature representation of the target

[0170] Next, a spatial attention mechanism is used, with each intersection in the initial target traffic map as the query and the representation of the corrected traffic map of physically connected intersections as the key, to extract spatial correlation features.

[0171]

[0172] α ij =softmax j (e ij )

[0173]

[0174]

[0175] in This represents the one-stage reconstruction and fusion feature representation of intersection i. This represents the modified fusion feature representation of adjacent intersection j. As a query, As the key, an attention network is used to extract spatial feature information. q W k W v ,v T Represents the parameter matrix. e ij Indicates the intersection The correlation feature is used to obtain the weight feature α through the softmax() function. ij By combining weight features and target value features, a spatial feature representation is obtained.

[0176] Finally, a decoder is used to obtain the final model output. Output target traffic to complete the data.

[0177]

[0178]

[0179] In step 605, the model loss value is calculated based on the historical traffic completion data, the preliminary traffic completion data, and the target traffic completion data.

[0180] Optionally, the historical traffic completion data, the preliminary traffic completion data, and the target traffic completion data are used as sample data, and the loss value of the sample data restoration is calculated by a multi-objective loss function, and the obtained loss value of the sample data restoration is used as the model loss value.

[0181] For example, the model uses the Huber function, which combines the characteristics of the MEA loss function and the MSE loss function. It closely approximates the absolute error when the error is large and has good robustness against outliers. The formula is shown below:

[0182]

[0183] The loss function of this method's model is: in This indicates the loss in restoring data by combining historical traffic with ID information. This indicates the loss in the initial traffic completion data restoration. This represents the loss in restoring the target traffic data.

[0184] These correspond to three metrics: non-target flow direction reconstruction, pure trajectory reconstruction, and spatial reconstruction. The loss values ​​for reconstructing the sample data are calculated using a multi-objective loss function. This facilitates the optimization and evaluation of the model's performance in the three aspects of flow direction, trajectory, and spatial relationships, allowing for more refined adjustments to optimize the model's performance across all three dimensions.

[0185] In step 606, a completion model for the traffic flow data is trained based on the model loss value.

[0186] Optionally, an adaptive optimization algorithm is used to decompose the multi-objective loss function into multiple single-objective loss functions, and each single-objective loss function is optimized step by step to improve the convergence speed of model training.

[0187] For example, in this embodiment, a step-by-step optimization training method is used when optimizing the multi-objective loss function. Step-by-step optimization refers to decomposing the multi-objective loss function into individual loss functions and optimizing these loss functions separately at different stages or steps.

[0188] By adopting stepwise optimization, each individual objective may converge to a local optimum more easily, avoiding conflicts and interference between objectives. This is conducive to more stable and controllable training, accelerating the convergence speed of model training, and thus generating target traffic flow completion data for target intersections more promptly. This reduces the error between the generated target traffic flow completion data and the real-time traffic flow data of the target intersections, providing a more effective reference for optimizing traffic management strategies.

[0189] Therefore, in this embodiment, when optimizing the model, first optimizes the loss of restoring historical traffic completion data, then optimizes the loss of restoring preliminary traffic completion data, and finally optimizes the loss of restoring target traffic completion data.

[0190] The training method provided in this embodiment solves the problem of missing real-time traffic flow data caused by high equipment costs, providing more complete data support for the optimization of traffic control systems. It offers a more accurate and reliable data foundation for transportation departments and urban traffic managers, helping to formulate more effective traffic management strategies and decisions.

[0191] Fifth embodiment

[0192] This embodiment also provides a training method for a traffic flow data completion model. The traffic flow data completion model includes a data processing module, a first fusion module, and a second fusion module, such as... Figure 9 As shown, the training method includes the following steps.

[0193] In step 901, road network data and vehicle driving data within the traffic area are acquired, as well as historical traffic flow data monitored at non-target intersections within the traffic area.

[0194] Optionally, step 901 can be implemented in any of the embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0195] In step 902, the trajectory mapping features between the road network data and the vehicle driving data are extracted by the data processing module.

[0196] Optionally, step 902 can be implemented in any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this, nor will they be described in detail.

[0197] In step 903, in the data processing module, the corresponding trajectory mapping features are encoded based on a pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data.

[0198] Optionally, in step 903, in the data processing module, the corresponding trajectory mapping features are input into the autoencoder learning model; the autoencoder learning model includes an encoder and a decoder.

[0199] The encoder outputs the corresponding encoded information after it has sequentially encoded the corresponding trajectory mapping features multiple times.

[0200] The corresponding encoded information is input to the decoder, which decodes the corresponding encoded information multiple times in sequence and outputs the historical traffic completion data and the preliminary traffic completion data accordingly.

[0201] In step 904, the historical traffic data, the historical traffic completion data, and the preliminary traffic completion data are input into the first fusion module to obtain the corrected traffic data for non-target intersections.

[0202] Optionally, step 904 can be implemented in any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0203] In step 905, the corrected traffic flow data of the non-target intersection, the preliminary traffic flow completion data, and the trajectory mapping features are input to the second fusion module to obtain the target traffic flow completion data of the target intersection.

[0204] Optionally, step 905 can be implemented in any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0205] In step 906, the model loss value is calculated based on the historical traffic completion data, the preliminary traffic completion data, and the target traffic completion data.

[0206] Optionally, step 906 can be implemented in any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0207] In step 907, a completion model for the traffic flow data is trained based on the model loss value.

[0208] Optionally, step 907 can be implemented in any of the various embodiments of this disclosure. The embodiments provided in this disclosure do not limit this and will not be described in detail.

[0209] The method provided in this embodiment solves the problem of missing real-time traffic flow data caused by high equipment costs, providing more complete data support for the optimization of traffic control systems. It offers a more accurate and reliable data foundation for transportation departments and urban traffic managers, helping to formulate more effective traffic management strategies and decisions.

[0210] Sixth Embodiment

[0211] To implement the above embodiments, this disclosure also proposes a traffic flow data completion device 1000. Figure 10 This is a structural block diagram of a traffic flow data completion device provided in this disclosure. Figure 10 As shown, the traffic flow data completion device may include:

[0212] The acquisition module 1001 is used to acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area.

[0213] The generation module 1002 is used to obtain historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections based on the trajectory mapping features between the road network data and the vehicle driving data; wherein, the historical traffic flow completion data is used to represent historical traffic flow data that was not obtained at the non-target intersections.

[0214] The generation module 1002 includes:

[0215] The first generation unit 11 is used to extract the trajectory mapping features between the road network data and the vehicle driving data.

[0216] The first generation unit 11 includes:

[0217] Embedding subunit 11a is used to obtain the identity recognition features and time features of the vehicle driving data, as well as the channelization features of the road network data, through embedding learning;

[0218] The residual subunit 11b is used to obtain the trajectory features of the vehicle driving data through residual learning.

[0219] The second generation unit 12 is used to encode the corresponding trajectory mapping features based on a pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data.

[0220] The second generation unit 12 includes:

[0221] Input subunit 12a is used to input the corresponding trajectory mapping features into the autoencoder learning model; the autoencoder learning model includes an encoder and a decoder;

[0222] The encoding subunit 12b is used to obtain the corresponding encoding information output by the encoder after the encoder has sequentially encoded the corresponding trajectory mapping features multiple times;

[0223] The decoding subunit 12c is used to input the corresponding encoded information to the decoder to obtain the historical traffic completion data and the preliminary traffic completion data output by the decoder after the corresponding encoded information is decoded multiple times in sequence.

[0224] The correction module 1003 is used to fuse the historical traffic completion data, the preliminary traffic completion data and the historical traffic data to obtain the corrected traffic data for non-target intersections;

[0225] Optionally, the correction module 1003 is used to generate a comprehensive spatial representation of the target traffic flow at the target intersection based on the trajectory mapping features, the corrected traffic flow data, and the preliminary traffic flow completion data, and to obtain the target traffic flow completion data.

[0226] Optionally, the correction module 1003 first employs a spatial attention mechanism to generate an intersection traffic spatial relationship matrix based on the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data; then, based on the intersection traffic spatial relationship matrix, it generates the intersection traffic comprehensive spatial representation; and finally, based on the intersection traffic comprehensive spatial representation, it obtains the target completed traffic data.

[0227] The fusion module 1004 is used to obtain the target traffic completion data of the target intersection based on the spatial correlation features between the trajectory mapping features, the corrected traffic data and the preliminary traffic completion data.

[0228] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0229] The device provided in this embodiment solves the problem of missing real-time traffic flow data caused by high equipment costs, providing more complete data support for the optimization of traffic control systems. It provides a more accurate and reliable data foundation for transportation departments and urban traffic managers, helping to formulate more effective traffic management strategies and decisions.

[0230] Seventh Embodiment

[0231] To implement the above embodiments, this disclosure also proposes a training device 1100 for a traffic flow data completion model. For example... Figure 11 As shown, the training device may include:

[0232] The acquisition module 1101 is used to acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area.

[0233] The generation module 1102 is used to input the trajectory mapping features between the road network data and the vehicle driving data into the data processing module to obtain the historical traffic flow completion data of the non-target intersection and the preliminary traffic flow completion data of the target intersection; wherein, the historical traffic flow completion data is used to represent the historical traffic flow data that was not obtained at the non-target intersection.

[0234] Optionally, the generation module 1102 includes:

[0235] Extraction unit 21 is used to extract trajectory mapping features between the road network data and the vehicle driving data through the data processing module;

[0236] The encoding unit 22 is used in the data processing module to encode the corresponding trajectory mapping features based on a pre-trained autoencoder model, thereby obtaining the historical traffic completion data and the preliminary traffic completion data.

[0237] Optionally, the encoding unit 22 includes:

[0238] Input subunit 22a is used in the data processing module to input the corresponding trajectory mapping features into the autoencoder learning model; the autoencoder learning model includes an encoder and a decoder;

[0239] Encoding subunit 22b is used to obtain the corresponding encoding information output by the encoder after the encoder has sequentially encoded the corresponding trajectory mapping features multiple times;

[0240] The decoding subunit 22c is used to input the corresponding encoded information to the decoder, and the decoder decodes the corresponding encoded information multiple times in sequence, and outputs the historical traffic completion data and the preliminary traffic completion data accordingly.

[0241] The correction module 1103 is used to input the historical traffic data, the historical traffic completion data and the preliminary traffic completion data into the first fusion module to obtain the corrected traffic data for non-target intersections.

[0242] The completion module 1104 is used to input the corrected traffic flow data of the non-target intersection, the preliminary traffic flow completion data, and the trajectory mapping features into the second fusion module to obtain the target traffic flow completion data of the target intersection.

[0243] Calculation module 1105 is used to calculate the model loss value based on the historical traffic completion data, the preliminary traffic completion data, and the target traffic completion data;

[0244] The optimization module 1106 is used to train a completion model of the traffic flow data based on the model loss value.

[0245] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0246] The device provided in this embodiment solves the problem of missing real-time traffic flow data caused by high equipment costs, providing more complete data support for the optimization of traffic control systems. It provides a more accurate and reliable data foundation for transportation departments and urban traffic managers, helping to formulate more effective traffic management strategies and decisions.

[0247] Eighth embodiment

[0248] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0249] Figure 12 A schematic block diagram of an example electronic device 1200 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0250] like Figure 12 As shown, device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1202 or a computer program loaded from storage unit 1208 into random access memory (RAM) 1203. The RAM 1203 may also store various programs and data required for the operation of device 1200. The computing unit 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. Input / output (I / O) interface 1205 is also connected to bus 1204.

[0251] Multiple components in device 1200 are connected to I / O interface 1205, including: input unit 1206, such as keyboard, mouse, etc.; output unit 1207, such as various types of monitors, speakers, etc.; storage unit 1208, such as disk, optical disk, etc.; and communication unit 1209, such as network card, modem, wireless transceiver, etc. Communication unit 1209 allows device 1200 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0252] The computing unit 1201 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1201 performs the various methods and processes described above. For example, in some embodiments, the methods may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 1208. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1200 via ROM 1202 and / or communication unit 1209. When the computer program is loaded into RAM 1203 and executed by the computing unit 1201, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to perform methods by any other suitable means (e.g., by means of firmware).

[0253] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0254] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0255] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0256] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0257] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0258] Computer systems may include clients and servers. Clients and servers are generally geographically separated and typically interact via a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. It should be understood that steps can be rearranged, added, or deleted using the various forms of processes shown above. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure are achieved, and this is not limited herein.

[0259] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for completing traffic flow data, comprising: Acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area; Based on the trajectory mapping characteristics between the road network data and the vehicle driving data, historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections are obtained. The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection. The historical traffic data, the preliminary traffic data, and the historical traffic data are fused together to obtain corrected traffic data for non-target intersections; Based on the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data, a comprehensive spatial representation of the target traffic at the target intersection is generated to obtain the target traffic completion data. The step of generating a comprehensive spatial representation of the target traffic flow at the target intersection based on the trajectory mapping features, the corrected traffic flow data, and the preliminary traffic flow completion data, to obtain the target traffic flow completion data, includes: A spatial attention mechanism is used to generate an intersection traffic spatial relationship matrix based on the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data. Based on the intersection traffic spatial relationship matrix, the comprehensive spatial representation of intersection traffic is generated; The target traffic flow completion data is obtained based on the intersection traffic flow spatial representation.

2. The method of claim 1, wherein, The process of obtaining historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections based on the trajectory mapping features between the road network data and the vehicle driving data includes: Extract the trajectory mapping features between the road network data and the vehicle driving data; The trajectory mapping features are encoded based on a pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data.

3. The method of claim 2, wherein, The extraction of trajectory mapping features between the road network data and the vehicle driving data includes: The vehicle driving data is obtained through embedding learning, which provides identification and temporal features, as well as channelization features of the road network data. The trajectory features of the vehicle driving data are obtained through residual learning.

4. The method of claim 2, wherein, The process of encoding the corresponding trajectory mapping features based on a pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data includes: The corresponding trajectory mapping features are input into an autoencoder learning model; the autoencoder learning model includes an encoder and a decoder. Obtain the corresponding encoded information output by the encoder after it has sequentially encoded the corresponding trajectory mapping features multiple times; The corresponding encoded information is input into the decoder to obtain the historical traffic completion data and the preliminary traffic completion data output by the decoder after the corresponding encoded information is decoded multiple times in sequence.

5. A method for training a traffic flow data completion model, the traffic flow data completion model comprising a data processing module, a first fusion module and a second fusion module, wherein, The training method includes: Acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area; The trajectory mapping features between the road network data and the vehicle driving data are input into the data processing module to obtain the historical traffic flow completion data of the non-target intersection and the preliminary traffic flow completion data of the target intersection; The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection. The historical traffic data, the historical traffic completion data, and the preliminary traffic completion data are input into the first fusion module to obtain the corrected traffic data for the non-target intersection; The corrected traffic flow data of the non-target intersection, the preliminary traffic flow completion data, and the trajectory mapping features are input into the second fusion module to obtain the target traffic flow completion data of the target intersection; The historical traffic completion data, the preliminary traffic completion data, and the target traffic completion data are used as sample data. The loss value of the sample data restoration is calculated by a multi-objective loss function, and the obtained loss value of the sample data restoration is used as the model loss value. Based on the model loss value, train a completion model for the traffic flow data; The step of training the traffic flow data completion model based on the model loss value includes: using an adaptive optimization algorithm to decompose the multi-objective loss function into multiple single-objective loss functions, and optimizing each single-objective loss function step by step to improve the convergence speed of model training.

6. The method of claim 5, wherein, The step of inputting the trajectory mapping features between the road network data and the vehicle driving data into the data processing module to obtain historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections includes: The data processing module extracts the trajectory mapping features between the road network data and the vehicle driving data. In the data processing module, the corresponding trajectory mapping features are encoded based on a pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data.

7. The method of claim 6, wherein, In the data processing module, the corresponding trajectory mapping features are encoded based on a pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data, including: In the data processing module, the corresponding trajectory mapping features are input into the autoencoder learning model; the autoencoder learning model includes an encoder and a decoder; Obtain the corresponding encoded information output by the encoder after it has sequentially encoded the corresponding trajectory mapping features multiple times; The corresponding encoded information is input to the decoder, which decodes the corresponding encoded information multiple times in sequence and outputs the historical traffic completion data and the preliminary traffic completion data accordingly.

8. A traffic flow data completion device, comprising: The acquisition module is used to acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area. The generation module is used to obtain historical traffic flow completion data for non-target intersections and preliminary traffic flow completion data for target intersections based on the trajectory mapping features between the road network data and the vehicle driving data. The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection. The correction module is used to fuse the historical traffic completion data, the preliminary traffic completion data, and the historical traffic data to obtain corrected traffic data for non-target intersections. The fusion module is used to obtain the target traffic completion data of the target intersection based on the spatial correlation features between the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data; The correction module is used to generate a comprehensive spatial representation of the target traffic flow at the target intersection and obtain the target traffic flow completion data based on the trajectory mapping features, the corrected traffic flow data, and the preliminary traffic flow completion data; specifically, it is used for: A spatial attention mechanism is used to generate an intersection traffic spatial relationship matrix based on the trajectory mapping features, the corrected traffic data, and the preliminary traffic completion data. Based on the intersection traffic spatial relationship matrix, the comprehensive spatial representation of intersection traffic is generated; The target traffic flow completion data is obtained based on the intersection traffic flow spatial representation.

9. The apparatus of claim 8, wherein, The generation module includes: The first generation unit is used to extract trajectory mapping features between the road network data and the vehicle driving data; The second generation unit is used to encode the corresponding trajectory mapping features based on a pre-trained autoencoder model to obtain the historical traffic completion data and the preliminary traffic completion data.

10. The apparatus of claim 9, wherein, The first generation unit includes: The embedding subunit is used to obtain the identity recognition features and time features of the vehicle driving data, as well as the channelization features of the road network data, through embedding learning; The residual sub-unit is used to obtain the trajectory features of the vehicle driving data through residual learning.

11. The apparatus of claim 9, wherein, The second generation unit includes: An input subunit is used to input the corresponding trajectory mapping features into an autoencoder learning model; the autoencoder learning model includes an encoder and a decoder. The encoding subunit is used to obtain the corresponding encoding information output by the encoder after it has sequentially encoded the corresponding trajectory mapping features multiple times. The decoding subunit is used to input the corresponding encoded information to the decoder to obtain the historical traffic completion data and the preliminary traffic completion data output by the decoder after the corresponding encoded information is decoded multiple times in sequence.

12. A device for training a traffic flow data completion model, the traffic flow data completion model comprising a data processing module, a first fusion module and a second fusion module, wherein, The training device includes: The acquisition module is used to acquire road network data and vehicle driving data within the traffic area, as well as historical traffic flow data monitored at non-target intersections within the traffic area. The generation module is used to input the trajectory mapping features between the road network data and the vehicle driving data into the data processing module to obtain the historical traffic flow completion data of the non-target intersection and the preliminary traffic flow completion data of the target intersection; The historical traffic data completion data is used to represent historical traffic data that was not obtained at the non-target intersection. The correction module is used to input the historical traffic data, the historical traffic completion data, and the preliminary traffic completion data into the first fusion module to obtain the corrected traffic data for non-target intersections; The completion module is used to input the corrected traffic flow data of the non-target intersection, the preliminary traffic flow completion data, and the trajectory mapping features into the second fusion module to obtain the target traffic flow completion data of the target intersection; The calculation module is used to take the historical traffic completion data, the preliminary traffic completion data and the target traffic completion data as sample data, calculate the loss value of the sample data restoration through a multi-objective loss function, and use the obtained loss value of the sample data restoration as the model loss value. The optimization module is used to train a complete model of the traffic flow data based on the model loss value; and specifically, it is used to decompose the multi-objective loss function into multiple single-objective loss functions using an adaptive optimization algorithm, and optimize each single-objective loss function step by step to improve the convergence speed of model training.

13. The apparatus of claim 12, wherein, The generation module includes: The extraction unit is used to extract the trajectory mapping features between the road network data and the vehicle driving data through the data processing module; The encoding unit is used in the data processing module to encode the corresponding trajectory mapping features based on a pre-trained autoencoder model, thereby obtaining the historical traffic completion data and the preliminary traffic completion data.

14. The apparatus of claim 13, wherein, The encoding unit includes: An input subunit is used in the data processing module to input the corresponding trajectory mapping features into the autoencoder learning model; the autoencoder learning model includes an encoder and a decoder. The encoding subunit is used to obtain the corresponding encoding information output by the encoder after it has sequentially encoded the corresponding trajectory mapping features multiple times. The decoding subunit is used to input the corresponding encoded information to the decoder, and the decoder decodes the corresponding encoded information multiple times in sequence, and outputs the historical traffic completion data and the preliminary traffic completion data accordingly.

15. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-4, or to perform the method of any one of claims 5-7.

16. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-4, or to be able to perform the method according to any one of claims 5-7.

17. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-4, or is capable of performing the method according to any one of claims 5-7.

Images