Road network green wave line identification model training method and device and electronic equipment

Abstract

This application provides a training method, apparatus, and electronic device for a road network green wave route identification model, comprising: performing data preprocessing and spatiotemporal alignment processing on the initial multi-source time-series data of each sample road segment in each sample road network to obtain target multi-source time-series data of each sample road segment in each sample road network; constructing a spatiotemporal map of each sample road network based on the target multi-source time-series data of each sample road segment in each sample road network; and training a road network green wave route identification model based on the spatiotemporal map of each sample road network, the actual green wave traffic tags of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. This solution can train a road network green wave route identification model that can accurately identify green wave routes based only on the collected incomplete initial multi-source time-series data, thereby effectively adapting to real-world scenarios with incomplete traffic control data, and improving the practicality and applicability of the solution.

Description

Technical Field

[0001] This application relates to the field of intelligent urban traffic management technology, and more specifically, to a training method, device, and electronic equipment for a road network green wave route recognition model. Background Technology

[0002] Green wave control in urban road networks is a key technology for improving the efficiency of trunk road traffic and reducing vehicle delays. Its core is to coordinate the signal timing of upstream and downstream intersections so that vehicles can pass through multiple intersections continuously at a specified speed, thus creating a green wave effect.

[0003] Existing technologies mostly rely on complete intersection signal timing data collected by dedicated signal control systems, such as cycle, phase difference, and green wave ratio. The core logic is to determine the feasibility of green waves by calculating the phase difference matching degree of upstream and downstream intersections.

[0004] However, due to limitations in data security and management permissions, complete intersection signal timing data cannot be collected, and the different standards for data collection by different devices severely restrict the implementation of the above methods. Summary of the Invention

[0005] The purpose of this application is to address the shortcomings of the prior art by providing a training method, apparatus, and electronic device for a green wave line recognition model for road networks. This will enable the training of a green wave recognition model that can effectively adapt to scenarios with incomplete signal control data, thereby improving the practicality and applicability of the green wave recognition model and enhancing recognition accuracy.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide a training method for a road network green wave line identification model, including: Collect initial multi-source time-series data of each sample road segment in at least one sample road network. The initial multi-source time-series data includes at least: vehicle driving data corresponding to each historical sampling time, traffic flow data corresponding to each historical sampling time, sample road segment topology data, and sample intersection attribute information. Data preprocessing and spatiotemporal alignment processing are performed on the initial multi-source time series data of each sample road segment in each sample road network to obtain the target multi-source time series data of each sample road segment in each sample road network. Based on the target multi-source time-series data of each sample road segment in each sample road network, a spatiotemporal map of each sample road network is constructed. The spatiotemporal map is an integrated representation of the spatial topology, temporal dynamics and signal constraints of the sample road network. Based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, the road network green wave route identification model is trained; the road network green wave route identification model is used to predict the green wave traffic information of each road segment in the target road network and generate the green wave potential heat map of the target road network.

[0007] Optionally, the step of constructing a spatiotemporal map of each sample road network based on the target multi-source time-series data of each sample road segment in each sample road network includes: Based on the identifiers of each sample road segment and each sample intersection in the sample road network, a first node set is determined, which includes: sample road segment nodes and sample intersection nodes. Based on the connectivity between sample road segments, the traffic flow transfer relationship between sample road segments, and the signal coordination relationship between sample intersections, an edge set is determined; the edge set includes: road segment connectivity edges, traffic flow transfer edges, and signal coordination edges. Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample road segment node in the first node set; Based on the sample intersection attribute information in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample intersection node in the first node set; Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of the connected edges of each road segment in the edge set; Based on the vehicle driving data, traffic flow data and sample intersection attribute information corresponding to each historical sampling time in the target multi-source time series data of each sample road segment, a dynamic attribute matrix is ​​determined. The dynamic attribute matrix is ​​used to characterize the dynamic change information of the operating status of the sample road network at each historical sampling time. The spatiotemporal map of the sample road network is constructed based on the first node set, the edge set, the static attribute information of each sample road segment node in the first node set, the static attribute information of each sample intersection node in the first node set, the static attribute information of each road segment connected edge in the edge set, and the dynamic attribute matrix.

[0008] Optionally, the step of training the road network green wave route recognition model based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network includes: The spatiotemporal maps of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network are used as training data and input into the initial road network green wave route identification model. The initial road network green wave route identification model determines the spatiotemporal feature information of each sample road segment in each sample road network based on the spatiotemporal maps of each sample road network, and predicts the predicted green wave traffic information and the predicted vehicle speed data of each sample road segment in each sample road network based on the spatiotemporal feature information of each sample road segment in each sample road network. Based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, the loss information of the initial road network green wave line identification model is determined. Based on the loss information, the initial road network green wave route identification model is iteratively trained, and the intermediate road network green wave route identification model at the time of iteration stop is used as the road network green wave route identification model.

[0009] Optionally, determining the spatiotemporal characteristic information of each sample road segment in each sample road network based on the spatiotemporal map of each sample road network includes: The initial feature vector of the sample road segment is determined based on the static attribute information of the sample road segment nodes in the sample road network and the signal timing information of the sample intersection nodes associated with the sample road segment nodes. Based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segment, the spatial dependency coding result of the sample road segment is determined. Based on the spatial dependency coding result and the spatial location coding result of the sample road segment, the spatial feature information of the sample road segment is determined; Based on the dynamic attribute matrix in the spatiotemporal map of the sample road network, determine the temporal sequence of the sample road segment; Based on the time sequence of the sample road segment, determine the time-series coding result of the sample road segment; Based on the target sampling time of each time series data in the time series of the sample road segment and the mapping relationship between the sampling time and the preset peak marker, the periodic sensing location coding result of the sample road segment is determined. Based on the temporal coding results of the sample road segment and the periodic sensing location coding results of the sample road segment, the temporal feature information of the sample road segment is determined; Based on the spatial feature information and temporal feature information of the sample road segment, a fusion process is performed to obtain the spatiotemporal feature information of the sample road segment.

[0010] Optionally, determining the spatial dependency coding result of the sample road segment based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segments of the sample road segment includes: Based on the edge set and the first node set in the spatiotemporal graph of the sample road network, the associated road segments of the sample road segment are determined; Based on the initial feature vector of the sample road segment and the initial feature vector of each associated road segment, the attention weight information between the sample road segment and each associated road segment is determined. Based on the initial feature vectors of each associated road segment and the attention weight information between the sample road segment and each associated road segment, the spatial dependency coding result of the sample road segment is determined.

[0011] Optionally, the method for determining the spatial location coding result of the sample road segment includes: Based on the location information of the sample road segment, spatial location coding is performed on the sample road segment to obtain the spatial location coding result of the sample road segment.

[0012] Optionally, determining the time-series coding result of the sample road segment based on its time-series sequence includes: The time sequence of the sample road segment is input into the time encoder, which determines the query matrix, key matrix and value matrix respectively. Based on the query matrix, key matrix, and value matrix, the time-series coding result of the sample road segment is determined.

[0013] Optionally, determining the periodic sensing location coding result of the sample road segment based on the target sampling time of each time-series data in the time-series of the sample road segment and the mapping relationship between the sampling time and the preset peak marker includes: Based on the target sampling time of each time series data in the time series sequence of the sample road segment, determine the hour label of each time series data; The target peak marker for each time series data is determined based on the target sampling time of each time series data and the mapping relationship between the sampling time and the preset peak marker. Based on the hourly labels and target peak markers of each time series data, the sample road segment is fused using a multilayer perceptron to obtain the periodic sensing location coding result.

[0014] Optionally, the step of fusing the spatial feature information and temporal feature information of the sample road segment to obtain the spatiotemporal feature information of the sample road segment includes: Based on the spatial feature information and temporal feature information of the sample road segment, a bidirectional cross operation is performed to obtain the temporal feature information of spatial guidance and the spatial feature information of temporal guidance, respectively. Based on the spatial feature information, temporal feature information, temporal feature information of the spatial guidance, and spatial feature information of the temporal guidance of the sample road segment, residual fusion processing is performed to obtain the spatiotemporal feature information of the sample road segment.

[0015] Optionally, determining the loss information of the initial road network green wave route identification model based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network includes: Based on the predicted green wave traffic information of each sample road segment in each sample road network and the actual green wave traffic label of each sample road segment in each sample road network, the classification loss information is determined. Based on the predicted vehicle speed data of each sample road segment in each sample road network and the actual vehicle speed data of each sample road segment in each sample road network, the speed regression loss information is determined. Based on the actual green wave trunk data of each sample road network and the predicted green wave traffic information of each sample road segment in each sample road network, the line continuity loss information is determined. Based on the classification loss information, the speed regression loss information, and the line continuity loss information, the loss information of the initial road network green wave line identification model is determined.

[0016] Optionally, it also includes: Collect multi-source data of each road segment in the target road network at the current time; The spatiotemporal map of the target road network is determined based on the multi-source data; The spatiotemporal map of the target road network is input into the road network green wave route identification model. The road network green wave route identification model predicts and outputs the predicted green wave traffic information of each road segment in the target road network. Based on the predicted green wave traffic information of each road segment, a green wave potential heat map of the target road network is generated. Based on the predicted green wave traffic information of each road segment, at least one green wave line under the target road network is searched and determined, and the green wave line consists of at least two consecutive road segments.

[0017] Optionally, the step of searching and determining at least one green wave route under the target road network based on the predicted green wave traffic information of each road segment includes: Based on the predicted green wave traffic information for each road segment, determine the green wave screening threshold; Based on the green wave screening threshold, each road segment is first screened to obtain each road segment after the first screening. Based on each of the first-selected road segments, a directed green wave reachability graph is constructed. The directed green wave reachability graph includes a second set of nodes and a set of directed edges. The second set of nodes includes road segment nodes corresponding to each of the first-selected road segments. The set of directed edges includes edges formed by consecutive road segment nodes in the same direction in the second set of nodes, and the weight of each edge is determined based on the predicted green wave traffic information of two interconnected road segment nodes. Based on the directed green wave reachability map and the green wave continuity constraint, a full-domain search is performed to determine each green wave route under the target road network; the green wave continuity constraint includes: the difference between the predicted green wave traffic information of continuous road segments meets a preset threshold and the turning angle of each road segment meets a preset angle.

[0018] Optionally, it also includes: Obtain the attribute characteristics of each green wave route, which include at least: the green wave route's identifier, starting point coordinates, ending point coordinates, sequence of intersections along the route, recommended driving direction, predicted green wave bandwidth, and green wave reliability; Based on the attribute characteristics of each green wave line, the evaluation indicators for each green wave line are determined. The evaluation indicators include: traffic efficiency indicators, green wave quality indicators, and spatiotemporal stability indicators. Based on the evaluation indicators of each green wave line, determine the comprehensive evaluation indicators for each green wave line; Based on the comprehensive evaluation indicators of each green wave line, the green wave control evaluation result of the target road network is determined; the green wave control evaluation result is used to indicate the green wave control level of the target road network.

[0019] Secondly, this application embodiment also provides a training device for a road network green wave line identification model, used to implement the training method of the road network green wave line identification model in the first aspect above, including: the device includes: a data acquisition module, a processing module, a construction module and a training module; The acquisition module is used to acquire initial multi-source time-series data of each sample road segment in at least one sample road network. The initial multi-source time-series data includes at least: vehicle driving data corresponding to each historical sampling time, traffic flow data corresponding to each historical sampling time, sample road segment topology data, and sample intersection attribute information. The processing module is used to perform data preprocessing and spatiotemporal alignment processing on the initial multi-source time series data of each sample road segment in each sample road network to obtain the target multi-source time series data of each sample road segment in each sample road network. The construction module is used to construct a spatiotemporal map of each sample road network based on the target multi-source time-series data of each sample road segment in each sample road network. The spatiotemporal map is an integrated representation of the spatial topology, temporal dynamics and signal constraints of the sample road network. The training module is used to train the green wave route identification model of the road network based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. The green wave route identification model of the road network is used to predict the green wave traffic information of each road segment in the target road network and generate the green wave potential heat map of the target road network.

[0020] Optionally, the construction module is specifically used to determine a first node set based on the identifiers of each sample road segment and each sample intersection in the sample road network. The first node set includes: sample road segment nodes and sample intersection nodes. Based on the connectivity between sample road segments, the traffic flow transfer relationship between sample road segments, and the signal coordination relationship between sample intersections, an edge set is determined; the edge set includes: road segment connectivity edges, traffic flow transfer edges, and signal coordination edges. Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample road segment node in the first node set; Based on the sample intersection attribute information in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample intersection node in the first node set; Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of the connected edges of each road segment in the edge set; Based on the vehicle driving data, traffic flow data and sample intersection attribute information corresponding to each historical sampling time in the target multi-source time series data of each sample road segment, a dynamic attribute matrix is ​​determined. The dynamic attribute matrix is ​​used to characterize the dynamic change information of the operating status of the sample road network at each historical sampling time. The spatiotemporal map of the sample road network is constructed based on the first node set, the edge set, the static attribute information of each sample road segment node in the first node set, the static attribute information of each sample intersection node in the first node set, the static attribute information of each road segment connected edge in the edge set, and the dynamic attribute matrix.

[0021] Optionally, the training module is specifically used to input the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network as training data into the initial road network green wave line recognition model. The initial road network green wave line recognition model determines the spatiotemporal feature information of each sample road segment in each sample road network based on the spatiotemporal map of each sample road network, and predicts the predicted green wave traffic information and the predicted vehicle speed data of each sample road segment in each sample road network based on the spatiotemporal feature information of each sample road segment in each sample road network. Based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, the loss information of the initial road network green wave line identification model is determined. Based on the loss information, the initial road network green wave route identification model is iteratively trained, and the intermediate road network green wave route identification model at the time of iteration stop is used as the road network green wave route identification model.

[0022] Optionally, the training module is specifically used to determine the initial feature vector of the sample road segment based on the static attribute information of the sample road segment nodes in the sample road network and the signal timing information of the sample intersection nodes associated with the sample road segment nodes. Based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segment, the spatial dependency coding result of the sample road segment is determined. Based on the spatial dependency coding result and the spatial location coding result of the sample road segment, the spatial feature information of the sample road segment is determined; Based on the dynamic attribute matrix in the spatiotemporal map of the sample road network, determine the temporal sequence of the sample road segment; Based on the time sequence of the sample road segment, determine the time-series coding result of the sample road segment; Based on the target sampling time of each time series data in the time series of the sample road segment and the mapping relationship between the sampling time and the preset peak marker, the periodic sensing location coding result of the sample road segment is determined. Based on the temporal coding results of the sample road segment and the periodic sensing location coding results of the sample road segment, the temporal feature information of the sample road segment is determined; Based on the spatial feature information and temporal feature information of the sample road segment, a fusion process is performed to obtain the spatiotemporal feature information of the sample road segment.

[0023] Optionally, the training module is specifically used to determine the associated road segments of the sample road segment based on the edge set and the first node set in the spatiotemporal graph of the sample road network; Based on the initial feature vector of the sample road segment and the initial feature vector of each associated road segment, the attention weight information between the sample road segment and each associated road segment is determined. Based on the initial feature vectors of each associated road segment and the attention weight information between the sample road segment and each associated road segment, the spatial dependency coding result of the sample road segment is determined.

[0024] Optionally, the training module is specifically used to perform spatial location encoding on the sample road segment based on the location information of the sample road segment, so as to obtain the spatial location encoding result of the sample road segment.

[0025] Optionally, the training module is specifically used to input the time sequence of the sample road segment into the time encoder, and the time encoder determines the query matrix, key matrix and value matrix respectively; Based on the query matrix, key matrix, and value matrix, the time-series coding result of the sample road segment is determined.

[0026] Optionally, the training module is specifically used to determine the hour label of each time series data according to the target sampling time of each time series data in the time series sequence of the sample road segment; The target peak marker for each time series data is determined based on the target sampling time of each time series data and the mapping relationship between the sampling time and the preset peak marker. Based on the hourly labels and target peak markers of each time series data, the sample road segment is fused using a multilayer perceptron to obtain the periodic sensing location coding result.

[0027] Optionally, the training module is specifically used to perform bidirectional cross-operation based on the spatial feature information and temporal feature information of the sample road segment to obtain spatial guidance temporal feature information and temporal guidance spatial feature information, respectively. Based on the spatial feature information, temporal feature information, temporal feature information of the spatial guidance, and spatial feature information of the temporal guidance of the sample road segment, residual fusion processing is performed to obtain the spatiotemporal feature information of the sample road segment.

[0028] Optionally, the training module is specifically used to determine classification loss information based on the predicted green wave traffic information of each sample road segment in each sample road network and the actual green wave traffic label of each sample road segment in each sample road network. Based on the predicted vehicle speed data of each sample road segment in each sample road network and the actual vehicle speed data of each sample road segment in each sample road network, the speed regression loss information is determined. Based on the actual green wave trunk data of each sample road network and the predicted green wave traffic information of each sample road segment in each sample road network, the line continuity loss information is determined. Based on the classification loss information, the speed regression loss information, and the line continuity loss information, the loss information of the initial road network green wave line identification model is determined.

[0029] Optionally, it may also include: a determination module and a prediction module; The acquisition module is also used to acquire multi-source data of each road segment in the target road network at the current time; The determining module is used to determine the spatiotemporal map of the target road network based on the multi-source data; The prediction module is used to input the spatiotemporal map of the target road network into the road network green wave line identification model, and the road network green wave line identification model predicts and outputs the predicted green wave traffic information of each road segment in the target road network, and generates a green wave potential heat map of the target road network based on the predicted green wave traffic information of each road segment. The determining module is used to search and determine at least one green wave line under the target road network based on the predicted green wave traffic information of each road segment. The green wave line consists of at least two consecutive road segments.

[0030] Optionally, the determining module is specifically used to determine the green wave screening threshold based on the predicted green wave traffic information of each road segment; Based on the green wave screening threshold, each road segment is first screened to obtain each road segment after the first screening. Based on each of the first-selected road segments, a directed green wave reachability graph is constructed. The directed green wave reachability graph includes a second set of nodes and a set of directed edges. The second set of nodes includes road segment nodes corresponding to each of the first-selected road segments. The set of directed edges includes edges formed by consecutive road segment nodes in the same direction in the second set of nodes, and the weight of each edge is determined based on the predicted green wave traffic information of two interconnected road segment nodes. Based on the directed green wave reachability map and the green wave continuity constraint, a full-domain search is performed to determine each green wave route under the target road network; the green wave continuity constraint includes: the difference between the predicted green wave traffic information of continuous road segments meets a preset threshold and the turning angle of each road segment meets a preset angle.

[0031] Optionally, the acquisition module is also used to acquire the attribute features of each green wave route, the attribute features including at least: the green wave route identifier, starting point coordinates, ending point coordinates, sequence of intersections along the route, recommended driving direction, predicted green wave bandwidth, and green wave reliability; The determining module is also used to determine the evaluation indicators of each green wave line based on the attribute characteristics of each green wave line. The evaluation indicators include: traffic efficiency indicators, green wave quality indicators, and spatiotemporal stability indicators. The determining module is also used to determine the comprehensive evaluation index of each green wave line based on the evaluation index of each green wave line. The determining module is further configured to determine the green wave control evaluation result of the target road network based on the comprehensive evaluation index of each green wave line; the green wave control evaluation result is used to indicate the green wave control level of the target road network.

[0032] Thirdly, embodiments of this application provide an electronic device, including: a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to implement the training method for the road network green wave line identification model provided in the first aspect.

[0033] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the training method for the road network green wave line identification model provided in the first aspect.

[0034] The beneficial effects of this application are: This application provides a training method, apparatus, and electronic device for a road network green wave route identification model, comprising: collecting initial multi-source time-series data of each sample road segment in at least one sample road network; performing data preprocessing and spatiotemporal alignment processing on the initial multi-source time-series data of each sample road segment in each sample road network to obtain target multi-source time-series data of each sample road segment in each sample road network; constructing a spatiotemporal map of each sample road network based on the target multi-source time-series data of each sample road segment in each sample road network; and training a road network green wave route identification model based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. This solution can collect training data without relying on a dedicated signal control system. It can train a road network green wave route identification model that can accurately identify green wave routes based solely on the incomplete initial multi-source time-series data collected by navigation software. This allows the solution to effectively adapt to real-world scenarios with incomplete signal control data, while improving the practicality and applicability of the solution.

[0035] Secondly, the road network green wave route identification model designed in this embodiment can achieve deep coupling of spatial topological features and temporal dynamic features through a two-way spatiotemporal cross-attention fusion mechanism, thereby obtaining spatiotemporal feature information with higher fusion accuracy. At the same time, the model can calculate multiple green wave potential quantification indicators based on spatiotemporal feature information through a uniquely designed four-way parallel decoding head, and output predicted green wave traffic information through indicator fusion, effectively improving the accuracy and efficiency of green wave route identification.

[0036] In addition, based on the trained road network green wave route identification model, the predicted green wave traffic information of each road segment in any road network can be accurately predicted. Based on the predicted green wave traffic information of each road segment, the green wave routes of the target road network can also be accurately identified. At the same time, combined with evaluation indicators, the green wave control level of the target road network can be effectively evaluated, providing better strategies for urban traffic control and providing precise support for green wave optimization. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 1 ; Figure 2 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 2 ; Figure 3 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 3 ; Figure 4 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 4 ; Figure 5 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 5 ; Figure 6 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 6 ; Figure 7 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 7 ; Figure 8 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 8 ; Figure 9 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 9 ; Figure 10 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 10 ; Figure 11 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 10 one; Figure 12 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 10 two; Figure 13 This is a schematic diagram of a training device for a road network green wave line recognition model according to an embodiment of this application; Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0040] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0041] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0042] Green wave control in urban road networks is a key technology for improving trunk road traffic efficiency and reducing vehicle delays. Its core is to coordinate signal timing at upstream and downstream intersections, allowing vehicles to pass through multiple intersections continuously at a specified speed, thus creating a green wave effect. Essentially, it achieves a synergistic match between spatial topological continuity and temporal traffic stability. Current green wave route identification and operation evaluation technologies have several limitations, specifically as follows: Existing technologies largely rely on complete intersection signal timing data (such as cycle, phase difference, and green ratio). The core logic is to determine the feasibility of green waves by calculating the phase difference matching degree of upstream and downstream intersections. However, in real-world scenarios, readily available internet-based traffic control data has significant shortcomings. This type of data mainly comes from various map providers and is generated by reverse calculation based on floating car trajectories. It can only obtain intersection signal cycle information and completely lacks information such as phase difference, resulting in severely incomplete data. It cannot directly determine the feasibility of green waves through phase difference calculation like traditional traffic control data. Furthermore, traditional traffic control data depends on the dedicated traffic control systems of each region. There are significant differences in the manufacturers, technical standards, and data formats of traffic control systems in different cities. Moreover, due to data security and management permissions, it is extremely difficult to obtain traditional traffic control data. It is impossible to achieve unified collection and aggregation of data from multiple cities across the country, and thus it is impossible to comprehensively evaluate the green wave control level of cities nationwide. This greatly limits the application scope of existing technologies and cannot meet the actual needs of green wave management and evaluation across the entire region and multiple cities.

[0043] Furthermore, existing green wave operation evaluations mostly focus on the traffic efficiency of a single arterial line, using only simple indicators such as travel time and number of stops. They lack a comprehensive quantitative assessment of the overall urban green wave control level, making it difficult to fully reflect the overall effectiveness and weaknesses of urban green wave management, and thus failing to provide accurate support for green wave optimization. Moreover, existing models are not specifically designed for green wave identification and evaluation tasks, failing to directly capture the spatial topological coherence and temporal traffic stability required for green wave formation. They also do not consider the practical constraints of incomplete traffic signal control data, making it difficult to reverse-engineer green wave routes from indirect traffic data, resulting in insufficient model applicability and identification accuracy.

[0044] Figure 1 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 1 The subject executing this method can be a computer device, such as... Figure 1 As shown, the method includes: S101. Collect initial multi-source time-series data of each sample road segment in at least one sample road network.

[0045] The initial multi-source time series data includes at least: vehicle driving data corresponding to each historical sampling time, traffic flow data corresponding to each historical sampling time, sample road segment topology data, and sample intersection attribute information.

[0046] A road network refers to a large network formed by connecting all the roads in a city. Each sample road network can correspond to a different city, meaning that sample road networks from different cities can be used as the source of training data.

[0047] Each sample road network is composed of multiple sample road segments connected together, and initial multi-source time-series data of each sample road segment in each sample road network can be collected. Specifically, the initial multi-source time-series data collected in this embodiment can be uploaded by navigation software, thereby avoiding dependence on a dedicated traffic control system.

[0048] When a vehicle traveling on a sample road segment uses navigation software, the navigation software's backend database will record historical data for each sample road segment, thus allowing the initial multi-source time-series data of each sample road segment in each sample road network to be extracted from the historical data.

[0049] The initial multi-source time series data includes multiple data points from each historical sampling time, arranged sequentially according to the sampling time. Among them, vehicle driving data includes, but is not limited to, vehicle speed and number of stops; sample road segment topology data includes, but is not limited to, road segment length and number of lanes; sample intersection attribute information includes, but is not limited to, intersection type, arterial road level, and intersection signal cycle.

[0050] S102. Perform data preprocessing and spatiotemporal alignment processing on the initial multi-source time series data of each sample road segment in each sample road network to obtain the target multi-source time series data of each sample road segment in each sample road network.

[0051] Data preprocessing methods include, but are not limited to: data cleaning, missing value repair, etc.

[0052] Data cleaning: For time-series data such as vehicle driving data and traffic flow data, outliers are removed using the 3σ criterion. Specifically, for any historical sampling time-series data x, if |x-μ|>3σ (μ is the data mean and σ is the data standard deviation), it is identified as an outlier and removed, and replaced with the mean of the data's time-series neighbors.

[0053] Assuming that the time series data x at the historical sampling time t is an outlier, we can calculate the mean of the time series data at time t-1 and time t+1 to replace x.

[0054] Missing value repair: Similarly, for vehicle driving data, traffic flow data, etc., interpolation methods based on spatiotemporal neighbors can be used for repair.

[0055] The calculation formula for the interpolation method is:

[0056] in, For road section The spatial neighbor set, i.e., road segment The upstream and downstream sections of the road, Historical sampling time The time-ordered neighbor set, The weights are determined by the spatiotemporal distance (the closer the spatial distance and the closer the time, the greater the weight).

[0057] Based on the results of data preprocessing, the initial multi-source time series data of each sample road segment in each sample road network can be unified to the same time granularity and spatial coordinate system, construct a spatiotemporally aligned dataset, and obtain the target multi-source time series data of each sample road segment in each sample road network, ensuring that the data of the same sample intersection or sample road segment at different historical sampling times can be correlated and compared.

[0058] S103. Based on the target multi-source time series data of each sample road segment in each sample road network, construct the spatiotemporal map of each sample road network.

[0059] The spatiotemporal map is an integrated representation of the spatial topology, temporal dynamics, and signal constraints of the sample road network.

[0060] Based on the target multi-source time-series data of each sample road segment in each sample road network obtained above, a spatiotemporal map of each sample road network can be constructed to represent the spatial topology, temporal dynamics and signal constraints of the sample road network in an integrated manner, so that the model can learn the conditions for green wave formation during subsequent model training.

[0061] S104. Based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, a road network green wave line recognition model is trained.

[0062] The road network green wave route identification model is used to predict the green wave traffic information of each road segment in the target road network and generate a heat map of the green wave potential of the target road network.

[0063] The actual green wave traffic label of the sample road segment is used to indicate whether the sample road segment belongs to the green wave segment or the non-green wave segment; the actual green wave trunk line data of the sample road network can refer to the number of green wave trunk lines; the actual vehicle speed data of the sample road segment can be taken as the average of the actual vehicle speeds of each vehicle on the sample road segment (here, each vehicle refers to vehicles using navigation software, those vehicles whose vehicle driving data can be collected through navigation software).

[0064] In some embodiments, the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network are used as training data to train a road network green wave line recognition model.

[0065] Based on the trained road network green wave route recognition model, the green wave route of any road network at any time can be identified, thereby guiding vehicles to pass through the green wave efficiently.

[0066] This embodiment can train a road network green wave route recognition model based solely on incomplete data collected by navigation software, eliminating the need for data collection from a dedicated signal control system. This effectively adapts to real-world scenarios with incomplete signal control data, enhancing the practicality and applicability of the solution.

[0067] In summary, the training method for the road network green wave route identification model provided in this embodiment includes: collecting initial multi-source time-series data of each sample road segment in at least one sample road network; performing data preprocessing and spatiotemporal alignment processing on the initial multi-source time-series data of each sample road segment in each sample road network to obtain target multi-source time-series data of each sample road segment in each sample road network; constructing a spatiotemporal map of each sample road network based on the target multi-source time-series data of each sample road segment in each sample road network; and training the road network green wave route identification model based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. This solution can collect training data without relying on a dedicated signal control system. It can train a road network green wave route identification model that can accurately identify green wave routes based solely on the incomplete initial multi-source time-series data collected by navigation software. This allows the solution to effectively adapt to real-world scenarios with incomplete signal control data, while improving the practicality and applicability of the solution.

[0068] Figure 2 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 2 In step S103, based on the target multi-source time-series data of each sample road segment in each sample road network, a spatiotemporal map of each sample road network is constructed, including: S201. Determine the first node set based on the identifiers of each sample road segment and each sample intersection in the sample road network.

[0069] The first set of nodes includes: sample road segment nodes and sample intersection nodes.

[0070] Definition of spatiotemporal graph: Abstracting the sample road network into a dynamic spatiotemporal graph structure. The first node set V contains sample road segment nodes. With sample intersection nodes Each node corresponds to unique spatial coordinates and attribute information.

[0071] The first node set can be generated based on the identifiers of each sample road segment and each sample intersection in the sample road network. Each sample road segment and each sample intersection uniquely corresponds to a node.

[0072] S202. Determine the edge set based on the connection relationship between each sample road segment, the traffic flow transfer relationship between each sample road segment, and the signal coordination relationship between each sample intersection.

[0073] The edge set includes: road segment connectivity edges, traffic flow transfer edges, and signal coordination edges. The edge set is used to describe the spatial connectivity relationships between nodes.

[0074] edge set It contains three types of edges: road segment connectivity edges, which connect adjacent road segment nodes; traffic flow transfer edges, which connect road segment nodes with traffic flow transfer relationships; and signal coordination edges, which connect intersection nodes with signal coordination relationships.

[0075] S203. Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample road segment node in the first node set.

[0076] Static attributes are embedded for each sample road segment node in the first node set. The static attribute information of the sample road segment is defined as follows: ;in, Indicates the length of the road segment; Indicates the number of lanes; Indicates the direction of travel; Indicates the road classification.

[0077] S204. Based on the sample intersection attribute information in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample intersection node in the first node set.

[0078] Static attributes are embedded for each sample intersection node in the first node set. The static attribute information of the sample intersection is defined as follows: ;in, Indicates the type of intersection; Indicates the trunk road classification of the intersection; This indicates the spatial coordinates of the intersection.

[0079] S205. Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of the connecting edges of each road segment in the edge set.

[0080] Static attributes are embedded into each segment-connected edge in the edge set. The static attribute information of the segment-connected edge is defined as follows: ;in, This represents the distance between adjacent road segments. Specifically, the distance between two adjacent road segments connected by the connecting edge of a road segment can be determined based on the road segment length in the sample road segment topology data.

[0081] S206. Determine the dynamic attribute matrix based on the vehicle driving data, traffic flow data, and sample intersection attribute information corresponding to each historical sampling time in the target multi-source time series data of each sample road segment.

[0082] The dynamic attribute matrix is ​​used to characterize the dynamic changes in the operating status of the sample road network at each historical sampling time.

[0083] The dynamic attribute matrix is ​​used to describe the dynamic changes in the road network's operating status at time t.

[0084] For any historical sampling time t, the dynamic attribute matrix is ​​represented as:

[0085] in, Represents the set of dynamic states of a node. , The average speed of vehicles on the road segment. Traffic volume on the road section This refers to the number of times a road segment stops or its congestion status. For the intersection signal cycle, , This represents the probability of traffic flow shift at historical sampling time t. This represents the signal coordination degree at historical sampling time t. , For the intersection The actual signal cycle difference between the intersection and the intersection j. For the intersection The optimal signal cycle difference between the intersection and the intersection j.

[0086] Dynamic attribute matrix It is continuously updated over time to realize the temporal expression of traffic operation status, traffic flow evolution process and signal control status.

[0087] S207. Based on the first node set, edge set, static attribute information of each sample road segment node in the first node set, static attribute information of each sample intersection node in the first node set, static attribute information of each road segment connected edge in the edge set, and dynamic attribute matrix, a spatiotemporal map of the sample road network is constructed.

[0088] Finally, based on the first set of nodes and edges mentioned above, and the embedded static attributes of nodes, static attributes of edges, and dynamic attribute matrices, a spatiotemporal graph of the sample road network can be constructed.

[0089] Figure 3 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 3 Optionally, in step S104, a road network green wave route recognition model is trained based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, including: S301. Input the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network as training data into the initial road network green wave route identification model. The initial road network green wave route identification model determines the spatiotemporal feature information of each sample road segment in each sample road network based on the spatiotemporal map of each sample road network, and predicts the predicted green wave traffic information and the predicted vehicle speed data of each sample road segment in each sample road network based on the spatiotemporal feature information of each sample road segment in each sample road network.

[0090] The initial road network green wave route recognition model has the same model architecture as the final road network green wave route recognition model after successful training; however, the network parameters of the initial road network green wave route recognition model have not yet been successfully trained.

[0091] After inputting the training data into the initial road network green wave route recognition model, iterative training begins. In any current iteration, the initial road network green wave route recognition model can construct the spatiotemporal feature information of each sample road segment in each sample road network based on the spatiotemporal map of each sample road network, and predict the predicted green wave traffic information and the predicted vehicle speed data of each sample road segment in each sample road network based on the spatiotemporal feature information of each sample road segment in each sample road network.

[0092] S302. Based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, determine the loss information of the initial road network green wave line identification model.

[0093] In this embodiment, the model's loss information can be constructed from multiple loss functions to comprehensively optimize the model from different dimensions. The predicted data and actual data mentioned above can be substituted into the corresponding loss functions to calculate the loss information of the road network green wave line identification model at the beginning of the current iteration.

[0094] S303. Based on the loss information, iteratively train the initial road network green wave route identification model, and use the intermediate road network green wave route identification model at the end of the iteration as the road network green wave route identification model.

[0095] If the loss information meets the preset loss threshold, the iteration can be stopped. Alternatively, the iteration can be stopped after the preset number of iterations has been reached. The intermediate road network green wave line recognition model at the time of iteration stop is then used as the trained road network green wave line recognition model for green wave line recognition processing in real-world scenarios.

[0096] Figure 4 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 4 Optionally, in step S301, based on the spatiotemporal map of each sample road network, the spatiotemporal characteristic information of each sample road segment in each sample road network is determined, including: S401. Determine the initial feature vector of the sample road segment based on the static attribute information of the sample road segment nodes in the sample road network and the signal timing information of the sample intersection nodes associated with the sample road segment nodes.

[0097] The road network green wave line identification model constructed in this embodiment may include a static topology embedding module, a multi-temporal state encoding module, a spatiotemporal cross-attention fusion module, and a green wave potential decoding head connected in sequence. Each module works together to realize the complete process of spatial feature extraction, temporal feature extraction, spatiotemporal feature fusion, and green wave potential decoding.

[0098] Static topology embedding module: used to extract the spatial features of road segments, characterizing the spatial connectivity of road segments and the spatial potential for green wave formation.

[0099] The following examples all use the calculation of spatial feature information, temporal feature information, and spatiotemporal feature information of any sample road segment as an example.

[0100] The static attribute information of the sample road segment nodes and the supplemented signal timing correlation features can be mapped into a low-dimensional dense initial feature vector, as shown in the formula: ,in For feature splicing, For signal timing-related features, MLP stands for Multilayer Perceptron, and the output dimension is... initial feature vector ,in, .

[0101] Among them, the supplemented signal timing related features can refer to the signal timing related information of the sample intersection nodes associated with the sample road segment nodes. The signal timing related information includes, but is not limited to, the signal cycle, signal start time, and signal end time of the intersection.

[0102] S402. Based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segment, determine the spatial dependency coding result of the sample road segment.

[0103] The associated road segments of a sample road segment can be determined based on the edge set in the spatiotemporal graph. The calculation of the initial feature vector of the associated road segments of a sample road segment can refer to the calculation of the initial feature vector of the sample road segment.

[0104] Based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segment, the spatial dependency relationship between the sample road segment and the associated road segment can be learned, and the spatial dependency encoding result of the sample road segment can be obtained.

[0105] S403. Based on the spatial dependency coding results and spatial location coding results of the sample road segments, determine the spatial feature information of the sample road segments.

[0106] Introducing spatial dependency coding results can enhance the model's perception of straight and continuous road segments. By concatenating the spatial dependency coding results of sample road segments with the spatial location coding results of sample road segments and performing layer normalization, the spatial feature information of sample road segments can be output.

[0107] S404. Determine the temporal sequence of the sample road segments based on the dynamic attribute matrix in the spatiotemporal map of the sample road network.

[0108] Multi-temporal state coding module: used to extract temporal features of sample road segments to characterize the traffic stability of road segments.

[0109] The time series sequence of a sample road segment can be extracted from the dynamic attribute matrix. The time series sequence includes multiple consecutive time steps, each corresponding to a historical sampling moment. Each time step includes multi-source data, which can include vehicle speed data, traffic flow data, and the number of stops. For example, the time series sequence of the sample road segment can be represented as follows: , This indicates vehicle speed data. This represents traffic flow data. Indicates the number of stops.

[0110] S405. Determine the timing coding result of the sample road segment based on the timing sequence of the sample road segment.

[0111] Temporal coding is performed on the temporal sequence of the sample road segments to capture the long-term temporal dependence and short-term fluctuations of traffic flow, and the temporal coding results of the sample road segments are obtained.

[0112] S406. Based on the target sampling time of each time series data in the time series of the sample road segment and the mapping relationship between the sampling time and the preset peak marker, determine the periodic sensing location coding result of the sample road segment.

[0113] Introducing periodic-aware location coding results enables the model to explicitly identify traffic periodic patterns at different times.

[0114] S407. Based on the temporal coding results of the sample road segments and the periodic sensing location coding results of the sample road segments, determine the temporal characteristic information of the sample road segments.

[0115] By concatenating the temporal coding results of the sample road segment with the periodic sensing location coding results of the sample road segment, and after layer normalization, the temporal feature information of the sample road segment can be output.

[0116] S408. Based on the spatial and temporal feature information of the sample road segments, perform fusion processing to obtain the spatiotemporal feature information of the sample road segments.

[0117] Spatiotemporal cross-attention fusion module: Through a bidirectional cross-attention mechanism, the spatial feature information and temporal feature information of the sample road segment can be deeply aligned. Based on residual fusion and normalization, the spatiotemporal feature information of the sample road segment can be output. The spatiotemporal feature information simultaneously includes the spatial topological coherence and temporal traffic stability of the sample road segment, comprehensively characterizing the green wave formation potential of the sample road segment.

[0118] Figure 5 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 5 Optionally, in step S402, the spatial dependency coding result of the sample road segment is determined based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segments, including: S501. Based on the edge set and the first node set in the spatiotemporal graph of the sample road network, determine the associated road segments of the sample road segments.

[0119] Based on the road segment connectivity edges in the edge set, the associated road segment nodes of the sample road segment can be determined. Based on the first node set, the identifier of the associated road segment node can be determined, thereby determining which specific road segment the associated road segment is.

[0120] S502. Based on the initial feature vector of the sample road segment and the initial feature vector of each associated road segment, determine the attention weight information between the sample road segment and each associated road segment.

[0121] An improved graph attention network can be used to learn the spatial dependencies between road segments. By adaptively allocating attention weights, the importance of trunk road segments and continuous straight road segments can be highlighted.

[0122] The formula for calculating the attention weight information between the sample road segment and the associated road segment is:

[0123] in, This represents the initial feature vector of the sample road segment. This represents the initial feature vector of the associated road segment j. Let k represent the initial feature vector of any associated road segment k. For attention weight vectors, Sample road sections The set of associated road segments, Indicates related road segments Relative to the sample road segment The importance of.

[0124] S503. Based on the initial feature vectors of each associated road segment and the attention weight information between the sample road segment and each associated road segment, determine the spatial dependency coding result of the sample road segment.

[0125] The formula for calculating the spatial dependency coding result of the sample road segment is as follows: That is, the initial feature vectors of each associated road segment are weighted and summed to obtain the spatial dependency coding result of the sample road segment.

[0126] Optionally, the spatial location coding result of the sample road segment is determined by: performing spatial location coding on the sample road segment based on its location information to obtain the spatial location coding result of the sample road segment.

[0127] The spatial location coding formula is:

[0128] in, Sample road sections Location coordinates on the main line Indexed by feature dimensions.

[0129] Based on this, the sample road section Spatial feature information The calculation formula is:

[0130] Figure 6 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 6Optionally, in step S405, the temporal coding result of the sample road segment is determined based on the temporal sequence of the sample road segment, including: S601. Input the time sequence of the sample road segment into the time encoder, and the time encoder determines the query matrix, key matrix and value matrix respectively.

[0131] In some embodiments, a Transformer encoder can be used for temporal encoding. The encoder can consist of multiple self-attention layers and one-dimensional convolutional layers (1D). After inputting the temporal sequence of the sample road segment into the Transformer encoder, the Transformer encoder can automatically determine the query matrix. Key matrix and value matrix .

[0132] The query matrix represents the traffic state features of the current time step to be analyzed; the key matrix represents the index of traffic state for all time steps, and its function is to match it with the query matrix to calculate the correlation between the current time step and a certain historical moment; the value matrix represents the real dynamic change information contained in the historical moment.

[0133] S602. Determine the time-series coding results of the sample road segments based on the query matrix, key matrix, and value matrix.

[0134] The formula for calculating the self-attention weight of the sample road segment is:

[0135] in, With the key matrix dimension as the reference, 1D convolutional layers are used to capture short-term traffic fluctuations, with a kernel size of 3.

[0136] The timing coding result is represented as follows: As in the aforementioned embodiments, This is the time sequence of the sample road segment, where the time-series coding result contains... .

[0137] Figure 7 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 7 In step S406, based on the target sampling time of each time-series data in the time-series sequence of the sample road segment and the mapping relationship between the sampling time and the preset peak marker, the periodic sensing location coding result of the sample road segment is determined, including: S701. Determine the hour label of each time series data according to the target sampling time of each time series data in the time series sequence of the sample road segment.

[0138] Based on the target sampling time of each time series data in the time series of the sample road segment, the hour label corresponding to each target sampling time can be determined. For example, if the target sampling time is 7:30, the hour label is determined to be 7; if the target sampling time is 13:20, the hour label is determined to be 13.

[0139] S702. Determine the target peak marker for each time series data based on the target sampling time of each time series data and the mapping relationship between the sampling time and the preset peak marker.

[0140] Different sampling times correspond to preset peak markers. For example, the peak period from 7:00 to 9:00 is predefined as the morning peak, from 9:00 to 17:00 as the off-peak, from 17:00 to 19:00 as the evening peak, and from 19:00 to 7:00 the next day as the night peak. Therefore, based on the time period range to which the target sampling time of the sample road segment belongs, the target peak marker for each time series data can be determined.

[0141] S703. Based on the hourly labels of each time series data and the target peak markers of each time series data, the data is fused using a multilayer perceptron to obtain the periodic sensing location coding results of the sample road segment.

[0142] The formula for periodic sensing position encoding is as follows:

[0143] in, Indicates hourly labels. Indicates the target peak marker. Indicates the target sampling time. Traffic cycle is 24 hours. This indicates a unique encoding.

[0144] Therefore, the periodic sensing location coding result of the sample road segment can be calculated. The periodic sensing location coding result contains the coding result corresponding to each time step in the time series data of the sample road segment.

[0145] Based on this, the temporal characteristic information of the sample road segments The calculation is as follows:

[0146] Figure 8 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 8 In step S408, the spatial and temporal feature information of the sample road segments are fused to obtain the spatiotemporal feature information of the sample road segments, including: S801. Based on the spatial feature information and temporal feature information of the sample road segments, perform bidirectional cross-operation to obtain the temporal feature information of spatial guidance and the spatial feature information of temporal guidance, respectively.

[0147] Temporal characteristics of spatial guidance Using spatial feature information as the query matrix Temporal feature information is used as the key matrix / Value matrix This allows for a focus on the temporal traffic stability of road sections with spatial advantages (such as trunk road sections). The core logic is that road sections with good spatial connectivity are more likely to form green waves if their temporal traffic is stable.

[0148] Temporally guided spatial feature information Using time-series feature information as the query matrix Spatial feature information as key matrix / Value matrix The focus is on the spatial connectivity of road sections with smooth temporal traffic. The core logic is that if road sections with smooth temporal traffic are spatially continuous, they are more likely to form green wave corridors.

[0149] S802. Based on the spatial feature information, temporal feature information, temporal feature information of spatial guidance, and spatial feature information of temporal guidance of the sample road segment, residual fusion processing is performed to obtain the spatiotemporal feature information of the sample road segment.

[0150] The fusion formula for spatial feature information, temporal feature information of sample road segments, temporal feature information of spatial guidance, and spatial feature information of temporal guidance is as follows:

[0151] in, This represents the spatiotemporal feature information obtained through fusion.

[0152] In some embodiments, the step of predicting the predicted green wave traffic information for each sample road segment in each sample road network based on the spatiotemporal characteristics of each sample road segment in each sample road network may include the following steps: The green wave potential decoder in the road network green wave line identification model can adopt a four-way parallel decoding structure to specifically decode green wave-related indicators, avoiding insufficient accuracy caused by general decoding. The four decoders correspond to the four core conditions for green wave formation, outputting their respective quantitative indicators: Traffic Smoothness Header: Outputs the traffic smoothness of sample road segments. The formula is: ,in For the weight vector, For bias, For the sigmoid function, The spatiotemporal characteristics of the sample road segments calculated above, .

[0153] Signal Coordination Matching Header: Outputs the signal coordination matching degree of the sample road segment. The formula is: Similarly, For the weight vector, For bias.

[0154] Traffic Flow Continuity Header: Outputs the traffic flow continuity of the sample road segment. The formula is: .

[0155] Green wave connectivity potential head: Outputs the green wave connectivity potential of the sample road segment. The formula is: .

[0156] By adaptively learning weights through the model and fusing the four decoding results, the predicted green wave traffic information (GWI) for each sample road segment can be obtained. The fusion formula is as follows: ,in The weights for adaptive learning of the model satisfy... , Indicates sample road segment The predicted green wave traffic information shows that the higher the GWI value, the greater the potential for a road segment to form a green wave.

[0157] Figure 9 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 9 Optionally, in step S302, based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, the loss information of the initial road network green wave line identification model is determined, including: S901. Based on the predicted green wave traffic information of each sample road segment in each sample road network and the actual green wave traffic label of each sample road segment in each sample road network, determine the classification loss information.

[0158] In this embodiment, the loss information is calculated using the following loss function formula:

[0159] in, To classify loss information, For velocity regression loss information, This is information about line continuity loss.

[0160] Classification loss information Cross-entropy loss can be used for calculation:

[0161] Where N represents the set of all sample road segments in the entire sample road network. Indicates sample road segment Predicted green wave traffic information, Indicates sample road segment The actual green wave traffic label, the actual green wave traffic label, the actual green wave traffic label takes the value of 0 or 1, 1 represents a green wave section, 0 represents a non-green wave section.

[0162] S902. Based on the predicted vehicle speed data of each sample road segment in each sample road network and the actual vehicle speed data of each sample road segment in each sample road network, determine the speed regression loss information.

[0163] Velocity regression loss information The mean square error loss can be used for calculation:

[0164] in, Indicates sample road segment Actual vehicle speed data Indicates sample road segment Predicted vehicle speed data.

[0165] S903. Based on the actual green wave trunk data of each sample road network and the predicted green wave traffic information of each sample road segment in each sample road network, determine the line continuity loss information.

[0166] Line continuity loss information L1 loss can be used for calculation:

[0167] in, The number of green wave trunk lines, For the first A collection of road sections along the green wave main road. For the first The average GWI value of the trunk line.

[0168] S904. Based on the classification loss information, speed regression loss information, and line continuity loss information, determine the loss information of the initial road network green wave line identification model.

[0169] The loss information is as follows:

[0170] in, Weighting coefficients (in this embodiment) , This is used to balance the training priorities of various supervised tasks. The training process employs the Adam optimizer with a learning rate of [missing information]. The iteration count is 100 rounds, until the loss information converges.

[0171] The road network green wave route identification model designed in this embodiment can achieve deep coupling of spatial topological features and temporal dynamic features through a two-way spatiotemporal cross-attention fusion mechanism, thereby obtaining spatiotemporal feature information with higher fusion accuracy. At the same time, the model can calculate multiple green wave potential quantification indicators based on spatiotemporal feature information through a uniquely designed four-way parallel decoding head, and output predicted green wave traffic information through indicator fusion, effectively improving the accuracy and efficiency of green wave route identification.

[0172] Next, the application and implementation of the green wave line identification model for the road network will be explained.

[0173] Figure 10 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 10 This method also includes: S1001. Collect multi-source data of each road segment in the target road network at the current time.

[0174] The target road network can be the road network of green wave lines to be predicted. The road network green wave line identification model can identify the green wave lines of the target road network at any time.

[0175] The data types contained in the multi-source data of each road segment at the current time are the same as those contained in the initial multi-source time-series data mentioned above. However, for the current time, the multi-source data is not time-series data, but data for a single moment.

[0176] S1002. Determine the spatiotemporal map of the target road network based on multi-source data.

[0177] The construction of the spatiotemporal map is described in steps S201-S207, and will not be repeated here.

[0178] S1003. Input the spatiotemporal map of the target road network into the road network green wave line identification model. The road network green wave line identification model predicts and outputs the predicted green wave traffic information of each road segment in the target road network. Based on the predicted green wave traffic information of each road segment, a green wave potential heat map of the target road network is generated.

[0179] The road network green wave route identification model can predict and output the predicted green wave traffic information (GWI) for each road segment in the target road network. Based on the predicted green wave traffic information of each road segment, a green wave potential heat map of the target road network can be output. Each pixel in the green wave potential heat map corresponds to the GWI value of a road segment, intuitively reflecting the green wave formation potential of all road segments in the target road network.

[0180] S1004. Based on the predicted green wave traffic information for each road segment, search for and determine at least one green wave line under the target road network.

[0181] A green wave route consists of at least two consecutive segments.

[0182] Based on the predicted green wave traffic information for each road segment, at least one high-potential green wave route under the target road network can be identified through multi-level filtering and searching.

[0183] Figure 11 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 10 Optionally, in step S1004, based on the predicted green wave traffic information for each road segment, at least one green wave route under the target road network is searched and determined, including: S1101. Determine the green wave screening threshold based on the predicted green wave traffic information for each road segment.

[0184] The green wave filtering threshold can be calculated based on the predicted green wave traffic information for each road segment:

[0185] in, Indicates the green wave filtering threshold. This represents the mean GWI of each road segment in the target road network. This represents the standard deviation of the GWI for each road segment.

[0186] S1102. Based on the green wave screening threshold, perform the first screening on each road segment to obtain each road segment after the first screening.

[0187] Based on the green wave screening threshold, GWI values ​​less than [a certain threshold] can be selected. Road sections with GWI values ​​higher than 0 were removed, and only those with GWI values ​​higher than 0 were retained. The road segments with high potential are selected as the first screening segments. That is, low-potential road segments are eliminated, and only high-potential road segments are retained to avoid subsequent omissions or misidentifications.

[0188] S1103. Based on each road segment after the first screening, construct a directed green wave reachability map.

[0189] The directed green wave reachability graph includes a second set of nodes and a set of directed edges. The second set of nodes includes the road segment nodes corresponding to each road segment after the first screening. The set of directed edges includes the edges formed by consecutive road segment nodes in the same direction in the second set of nodes, and the weight of each edge is determined according to the predicted green wave traffic information of two interconnected road segment nodes.

[0190] Construct a directed green wave reachable graph ,in, This represents the second set of nodes, including the nodes of each road segment after the first selection. This represents a set of directed edges, retaining only connections between consecutive road segment nodes in the same direction, and prohibiting unreasonable jumps across directions or main roads. The weight of each edge is the average of the GWIs of the two road segment nodes connected by the edge.

[0191] S1104. Based on the directed green wave reachability graph and the green wave continuity constraint, perform a global search to determine each green wave line under the target road network.

[0192] The green wave continuity constraints include: the difference between the predicted green wave traffic information of continuous road segments meets the preset threshold and the turning angle of each road segment meets the preset angle.

[0193] This embodiment employs a modified Dijkstra's Shortest Path Algorithm (Dijkstra). It uses the -GWI value as a weight, prioritizing the search of road segments with high GWI values, and searches for the longest continuous green wave route across the entire region. Simultaneously, a green wave continuity constraint is introduced: the GWI difference between continuous road segments... And the turning angle of the road section This prevents the searched routes from having breaks, sharp turns, or other issues that do not meet the requirements for green wave traffic. The core logic of the algorithm is as follows: 1. Initialize the distance array ,in This indicates that the initial distance is negative infinity, corresponding to the longest path, and the starting segment. ; 2. Use a priority queue (max heap) to store the road segments to be processed, and prioritize processing road segments with high GWI values; 3. For each current road segment Traverse its adjacent road segments If the continuity constraint is satisfied, and Then update and will Add to priority queue; 4. Iterate until the priority queue is empty, and output all the longest paths, which are the longest continuous green wave line, the second longest green wave line, and the green wave corridor.

[0194] For the searched similar green wave routes (segment overlap rate ≥ 80%), clustering and deduplication are performed using a hierarchical clustering algorithm, with the clustering distance being the Euclidean distance between the routes. , These are the segment GWI values ​​of the two lines respectively. After merging similar lines, the final output is all the green wave lines.

[0195] Figure 12 A flowchart illustrating the training method for the road network green wave line recognition model provided in this application embodiment. Figure 10 2; optionally, it also includes: S1201. Obtain the attribute characteristics of each green wave line.

[0196] The attribute features include at least: the green wave route identifier, the starting point coordinates, the ending point coordinates, the sequence of intersections along the route, the recommended driving direction, the predicted green wave bandwidth, and the green wave reliability.

[0197] The attribute characteristics of each green wave line can be obtained, including the predicted green wave bandwidth. , The minimum signal cycle at the intersection is given by the green wave reliability R = effective green wave duration / total duration.

[0198] S1202. Based on the attribute characteristics of each green wave line, determine the evaluation index of each green wave line.

[0199] The evaluation indicators include: traffic efficiency, green wave quality, and spatiotemporal stability.

[0200] Based on the attribute characteristics of each green wave route and the measured traffic data, a multi-dimensional evaluation index system is constructed. The quantitative formulas and meanings of each index are as follows: Traffic efficiency indicators specifically include: (1) Green wave coverage: , It is the total number of signalized intersections. This represents the number of intersections included in the green wave route. The higher the value, the more complete the green wave infrastructure.

[0201] (2) Continuous mileage, every 5 kilometers, the mileage without stopping. Used to characterize the effectiveness of the green wave, the larger the value, the smoother the flow.

[0202] (3) Average travel time along the green wave route: ,in This represents the number of road segments included in the green wave route. For road section The actual travel time The smaller the value, the higher the passage efficiency.

[0203] (4) Number of stops: , The smaller the value, the higher the passage efficiency.

[0204] (5) Parking rate: ,in This represents the total number of vehicles on the green wave route. The smaller the value, the higher the passage efficiency.

[0205] (6) Free Flow Velocity Achievement Rate: ,in The average speed of vehicles on the green wave line. For free-flow vehicle speed, The higher the value, the higher the traffic efficiency.

[0206] The specific green wave quality indicators include: (1) Green wave bandwidth utilization: ,in This is the actual green wave bandwidth. To predict green wave bandwidth, The higher the value, the better the quality of the green wave.

[0207] (2) Probability of green wave passing: ,in This represents the number of vehicles that continuously pass through all intersections within the green wave route. This represents the total number of vehicles on the green wave route. The higher the value, the better the quality of the green wave.

[0208] (3) Phase difference matching consistency: , For vehicles traveling in the direction of the green wave route from the section of road Drive to the section of road The actual time Indicates road segment and road sections The actual phase difference. The higher the value, the better the signal coordination effect.

[0209] Spatiotemporal stability indices specifically include: (1) Fluctuation index of green wave effect in different time periods: ,in The standard deviation of GWI values ​​within the time period. The mean of GWI values ​​over the time period. The smaller the value, the better the stability.

[0210] (2) Green wave effect uniformity between road segments: , The larger the value, the better the balance.

[0211] S1203. Based on the evaluation indicators of each green wave line, determine the comprehensive evaluation indicators of each green wave line.

[0212] The evaluation indicators for each green wave line can be standardized to [0, 100] using a linear normalization method. For positive indicators (the larger the value, the better), the normalization formula is: For negative indicators (the smaller the value, the better), the normalization formula is: ,in , These are the maximum and minimum values ​​of the indicator, respectively.

[0213] The comprehensive evaluation index of a single green wave line is calculated using a weighted scoring method, and the formula is as follows: ,in Assigning weights to each indicator. For the total number of indicators, The scores are the standardized scores for each indicator.

[0214] Green wave lines can be classified into different levels based on comprehensive evaluation indicators, such as: Excellent (…). ≥90), Good (80≤) <90), General (60≤) <80), Poor ( <60).

[0215] S1204. Based on the comprehensive evaluation indicators of each green wave line, determine the green wave control evaluation results of the target road network.

[0216] The green wave control evaluation results are used to indicate the level of green wave control in the target road network.

[0217] The evaluation results of green wave control for the target road network are as follows:

[0218] in, For the first Comprehensive evaluation indicators for green wave lines, For the first The length of the green wave line, This is the evaluation result of green wave control for the target road network. The higher the value, the better the city's green wave management level.

[0219] In some embodiments, the green wave control evaluation results of the target road network can be used as a basis. Locate weak sections and uncoordinated intersections in green wave control, and analyze the causes of the problems, which mainly include: signal timing error due to incomplete signal control data, unreasonable signal timing (low phase difference matching degree), road network topology limitations (multiple road turns and poor continuity), and excessive traffic flow fluctuations.

[0220] For different problems, provide targeted optimization and adjustment suggestions, such as: Issues with incomplete traffic control data: Optimize the traffic control data supplementation algorithm, increase matching dimensions for similar intersections (such as intersection type and time-of-day traffic similarity), and improve supplementation accuracy.

[0221] Issues with unreasonable signal timing: Adjust the signal cycle and phase difference at upstream and downstream intersections based on the matching degree of green wave traffic information (GWI) and phase difference.

[0222] Road network topology constraints: Continuous straight road sections should be prioritized as green wave trunk lines, while sections with many turns and dense intersections should be avoided.

[0223] Traffic flow fluctuation problem: Based on the prediction results of time series features, the green wave bandwidth and recommended driving speed are dynamically adjusted to improve the stability of the green wave.

[0224] In summary, the training method for the road network green wave route identification model provided in this embodiment includes: collecting initial multi-source time-series data of each sample road segment in at least one sample road network; performing data preprocessing and spatiotemporal alignment processing on the initial multi-source time-series data of each sample road segment in each sample road network to obtain target multi-source time-series data of each sample road segment in each sample road network; constructing a spatiotemporal map of each sample road network based on the target multi-source time-series data of each sample road segment in each sample road network; and training the road network green wave route identification model based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. This solution can collect training data without relying on a dedicated signal control system. It can train a road network green wave route identification model that can accurately identify green wave routes based solely on the incomplete initial multi-source time-series data collected by navigation software. This allows the solution to effectively adapt to real-world scenarios with incomplete signal control data, while improving the practicality and applicability of the solution.

[0225] Secondly, the road network green wave route identification model designed in this embodiment can achieve deep coupling of spatial topological features and temporal dynamic features through a two-way spatiotemporal cross-attention fusion mechanism, thereby obtaining spatiotemporal feature information with higher fusion accuracy. At the same time, the model can calculate multiple green wave potential quantification indicators based on spatiotemporal feature information through a uniquely designed four-way parallel decoding head, and output predicted green wave traffic information through indicator fusion, effectively improving the accuracy and efficiency of green wave route identification.

[0226] In addition, based on the trained road network green wave route identification model, the predicted green wave traffic information of each road segment in any road network can be accurately predicted. Based on the predicted green wave traffic information of each road segment, the green wave routes of the target road network can also be accurately identified. At the same time, combined with evaluation indicators, the green wave control level of the target road network can be effectively evaluated, providing better strategies for urban traffic control and providing precise support for green wave optimization.

[0227] The following describes the apparatus, equipment, and storage medium used to implement the training method for the road network green wave line identification model provided in this application. The specific implementation process and technical effects are described above and will not be repeated below.

[0228] Figure 13 This is a schematic diagram of the training device for the road network green wave line recognition model according to an embodiment of this application, as shown below. Figure 13 As shown, the device includes: a data acquisition module 100, a processing module 200, a construction module 300, and a training module 400; The acquisition module 100 is used to acquire initial multi-source time series data of each sample road segment in at least one sample road network. The initial multi-source time series data includes at least: vehicle driving data corresponding to each historical sampling time, traffic flow data corresponding to each historical sampling time, sample road segment topology data, and sample intersection attribute information. The processing module 200 is used to perform data preprocessing and spatiotemporal alignment processing on the initial multi-source time series data of each sample road segment in each sample road network to obtain the target multi-source time series data of each sample road segment in each sample road network. The construction module 300 is used to construct the spatiotemporal map of each sample road network based on the target multi-source time series data of each sample road segment in each sample road network. The spatiotemporal map is an integrated representation of the spatial topology, temporal dynamics and signal constraints of the sample road network. Training module 400 is used to train a road network green wave route identification model based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. The road network green wave route identification model is used to predict the green wave traffic information of each road segment in the target road network and generate a green wave potential heat map of the target road network.

[0229] Optionally, the construction module 300 is specifically used to determine the first node set based on the identifiers of each sample road segment and each sample intersection in the sample road network. The first node set includes: sample road segment nodes and sample intersection nodes. Based on the connectivity between sample road segments, the traffic flow transfer relationship between sample road segments, and the signal coordination relationship between sample intersections, the edge set is determined; the edge set includes: road segment connectivity edges, traffic flow transfer edges, and signal coordination edges. Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample road segment node in the first node set; Based on the sample intersection attribute information in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample intersection node in the first node set; Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of the connected edges of each road segment in the edge set; Based on the vehicle driving data, traffic flow data and sample intersection attribute information corresponding to each historical sampling time in the target multi-source time series data of each sample road segment, a dynamic attribute matrix is ​​determined. The dynamic attribute matrix is ​​used to characterize the dynamic changes in the operating status of the sample road network at each historical sampling time. Based on the first node set, edge set, static attribute information of each sample road segment node in the first node set, static attribute information of each sample intersection node in the first node set, static attribute information of each road segment connected edge in the edge set, and dynamic attribute matrix, a spatiotemporal graph of the sample road network is constructed.

[0230] Optionally, the training module 400 is specifically used to input the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network as training data into the initial road network green wave line recognition model. The initial road network green wave line recognition model determines the spatiotemporal feature information of each sample road segment in each sample road network based on the spatiotemporal map of each sample road network, and predicts the predicted green wave traffic information and the predicted vehicle speed data of each sample road segment in each sample road network based on the spatiotemporal feature information of each sample road segment in each sample road network. Based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, the loss information of the initial road network green wave line identification model is determined. Based on the loss information, the initial road network green wave route identification model is trained iteratively, and the intermediate road network green wave route identification model at the end of the iteration is used as the road network green wave route identification model.

[0231] Optionally, the training module 400 is specifically used to determine the initial feature vector of the sample road segment based on the static attribute information of the sample road segment nodes in the sample road network and the signal timing information of the sample intersection nodes associated with the sample road segment nodes. Based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segment, the spatial dependency coding result of the sample road segment is determined. Based on the spatial dependency coding results and spatial location coding results of the sample road segments, the spatial feature information of the sample road segments is determined. Based on the dynamic attribute matrix in the spatiotemporal map of the sample road network, the temporal sequence of the sample road segments is determined; Based on the time sequence of the sample road segment, determine the time sequence coding result of the sample road segment; Based on the target sampling time of each time series data in the time series of the sample road segment and the mapping relationship between the sampling time and the preset peak marker, the periodic sensing location coding result of the sample road segment is determined. Based on the temporal coding results and the periodic sensing location coding results of the sample road segments, the temporal feature information of the sample road segments is determined. Based on the spatial and temporal characteristics of the sample road segments, a fusion process is performed to obtain the spatiotemporal characteristics of the sample road segments.

[0232] Optionally, the training module 400 is specifically used to determine the associated road segments of the sample road segments based on the edge set and the first node set in the spatiotemporal graph of the sample road network. Based on the initial feature vectors of the sample road segments and the initial feature vectors of each associated road segment, the attention weight information between the sample road segments and each associated road segment is determined. Based on the initial feature vectors of each associated road segment and the attention weight information between the sample road segment and each associated road segment, the spatial dependency coding result of the sample road segment is determined.

[0233] Optionally, the training module 400 is specifically used to perform spatial location encoding on the sample road segments based on their location information, and obtain the spatial location encoding results of the sample road segments.

[0234] Optionally, the training module 400 is specifically used to input the time sequence of the sample road segment into the time encoder, and the time encoder determines the query matrix, key matrix and value matrix respectively; Based on the query matrix, key matrix, and value matrix, the time-series coding results of the sample road segments are determined.

[0235] Optionally, the training module 400 is specifically used to determine the hour label of each time series data according to the target sampling time of each time series data in the time series sequence of the sample road segment; Based on the target sampling time of each time series data and the mapping relationship between the sampling time and the preset peak marker, the target peak marker of each time series data is determined. Based on the hourly labels and target peak markers of each time series data, the sample road segments are fused using a multilayer perceptron to obtain the periodic sensing location coding results.

[0236] Optionally, the training module 400 is specifically used to perform bidirectional cross-operation based on the spatial feature information and temporal feature information of the sample road segments to obtain the temporal feature information of spatial guidance and the spatial feature information of temporal guidance, respectively. Based on the spatial feature information, temporal feature information, temporal feature information of spatial guidance, and spatial feature information of temporal guidance of the sample road segment, residual fusion processing is performed to obtain the spatiotemporal feature information of the sample road segment.

[0237] Optionally, the training module 400 is specifically used to determine the classification loss information based on the predicted green wave traffic information of each sample road segment in each sample road network and the actual green wave traffic label of each sample road segment in each sample road network. Based on the predicted vehicle speed data of each sample road segment in each sample road network and the actual vehicle speed data of each sample road segment in each sample road network, the speed regression loss information is determined. Based on the actual green wave trunk data of each sample road network and the predicted green wave traffic information of each sample road segment in each sample road network, the line continuity loss information is determined. Based on classification loss information, velocity regression loss information, and line continuity loss information, the loss information of the initial road network green wave line identification model is determined.

[0238] Optionally, it may also include: a determination module and a prediction module; The acquisition module 100 is also used to acquire multi-source data of each road segment in the target road network at the current time; The determination module is used to determine the spatiotemporal map of the target road network based on multi-source data; The prediction module is used to input the spatiotemporal map of the target road network into the road network green wave line identification model. The road network green wave line identification model predicts and outputs the predicted green wave traffic information of each road segment in the target road network, and generates a green wave potential heat map of the target road network based on the predicted green wave traffic information of each road segment. The determination module is used to search for and determine at least one green wave line under the target road network based on the predicted green wave traffic information of each road segment. The green wave line consists of at least two consecutive road segments.

[0239] Optionally, the module is specifically used to determine the green wave screening threshold based on the predicted green wave traffic information of each road segment; Based on the green wave screening threshold, each road segment is first screened to obtain each road segment after the first screening; Based on each of the first-selected road segments, a directed green wave reachability graph is constructed. The directed green wave reachability graph includes a second set of nodes and a set of directed edges. The second set of nodes includes the road segment nodes corresponding to each of the first-selected road segments. The set of directed edges includes the edges formed by consecutive road segment nodes in the same direction in the second set of nodes, and the weight of each edge is determined based on the predicted green wave traffic information of the two connected road segment nodes. Based on the directed green wave reachability map and the green wave continuity constraint, a full-domain search is performed to determine each green wave route under the target road network. The green wave continuity constraint includes: the difference between the predicted green wave traffic information of continuous road segments meets a preset threshold and the turning angle of each road segment meets a preset angle.

[0240] Optionally, the acquisition module 100 is also used to acquire the attribute characteristics of each green wave route. The attribute characteristics include at least: the green wave route's identifier, starting point coordinates, ending point coordinates, sequence of intersections along the route, recommended driving direction, predicted green wave bandwidth, and green wave reliability. The determination module is also used to determine the evaluation indicators of each green wave line based on the attribute characteristics of each green wave line. The evaluation indicators include: traffic efficiency indicators, green wave quality indicators, and spatiotemporal stability indicators. The determination module is also used to determine the comprehensive evaluation index of each green wave line based on the evaluation index of each green wave line; The determination module is also used to determine the green wave control evaluation results of the target road network based on the comprehensive evaluation indicators of each green wave line; the green wave control evaluation results are used to indicate the green wave control level of the target road network.

[0241] The above-described device is used to execute the method provided in the foregoing embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

[0242] These modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital signal processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). Alternatively, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a system-on-a-chip (SOC).

[0243] The modules described above can be connected or communicate with each other via wired or wireless connections. Wired connections may include metal cables, optical fibers, hybrid cables, or any combination thereof. Wireless connections may include connections via LAN, WAN, Bluetooth, ZigBee, or NFC, or any combination thereof. Two or more units can be combined into a single module, and any module can be divided into two or more units. Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here.

[0244] Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The device may be a computing device with data processing capabilities.

[0245] The device includes a processor 801, a storage medium 802, and a bus 803. The storage medium 802 stores program instructions that can be executed by the processor 801. When the electronic device is running, the processor 801 and the storage medium 802 communicate through the bus 803. The processor 801 executes the program instructions to implement the training method of the road network green wave line identification model as described in the embodiment.

[0246] The storage medium 802 stores program code, which, when executed by the processor 801, causes the processor 801 to perform various steps in the training method of the road network green wave line identification model according to various exemplary embodiments of this application as described in the "Exemplary Methods" section above.

[0247] The processor 801 can be a general-purpose processor, such as a central processing unit (CPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0248] Storage medium 802, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The storage medium can include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type storage medium, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic storage medium, magnetic disk, optical disk, etc. The storage medium is any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In the embodiments of this application, storage medium 802 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0249] Optionally, this application also provides a program product, such as a computer-readable storage medium, including a program that, when executed by a processor, performs the above-described method embodiments.

[0250] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0251] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0252] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0253] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A training method for a road network green wave line recognition model, characterized in that, include: Collect initial multi-source time-series data of each sample road segment in at least one sample road network. The initial multi-source time-series data includes at least: vehicle driving data corresponding to each historical sampling time, traffic flow data corresponding to each historical sampling time, sample road segment topology data, and sample intersection attribute information. Data preprocessing and spatiotemporal alignment processing are performed on the initial multi-source time series data of each sample road segment in each sample road network to obtain the target multi-source time series data of each sample road segment in each sample road network. Based on the target multi-source time-series data of each sample road segment in each sample road network, a spatiotemporal map of each sample road network is constructed. The spatiotemporal map is an integrated representation of the spatial topology, temporal dynamics and signal constraints of the sample road network. Based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, the road network green wave route identification model is trained; the road network green wave route identification model is used to predict the green wave traffic information of each road segment in the target road network and generate the green wave potential heat map of the target road network.

2. The method according to claim 1, characterized in that, The step of constructing a spatiotemporal map of each sample road network based on the target multi-source time-series data of each sample road segment in each sample road network includes: Based on the identifiers of each sample road segment and each sample intersection in the sample road network, a first node set is determined, which includes: sample road segment nodes and sample intersection nodes. Based on the connectivity between sample road segments, the traffic flow transfer relationship between sample road segments, and the signal coordination relationship between sample intersections, an edge set is determined; the edge set includes: road segment connectivity edges, traffic flow transfer edges, and signal coordination edges. Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample road segment node in the first node set; Based on the sample intersection attribute information in the target multi-source time series data of each sample road segment, determine the static attribute information of each sample intersection node in the first node set; Based on the sample road segment topology data in the target multi-source time series data of each sample road segment, determine the static attribute information of the connected edges of each road segment in the edge set; Based on the vehicle driving data, traffic flow data and sample intersection attribute information corresponding to each historical sampling time in the target multi-source time series data of each sample road segment, a dynamic attribute matrix is ​​determined. The dynamic attribute matrix is ​​used to characterize the dynamic change information of the operating status of the sample road network at each historical sampling time. The spatiotemporal map of the sample road network is constructed based on the first node set, the edge set, the static attribute information of each sample road segment node in the first node set, the static attribute information of each sample intersection node in the first node set, the static attribute information of each road segment connected edge in the edge set, and the dynamic attribute matrix.

3. The method according to claim 2, characterized in that, The method involves training a green wave route recognition model for each road network based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. This model includes: The spatiotemporal maps of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network are used as training data and input into the initial road network green wave route identification model. The initial road network green wave route identification model determines the spatiotemporal feature information of each sample road segment in each sample road network based on the spatiotemporal maps of each sample road network, and predicts the predicted green wave traffic information and the predicted vehicle speed data of each sample road segment in each sample road network based on the spatiotemporal feature information of each sample road segment in each sample road network. Based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network, the loss information of the initial road network green wave line identification model is determined. Based on the loss information, the initial road network green wave route identification model is iteratively trained, and the intermediate road network green wave route identification model at the time of iteration stop is used as the road network green wave route identification model.

4. The method according to claim 3, characterized in that, The step of determining the spatiotemporal characteristic information of each sample road segment in each sample road network based on the spatiotemporal map of each sample road network includes: The initial feature vector of the sample road segment is determined based on the static attribute information of the sample road segment nodes in the sample road network and the signal timing information of the sample intersection nodes associated with the sample road segment nodes. Based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segment, the spatial dependency coding result of the sample road segment is determined. Based on the spatial dependency coding result and the spatial location coding result of the sample road segment, the spatial feature information of the sample road segment is determined; Based on the dynamic attribute matrix in the spatiotemporal map of the sample road network, determine the temporal sequence of the sample road segment; Based on the time sequence of the sample road segment, determine the time-series coding result of the sample road segment; Based on the target sampling time of each time series data in the time series of the sample road segment and the mapping relationship between the sampling time and the preset peak marker, the periodic sensing location coding result of the sample road segment is determined. Based on the temporal coding results of the sample road segment and the periodic sensing location coding results of the sample road segment, the temporal feature information of the sample road segment is determined; Based on the spatial feature information and temporal feature information of the sample road segment, a fusion process is performed to obtain the spatiotemporal feature information of the sample road segment.

5. The method according to claim 4, characterized in that, The step of determining the spatial dependency coding result of the sample road segment based on the initial feature vector of the sample road segment and the initial feature vector of the associated road segment includes: Based on the edge set and the first node set in the spatiotemporal graph of the sample road network, the associated road segments of the sample road segment are determined; Based on the initial feature vector of the sample road segment and the initial feature vector of each associated road segment, the attention weight information between the sample road segment and each associated road segment is determined. Based on the initial feature vectors of each associated road segment and the attention weight information between the sample road segment and each associated road segment, the spatial dependency coding result of the sample road segment is determined.

6. The method according to claim 4, characterized in that, The methods for determining the spatial location coding results of the sample road segments include: Based on the location information of the sample road segment, spatial location encoding is performed on the sample road segment to obtain the spatial location encoding result of the sample road segment.

7. The method according to claim 4, characterized in that, Determining the temporal coding result of the sample road segment based on its temporal sequence includes: The time sequence of the sample road segment is input into the time encoder, which determines the query matrix, key matrix and value matrix respectively. Based on the query matrix, key matrix, and value matrix, the time-series coding result of the sample road segment is determined.

8. The method according to claim 4, characterized in that, The step of determining the periodic sensing location coding result of the sample road segment based on the target sampling time of each time-series data in the time-series of the sample road segment and the mapping relationship between the sampling time and the preset peak marker includes: Based on the target sampling time of each time series data in the time series sequence of the sample road segment, determine the hour label of each time series data; Based on the target sampling time of each time series data and the mapping relationship between the sampling time and the preset peak marker, the target peak marker of each time series data is determined. Based on the hourly labels and target peak markers of each time series data, the sample road segment is fused using a multilayer perceptron to obtain the periodic sensing location coding result.

9. The method according to claim 4, characterized in that, The step of fusing the spatial and temporal feature information of the sample road segments to obtain the spatiotemporal feature information of the sample road segments includes: Based on the spatial feature information and temporal feature information of the sample road segment, a bidirectional cross operation is performed to obtain the temporal feature information of spatial guidance and the spatial feature information of temporal guidance, respectively. Based on the spatial feature information, temporal feature information, temporal feature information of the spatial guidance, and spatial feature information of the temporal guidance of the sample road segment, residual fusion processing is performed to obtain the spatiotemporal feature information of the sample road segment.

10. The method according to claim 3, characterized in that, The loss information of the initial road network green wave route identification model is determined based on the predicted green wave traffic information of each sample road segment in each sample road network, the predicted vehicle speed data of each sample road segment in each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. This includes: Based on the predicted green wave traffic information of each sample road segment in each sample road network and the actual green wave traffic label of each sample road segment in each sample road network, the classification loss information is determined. Based on the predicted vehicle speed data of each sample road segment in each sample road network and the actual vehicle speed data of each sample road segment in each sample road network, the speed regression loss information is determined. Based on the actual green wave trunk data of each sample road network and the predicted green wave traffic information of each sample road segment in each sample road network, the line continuity loss information is determined. Based on the classification loss information, the speed regression loss information, and the line continuity loss information, the loss information of the initial road network green wave line identification model is determined.

11. The method according to claim 1, characterized in that, Also includes: Collect multi-source data of each road segment in the target road network at the current time; The spatiotemporal map of the target road network is determined based on the multi-source data; The spatiotemporal map of the target road network is input into the road network green wave route identification model. The road network green wave route identification model predicts and outputs the predicted green wave traffic information of each road segment in the target road network. Based on the predicted green wave traffic information of each road segment, a green wave potential heat map of the target road network is generated. Based on the predicted green wave traffic information of each road segment, at least one green wave line under the target road network is searched and determined, wherein the green wave line consists of at least two consecutive road segments.

12. The method according to claim 11, characterized in that, The step of searching and determining at least one green wave route under the target road network based on the predicted green wave traffic information of each road segment includes: Based on the predicted green wave traffic information for each road segment, determine the green wave screening threshold; Based on the green wave screening threshold, each road segment is first screened to obtain each road segment after the first screening. Based on each of the first-selected road segments, a directed green wave reachability graph is constructed. The directed green wave reachability graph includes a second set of nodes and a set of directed edges. The second set of nodes includes road segment nodes corresponding to each of the first-selected road segments. The set of directed edges includes edges formed by consecutive road segment nodes in the same direction in the second set of nodes, and the weight of each edge is determined based on the predicted green wave traffic information of two interconnected road segment nodes. Based on the directed green wave reachability map and the green wave continuity constraint, a full-domain search is performed to determine each green wave route under the target road network; the green wave continuity constraint includes: the difference between the predicted green wave traffic information of continuous road segments meets a preset threshold and the turning angle of each road segment meets a preset angle.

13. The method according to claim 11, characterized in that, Also includes: Obtain the attribute characteristics of each green wave route, which include at least: the green wave route's identifier, starting point coordinates, ending point coordinates, sequence of intersections along the route, recommended driving direction, predicted green wave bandwidth, and green wave reliability; Based on the attribute characteristics of each green wave line, the evaluation indicators for each green wave line are determined. The evaluation indicators include: traffic efficiency indicators, green wave quality indicators, and spatiotemporal stability indicators. Based on the evaluation indicators of each green wave line, determine the comprehensive evaluation indicators for each green wave line; Based on the comprehensive evaluation indicators of each green wave line, the green wave control evaluation result of the target road network is determined; the green wave control evaluation result is used to indicate the green wave control level of the target road network.

14. A training apparatus for a road network green wave route identification model, used to implement the training method for the road network green wave route identification model according to any one of claims 1-13, characterized in that, The device includes: an acquisition module, a processing module, a construction module, and a training module; The acquisition module is used to acquire initial multi-source time-series data of each sample road segment in at least one sample road network. The initial multi-source time-series data includes at least: vehicle driving data corresponding to each historical sampling time, traffic flow data corresponding to each historical sampling time, sample road segment topology data, and sample intersection attribute information. The processing module is used to perform data preprocessing and spatiotemporal alignment processing on the initial multi-source time series data of each sample road segment in each sample road network to obtain the target multi-source time series data of each sample road segment in each sample road network. The construction module is used to construct a spatiotemporal map of each sample road network based on the target multi-source time series data of each sample road segment in each sample road network. The spatiotemporal map is an integrated representation of the spatial topology, temporal dynamics and signal constraints of the sample road network. The training module is used to train the green wave route identification model of the road network based on the spatiotemporal map of each sample road network, the actual green wave traffic labels of each sample road segment in each sample road network, the actual vehicle speed data of each sample road segment in each sample road network, and the actual green wave trunk line data of each sample road network. The green wave route identification model of the road network is used to predict the green wave traffic information of each road segment in the target road network and generate the green wave potential heat map of the target road network.

15. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores program instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to implement the method as described in any one of claims 1 to 13.

Images