Intelligent traffic signal optimization method and system fusing beidou-3 and 5g

Abstract

The present application relates to the technical field of intelligent transportation, and more particularly to an intelligent traffic signal optimization method and system fusing Beidou-3 and 5G. The fusion positioning is obtained by cross-verification of Beidou position and 5G communication state for auxiliary positioning correction, the traffic situation awareness is obtained by combining lane-level topology identification with intention, the regional-level prediction is generated by predicting lane-level vehicle sequence through 5G edge computing distributed processing, and the global coordination timing scheme is obtained by modeling multi-intersection signal timing as multi-agent game iterative calculation of optimal strategy combination based on the prediction. The method improves the positioning accuracy and the global coordination of traffic signal control, and improves the traffic efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to an intelligent traffic signal optimization method and system that integrates BeiDou-3 and 5G. Background Technology

[0002] With the acceleration of urbanization, traffic congestion has become an increasingly prominent problem, making intelligent traffic signal optimization technology a key means to alleviate urban traffic pressure. Current conventional practices typically rely on global navigation satellite systems (such as GPS or BeiDou-2) to obtain vehicle location information, combined with fixed road detection equipment (such as inductive loop detectors and microwave detectors) to collect traffic flow data, and dynamically adjust intersection signal timing through centralized signal control algorithms (such as SCATS and SCOOT systems). In addition, some solutions attempt to introduce vehicle-to-everything (V2X) communication technologies (such as DSRC or 4G LTE networks) to achieve vehicle-to-infrastructure information exchange, thereby improving the real-time nature of traffic perception. However, these approaches have several limitations in practical applications.

[0003] Existing positioning technologies are susceptible to building obstruction and multipath effects in complex scenarios such as urban canyons, underpasses, and tunnels, leading to satellite signal attenuation or distortion and a significant decrease in vehicle positioning accuracy. This accuracy degradation makes it difficult for the system to accurately identify the specific lane a vehicle is in, thus failing to provide reliable micro-level traffic information for signal optimization. Furthermore, traditional traffic signal control often employs a centralized architecture, requiring all detection data to be transmitted back to a central server for processing before issuing commands to intersections. This results in high communication latency and concentrated computational load, making it difficult to adapt to rapidly changing traffic flow. Current practices also generally neglect real-time perception of vehicle driving intentions, with signal timing often based on historical statistics or cross-sectional flow, lacking refined consideration of dynamic factors such as turning demand and queue length. This leads to a mismatch between green light time allocation and actual traffic flow, exacerbating intersection overflow or empty lane phenomena. Therefore, a technical solution that can overcome these shortcomings and improve signal optimization accuracy is urgently needed. Summary of the Invention

[0004] This invention provides a method and system for optimizing intelligent traffic signals that integrates BeiDou-3 and 5G, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for optimizing intelligent traffic signals that integrates BeiDou-3 and 5G, comprising:

[0006] For each moving target within the road network, the location information provided by BeiDou-3 is cross-validated with the communication connection status information provided by the 5G network. When the location information is found to have multipath interference or obstruction that causes accuracy degradation, the signal arrival time difference and signal arrival angle difference in the communication connection status information are used for auxiliary positioning correction to generate a fused positioning result.

[0007] Based on the fusion positioning results and the road network lane-level topology, the lane position and driving intention of each moving target are identified to obtain lane-level traffic situation awareness information;

[0008] The lane-level traffic situation awareness information is distributed through the edge computing nodes of the 5G network. The lane-level arrival vehicle sequence of each approach lane in the future time window is predicted on the edge computing nodes corresponding to each intersection and then aggregated to the regional coordination and control node to generate regional traffic flow prediction information.

[0009] Based on the regional traffic flow prediction information, the signal timing problem of multiple intersections in the road network is modeled as a multi-agent game framework. The optimal strategy combination of each intersection in the multi-agent game framework is iteratively calculated. During the iteration process, the state space of each agent is updated in real time using the lane-level traffic situation perception information to obtain a globally coordinated signal timing scheme.

[0010] When the location information is found to have multipath interference or occlusion causing accuracy degradation, auxiliary positioning correction is performed using the signal arrival time difference and signal arrival angle difference in the communication connection status information to generate a fused positioning result, including:

[0011] Based on the signal arrival time difference between multiple 5G base stations and the mobile target in the communication connection status information, the distance difference between the mobile target and different 5G base stations is calculated. A hyperboloid is constructed with any two 5G base stations as the focus and the distance difference as the parameter. Multiple hyperboloids form a set of hyperbolic spatial positioning constraints.

[0012] The arrival angle information of the uplink signal of the mobile target received by multiple 5G base stations is extracted, the signal arrival angle difference between different 5G base stations is calculated, and the arrival angle information is verified for consistency based on the spatial geometric relationship between the signal arrival angle difference and each 5G base station. Abnormal angle measurement values ​​that cause the signal arrival angle difference to not meet the spatial geometric consistency constraint are eliminated.

[0013] Based on the arrival angle information after consistency verification, the ray direction of each 5G base station pointing to the moving target is determined. A spatial ray is constructed with the location of each 5G base station as the starting point and the ray direction as the direction vector. Multiple spatial rays form a set of ray intersection spatial positioning constraints.

[0014] Calculate the intersection region in space between the hyperbolic spatial positioning constraint set and the ray intersection spatial positioning constraint set, and take the geometric center of the intersection region as the auxiliary positioning coordinates;

[0015] The deviation between the location information and the auxiliary positioning coordinates is calculated. A correction weighting coefficient is determined based on the degree of accuracy degradation of the location information. The deviation is then weighted and compensated using the correction weighting coefficient and added to the location information to generate the fused positioning result.

[0016] Based on the fused localization results and the road network lane-level topology, the lane position and driving intention of each moving target are identified, resulting in lane-level traffic situation awareness information, including:

[0017] The fused positioning results are projected onto the road network lane-level topology, and the lane position of each moving target is determined by calculating the vertical distance between the spatial coordinates of each moving target and the centerline of each lane.

[0018] Extract the fusion positioning results of each moving target in a continuous time series, construct multiple accessible lane connection paths starting from the current lane position based on the lane-level topology of the road network, calculate the fitting deviation between the spatial trajectory formed by the continuous fusion positioning results and each accessible lane connection path, select the accessible lane connection path with the smallest fitting deviation as the driving path, determine the end lane of the driving path as the target lane, and identify the driving intention of each moving target.

[0019] The lane position is associated with the driving intention, and the number of moving targets in each lane, the speed distribution of moving targets in each lane, and the lane changing behavior characteristics of moving targets are statistically analyzed to generate the lane-level traffic situation awareness information.

[0020] The lane-level traffic situation awareness information is processed in a distributed manner through the edge computing nodes of the 5G network. At the edge computing nodes corresponding to each intersection, the lane-level arrival vehicle sequences for each approach lane within a future time window are predicted and aggregated to the regional coordination and control node, generating regional-level traffic flow prediction information, including:

[0021] The lane-level traffic situation awareness information is distributed to the edge computing nodes of the corresponding intersections according to the intersection affiliation of the lane position of each moving target. The lane position, driving intention and moving speed of each moving target are extracted at each edge computing node. The approach lane to which each moving target will arrive is determined based on the lane position and driving intention of each moving target. The estimated arrival time of each moving target at the corresponding approach lane is calculated based on the moving speed of each moving target and the spatial distance between the current lane position of each moving target and the corresponding approach lane.

[0022] Each moving target is sorted according to its expected arrival time on each approach lane within a future time window, forming a lane-level arrival vehicle sequence for each approach lane within the future time window. During the sorting process, the turning requirements of each moving target after arriving at the approach lane are identified based on the driving intention of each moving target, and the turning requirements are marked in the lane-level arrival vehicle sequence.

[0023] The lane-level arrival vehicle sequences generated on each edge computing node are aggregated to the regional coordination control node. In the regional coordination control node, based on the road network spatiotemporal propagation map, the cross-intersection propagation delay calculation and timing correction are performed on the lane-level arrival vehicle sequences of multiple intersections to generate the regional traffic flow prediction information.

[0024] In the regional coordinated control node, based on the road network spatiotemporal propagation map, the cross-intersection propagation delay calculation and timing correction are performed on the lane-level arrival vehicle sequences at multiple intersections to generate the regional-level traffic flow prediction information, including:

[0025] Each lane in the lane-level topology of the road network is abstracted as a node, and the connection relationship between lanes and the transfer time of vehicles between lanes are abstracted as directed edges. A spatiotemporal propagation graph of the road network is constructed in the regional coordination control node.

[0026] In the spatiotemporal propagation diagram, a multi-hop propagation path from the current lane position to the final target lane is constructed for each moving target. The lane transfer direction of each moving target at each intersection is determined according to the steering requirements. The theoretical time of passage of each moving target at each intermediate lane node without delay is calculated along the multi-hop propagation path.

[0027] The system obtains the traffic light status and the current number of vehicles in the queue at each intermediate lane node, and calculates the queuing delay time of each moving target at each intermediate lane node by combining the aforementioned no-delay theory. The queuing delay time is then added to the estimated arrival time of each moving target at the downstream intersection entrance.

[0028] The estimated arrival times after accumulating queuing delays are reordered according to time sequence for the lane-level arrival vehicle sequences at the downstream intersection entrances, and then propagated back to the corresponding downstream lane nodes in the spatiotemporal propagation graph to update the state of each node in the spatiotemporal propagation graph.

[0029] The lane-level arrival vehicle sequences of each node in the updated spatiotemporal propagation graph are extracted and organized according to the intersection affiliation to generate the regional traffic flow prediction information.

[0030] Based on the aforementioned regional traffic flow prediction information, the signal timing problem at multiple intersections within the road network is modeled as a multi-agent game framework, including:

[0031] The lane-level arrival vehicle sequences at each intersection in the regional traffic flow prediction information are grouped according to the intersection identifier. The expected arrival time and turning demand of vehicles at each entrance lane of each intersection are extracted as the initial input of the agent's state space, and an independent agent is constructed for each intersection in the road network.

[0032] The combination of the signal phase sequence, the green light duration of each phase, and the phase switching order that each independent agent can choose is defined as the action space of the independent agent.

[0033] For each independent intelligent agent, a revenue indicator calculation rule is constructed. The revenue indicator calculation rule calculates the average delay time of all vehicles in the intersection controlled by the independent intelligent agent as the delay cost item based on the signal phase scheme and phase duration combination selected by the independent intelligent agent.

[0034] The vehicle overflow penalty for adjacent intersections is calculated based on the difference between the vehicle outflow volume of each exit lane of the independent intelligent agent controlled intersection and the vehicle carrying capacity of the corresponding entrance lane of the adjacent independent intelligent agent controlled intersection.

[0035] The delay cost item and the vehicle overflow penalty item at the adjacent intersection are weighted and combined according to a preset weight coefficient to obtain the revenue index of the independent intelligent agent, and a multi-agent game framework including the state space, action space and revenue index of each independent intelligent agent is established.

[0036] The optimal strategy combination for each intersection in the multi-agent game framework is calculated iteratively. During the iteration process, the state space of each agent is updated in real time using the lane-level traffic situation awareness information to obtain a globally coordinated signal timing scheme, including:

[0037] In the multi-agent game framework, a policy mapping mechanism is constructed for each agent. The policy mapping mechanism outputs the signal phase scheme and phase duration combination of the agent in the current round based on the agent's current state and the historical actions of neighboring agents. Each agent calculates the agent's reward index value based on the signal phase scheme and phase duration combination output by the policy mapping mechanism, and then backpropagates the reward index value to the policy mapping mechanism to adjust the mapping parameters of the policy mapping mechanism.

[0038] The real-time vehicle arrival rate and real-time queue length of each intersection entrance lane are extracted from the lane-level traffic situation perception information. The initial state of each agent is corrected based on the real-time vehicle arrival rate and real-time queue length. The corrected initial state is input into the strategy mapping mechanism of each agent to recalculate the signal phase scheme and phase duration combination. The output calculation of the strategy mapping mechanism and the back propagation process of the revenue index are iteratively executed until the revenue index values ​​of each agent converge.

[0039] The signal phase schemes and phase duration combinations output by the policy mapping mechanisms of each agent after convergence are merged according to the intersection identifiers to generate a globally coordinated signal timing scheme.

[0040] A second aspect of this invention provides an intelligent traffic signal optimization system integrating BeiDou-3 and 5G, comprising:

[0041] The positioning fusion unit is used to cross-verify the location information provided by BeiDou-3 with the communication connection status information provided by 5G network for each moving target in the road network. When the location information is found to have multipath interference or obstruction that causes accuracy degradation, the unit uses the signal arrival time difference and signal arrival angle difference in the communication connection status information to perform auxiliary positioning correction and generate a fused positioning result.

[0042] The lane perception unit is used to identify the lane position and driving intention of each moving target based on the fused positioning result and the road network lane-level topology, and obtain lane-level traffic situation perception information.

[0043] The traffic flow prediction unit is used to distribute the lane-level traffic situation perception information through the edge computing nodes of the 5G network, predict the lane-level arrival vehicle sequence of each approach lane in the future time window on the edge computing nodes corresponding to each intersection, and aggregate it to the regional coordination control node to generate regional traffic flow prediction information.

[0044] The signal timing unit is used to model the signal timing problem of multiple intersections in the road network as a multi-agent game framework based on the regional traffic flow prediction information, iteratively calculate the optimal strategy combination of each intersection in the multi-agent game framework, and use the lane-level traffic situation perception information to update the state space of each agent in real time during the iteration process to obtain a globally coordinated signal timing scheme.

[0045] A third aspect of the present invention provides an electronic device, comprising:

[0046] processor;

[0047] Memory used to store processor-executable instructions;

[0048] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0049] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0050] The beneficial effects of this application are as follows:

[0051] To address the accuracy degradation of BeiDou-3 positioning caused by multipath interference or obstruction in complex urban environments, this method utilizes the signal arrival time difference and angle of arrival difference from 5G network communication connection status information for auxiliary correction, significantly improving positioning accuracy and robustness. The fused positioning results compensate for the shortcomings of single BeiDou positioning, achieving high-precision location acquisition even in scenarios such as tunnels and densely populated high-rise areas, providing a reliable data foundation for subsequent traffic analysis and ensuring the system's continuous and stable operation under harsh conditions.

[0052] Based on high-precision fusion positioning and road network lane-level topology, it can accurately identify the lane position and driving intention of each moving target, achieving fine perception of road network traffic status. Lane-level traffic situation perception information includes micro-details such as vehicle distribution and turning intentions, breaking through the limitations of traditional macro-models that cannot capture lane-level fluctuations. This enables traffic management to carry out differentiated control for different lanes and different flow directions, improving the efficiency of spatial resource utilization and the smoothness of traffic flow.

[0053] By leveraging 5G edge computing nodes to perform distributed processing of lane-level situational information, each intersection independently predicts lane-level arrival vehicle sequences within future time windows, which are then aggregated to a regional coordination and control node to generate regional traffic flow prediction information. This layered architecture significantly reduces data transmission volume and cloud processing latency, enhancing the system's real-time response capabilities. Multi-intersection signal timing is modeled as a multi-agent game framework, iteratively calculating the optimal strategy combination based on real-time updated lane-level state information to achieve a globally coordinated signal timing scheme. This scheme significantly reduces vehicle stops, delay time, and fuel consumption, effectively alleviating traffic congestion while balancing fairness among different intersections with the overall road network capacity, thus improving regional traffic operation efficiency. Attached Figure Description

[0054] Figure 1 A flowchart illustrating the intelligent traffic signal optimization method integrating BeiDou-3 and 5G;

[0055] Figure 2 A schematic diagram of the fusion positioning process for 5G-assisted positioning correction. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0058] Figure 1 This is a flowchart illustrating the intelligent traffic signal optimization method integrating BeiDou-3 and 5G according to an embodiment of the present invention.

[0059] The intelligent traffic signal optimization methods integrating BeiDou-3 and 5G include:

[0060] For each moving target within the road network, the location information provided by BeiDou-3 is cross-validated with the communication connection status information provided by the 5G network. When the location information is found to have multipath interference or obstruction that causes accuracy degradation, the signal arrival time difference and signal arrival angle difference in the communication connection status information are used for auxiliary positioning correction to generate a fused positioning result.

[0061] Based on the fusion positioning results and the road network lane-level topology, the lane position and driving intention of each moving target are identified to obtain lane-level traffic situation awareness information;

[0062] The lane-level traffic situation awareness information is distributed through the edge computing nodes of the 5G network. The lane-level arrival vehicle sequence of each approach lane in the future time window is predicted on the edge computing nodes corresponding to each intersection and then aggregated to the regional coordination and control node to generate regional traffic flow prediction information.

[0063] Based on the regional traffic flow prediction information, the signal timing problem of multiple intersections in the road network is modeled as a multi-agent game framework. The optimal strategy combination of each intersection in the multi-agent game framework is iteratively calculated. During the iteration process, the state space of each agent is updated in real time using the lane-level traffic situation perception information to obtain a globally coordinated signal timing scheme.

[0064] In one optional implementation, when multipath interference or occlusion causing accuracy degradation is detected in the location information, auxiliary positioning correction is performed using the signal arrival time difference and signal arrival angle difference in the communication connection status information to generate a fused positioning result, including:

[0065] Based on the signal arrival time difference between multiple 5G base stations and the mobile target in the communication connection status information, the distance difference between the mobile target and different 5G base stations is calculated. A hyperboloid is constructed with any two 5G base stations as the focus and the distance difference as the parameter. Multiple hyperboloids form a set of hyperbolic spatial positioning constraints.

[0066] The arrival angle information of the uplink signal of the mobile target received by multiple 5G base stations is extracted, the signal arrival angle difference between different 5G base stations is calculated, and the arrival angle information is verified for consistency based on the spatial geometric relationship between the signal arrival angle difference and each 5G base station. Abnormal angle measurement values ​​that cause the signal arrival angle difference to not meet the spatial geometric consistency constraint are eliminated.

[0067] Based on the arrival angle information after consistency verification, the ray direction of each 5G base station pointing to the moving target is determined. A spatial ray is constructed with the location of each 5G base station as the starting point and the ray direction as the direction vector. Multiple spatial rays form a set of ray intersection spatial positioning constraints.

[0068] Calculate the intersection region in space between the hyperbolic spatial positioning constraint set and the ray intersection spatial positioning constraint set, and take the geometric center of the intersection region as the auxiliary positioning coordinates;

[0069] The deviation between the location information and the auxiliary positioning coordinates is calculated. A correction weighting coefficient is determined based on the degree of accuracy degradation of the location information. The deviation is then weighted and compensated using the correction weighting coefficient and added to the location information to generate the fused positioning result.

[0070] like Figure 2 As shown, the method includes:

[0071] In scenarios where the accuracy of BeiDou-3 positioning signals is degraded due to multipath interference or building obstruction, relying solely on satellite positioning is insufficient to meet the accuracy requirements of lane-level positioning. In such cases, by utilizing the communication connection status information between multiple base stations and the moving target in the 5G network, spatial constraints are constructed using two independent measurement quantities: signal arrival time difference and signal arrival angle difference, to assist in correcting the BeiDou-3 positioning results.

[0072] To address the processing of signal arrival time difference (ATD), the timestamps of the uplink signal from the moving target arriving at different 5G base stations are extracted, and the time difference between any two base stations receiving the same signal is calculated. Since electromagnetic waves propagate at the speed of light in free space, there is a definite linear relationship between time difference and distance difference. Therefore, the time difference for each pair of base stations can be converted into the distance difference between the moving target and these two base stations. In three-dimensional space, the trajectory of points whose distance difference to two fixed foci is constant forms a hyperboloid. Using the position coordinates of any two 5G base stations as the two foci of the hyperboloid and the corresponding distance difference as parameters, a hyperboloid equation can be constructed. When the number of 5G base stations participating in the positioning is... At this time, theoretically it is possible to construct Each base station pair corresponds to There are several hyperboloids, which form a set of hyperbolic spatial positioning constraints in three-dimensional space. The true position of the moving target must simultaneously satisfy the constraints of all hyperboloids in this set, thus transforming the positioning problem into a problem of solving the spatial intersection of multiple hyperboloids. In actual calculations, considering measurement errors caused by factors such as non-line-of-sight propagation, not all hyperboloids can intersect precisely at the same point. Therefore, a weighted least squares method is used to jointly solve the constraints of each hyperboloid to reduce the impact of measurement noise on the positioning results.

[0073] To address the signal arrival angle difference (AOP) processing, the AOP information of the uplink signal received from the moving target by each 5G base station antenna array is extracted, including both azimuth and elevation components. Since different base stations observe the same moving target from their respective locations, there are inherent constraints on the AOPs between each base station determined by spatial geometry. Specifically, if the three-dimensional coordinates of each base station are known, the AOP difference between any two base stations observing the same target should be consistent with the spatial azimuth difference between these two base stations relative to the target. Consistency verification is performed on the AOP differences of each pair of base stations, calculating the deviation between the measured angle difference and the theoretical angle difference calculated based on the currently estimated target position. If the AOP measurement value of a base station causes the angle difference deviation to exceed a preset threshold, the measurement value is determined to be an abnormal angle measurement value and discarded to avoid interference from non-line-of-sight reflections or antenna array measurement errors in subsequent positioning calculations. After consistency verification and filtering, valid AOP measurements that satisfy the spatial geometric consistency constraints are retained.

[0074] Based on the validated angle of arrival information, using the three-dimensional coordinates of each 5G base station as the starting point and the direction vector determined by the angle of arrival as the direction, spatial rays are constructed in three-dimensional space. Each valid base station corresponds to one spatial ray, and multiple spatial rays form a set of ray intersection spatial positioning constraints in space. The true position of the moving target should be located near the intersection area of ​​all rays. Due to errors in actual measurements, multiple rays usually do not intersect precisely at one point, but rather form a spatial intersection area. The least common perpendicular method or weighted least squares method is used to jointly solve for multiple rays, calculating the spatial point that minimizes the sum of the squared distances from each ray to that point, which is taken as the optimal estimated position of the ray intersection constraint.

[0075] The hyperbolic spatial positioning constraint set and the ray intersection spatial positioning constraint set are jointly solved in three-dimensional space to calculate the intersection region that simultaneously satisfies both types of constraints. The hyperbolic constraint provides a circumferential constraint regarding distance difference, while the ray constraint provides a linear constraint regarding direction. These two types of constraints are geometrically complementary, and their combined use can significantly reduce the positioning uncertainty area. The geometric center coordinates of the intersection region are taken as auxiliary positioning coordinates, denoted as... This coordinate represents the estimated position of the moving target calculated based on 5G communication measurements.

[0076] After obtaining the auxiliary positioning coordinates, the three-dimensional deviation between the original position information provided by BeiDou-3 and the auxiliary positioning coordinates is calculated and denoted as . , , These correspond to the coordinate deviation components in the east, north, and zenith directions, respectively. The correction weighting coefficients are determined based on the degree of accuracy degradation of the BeiDou-3 positioning information. The degree of accuracy degradation can be comprehensively evaluated using indicators such as the positioning accuracy factor, signal-to-noise ratio, and number of visible satellites in satellite positioning. When the degree of accuracy degradation is high, A value close to 1 indicates a higher level of confidence in the auxiliary positioning coordinates and a greater correction to the original BeiDou-3 position information; when the accuracy degradation is low, A value close to 0 indicates that the BeiDou-3 positioning information still has high reliability, requiring only minor adjustments. This is achieved using the adjustment weighting coefficient. The deviations are weighted and then superimposed on the original BeiDou-3 position information to generate a fused positioning result. This fused positioning result combines the global coverage advantage of BeiDou-3 with the geometric positioning advantage of 5G networks in dense urban environments. It effectively maintains lane-level positioning accuracy even under multipath interference or obstruction scenarios, providing a reliable location basis for subsequent lane position recognition and driving intention determination.

[0077] At the engineering implementation level, the aforementioned assisted positioning correction process is executed in real time on the edge computing nodes of the 5G network to meet the stringent requirements of traffic signal optimization for positioning latency. The positioning correction results for each moving target are updated at a fixed frequency to ensure that the fused positioning results can reflect the real-time motion status of the target and support the continuous and stable operation of subsequent lane-level traffic situation awareness.

[0078] In one optional implementation, based on the fused positioning results and the road network lane-level topology, the lane position and driving intention of each moving target are identified to obtain lane-level traffic situation awareness information, including:

[0079] The fused positioning results are projected onto the road network lane-level topology, and the lane position of each moving target is determined by calculating the vertical distance between the spatial coordinates of each moving target and the centerline of each lane.

[0080] Extract the fusion positioning results of each moving target in a continuous time series, construct multiple accessible lane connection paths starting from the current lane position based on the lane-level topology of the road network, calculate the fitting deviation between the spatial trajectory formed by the continuous fusion positioning results and each accessible lane connection path, select the accessible lane connection path with the smallest fitting deviation as the driving path, determine the end lane of the driving path as the target lane, and identify the driving intention of each moving target.

[0081] The lane position is associated with the driving intention, and the number of moving targets in each lane, the speed distribution of moving targets in each lane, and the lane changing behavior characteristics of moving targets are statistically analyzed to generate the lane-level traffic situation awareness information.

[0082] After obtaining the fused positioning results, they need to be spatially correlated with the road network lane-level topology to determine the precise lane location of each moving target. The road network lane-level topology is pre-stored in the form of a digital map, where each lane is represented by a centerline. The centerline consists of a series of ordered spatial nodes, with adjacent nodes connected by line segments, forming a polygonal geometric representation of the lane centerline. When projecting the spatial coordinates of the fused positioning results onto this topology, for each moving target, the vertical distance from its current spatial coordinates to the centerlines of each lane within the road network is calculated. The vertical distance is calculated as follows: draw perpendicular lines from the moving target's coordinates to each segment of the lane centerline. The length of the perpendicular when the foot of the perpendicular falls within the range of the segment is taken as the local distance corresponding to that segment. If the foot of the perpendicular is not within the range of the segment, the distance from the endpoint is taken as the local distance. Finally, the minimum value of all local distances is taken as the vertical distance from the moving target to the lane centerline. After calculating the perpendicular distance to all candidate lanes within the road network, the lanes are selected. The lane with the smallest width that satisfies the lane width constraint is taken as the lane position of the moving target. Lane width constraints are usually set according to road design specifications with threshold values. When the lane exceeds half the lane width plus a certain tolerance, the lane is not considered a valid candidate, thus avoiding mismatching the target to an adjacent parallel lane.

[0083] Based on the determined lane positions, the driving intentions of each moving target are further identified. The identification of driving intentions relies on the spatial trajectories formed by fused localization results over a continuous time series. Several fused localization results of the moving target over a past period are extracted and arranged chronologically to form a discrete spatial trajectory point sequence. Starting from the current lane position, based on the lane connection relationships in the road network's lane-level topology, all lane connection paths reachable from the current lane within a certain future path length are enumerated. Each reachable lane connection path consists of several sequentially connected lane centerline segments, forming a candidate driving path set. For each candidate driving path, the fitting deviation between the spatial trajectory point sequence and the path is calculated. The fitting deviation is calculated by calculating the perpendicular distance from each point in the trajectory point sequence to the centerline of the candidate path. ,in For the time series index of the trajectory points, take the root mean square of the vertical distances between all trajectory points to obtain the comprehensive fitting bias corresponding to the candidate path. ,Right now ,in This represents the total number of trajectory points involved in the calculation. The overall fitting bias is calculated for each path in the candidate driving path set. Then, select The smallest candidate path is taken as the most likely current travel path for the moving target, and the lane ending at that path is designated as the target lane. The type of the target lane is directly mapped to the driving intention category: if the target lane and the current lane are in the same straight-line direction, it is identified as a straight-line intention; if the target lane is located to the left of the current lane, it is identified as a left-turn intention; if the target lane is located to the right of the current lane, it is identified as a right-turn intention; if the target lane is a dedicated U-turn lane, it is identified as a U-turn intention. The above intention classification is completed based on the steering attribute annotation of each lane in the road network topology, without the need for an additional intention classification model.

[0084] After determining the lane position and driving intention, the two are correlated to form a complete lane-level state description for each moving target. Using each lane as a statistical unit, the number of moving targets carried in each lane within the road network at the current moment is counted to obtain the real-time occupied target count for each lane. This indicator directly reflects the congestion level of each lane. The instantaneous speeds of all moving targets in each lane are summarized, and the mean and variance of the target speeds within the lane are calculated to form a speed distribution description. Instantaneous speed is calculated by removing the spatial position of two adjacent fused positioning results according to the corresponding time interval. Received, among which This represents the timestamp difference between adjacent fusion positioning results. The mean of the speed distribution reflects the overall traffic efficiency of the lane, while the variance reflects the uniformity of traffic flow within the lane. A large variance indicates significant speed dispersion within the lane, corresponding to situations where low-speed targets within the lane are obstructing the normal movement of subsequent targets.

[0085] Lane changing behavior features are extracted based on the analysis of the historical lane position sequence of moving targets. When the lane affiliation of a moving target changes within a continuous time step (i.e., the lane it belonged to at the previous moment is different from the lane it belongs to at the current moment), a lane changing event is recorded, and the starting and target lane numbers, the time of the change, and the target's speed state during the change are labeled. The number of changing events for each lane pair within a time window is statistically analyzed to form a lane changing frequency matrix. The matrix contains the lane changing frequency of the first lane pair. Line 1 The elements of the column represent the number of elements from the first column. Lane #1 changes to Lane #2 The number of events occurring in lane number 1. Lane changing behavior characteristics, together with driving intentions, constitute a description of the future movement trend of a moving target, and can provide richer input features for predicting the sequence of vehicles arriving at subsequent intersection approach lanes.

[0086] The system integrates the number of targets, speed distribution, lane-changing behavior characteristics, and driving intention labels of each target in each lane to form structured lane-level traffic situation awareness information. This information is organized by lane, covering all lanes within the monitoring range of the road network, and is updated on a rolling basis at fixed time steps to ensure the timeliness of the situation awareness information. In actual deployment, the update frequency of the fused positioning results is synchronized with the communication refresh cycle of the 5G network, typically reaching the order of 100 milliseconds. Therefore, the refresh latency of the lane-level traffic situation awareness information is extremely low, supporting the real-time prediction needs of subsequent edge computing nodes for vehicle arrival sequences at intersections. After generation, the lane-level traffic situation awareness information is reported to the corresponding edge computing nodes at each intersection via the low-latency transmission capability of the 5G network, serving as the basic data input for distributed traffic flow prediction and signal timing optimization.

[0087] In one optional implementation, the lane-level traffic situation awareness information is distributed through edge computing nodes of the 5G network. At the edge computing nodes corresponding to each intersection, the lane-level arrival vehicle sequences for each approach lane within a future time window are predicted and aggregated to the regional coordination control node, generating regional-level traffic flow prediction information, including:

[0088] The lane-level traffic situation awareness information is distributed to the edge computing nodes of the corresponding intersections according to the intersection affiliation of the lane position of each moving target. The lane position, driving intention and moving speed of each moving target are extracted at each edge computing node. The approach lane to which each moving target will arrive is determined based on the lane position and driving intention of each moving target. The estimated arrival time of each moving target at the corresponding approach lane is calculated based on the moving speed of each moving target and the spatial distance between the current lane position of each moving target and the corresponding approach lane.

[0089] Each moving target is sorted according to its expected arrival time on each approach lane within a future time window, forming a lane-level arrival vehicle sequence for each approach lane within the future time window. During the sorting process, the turning requirements of each moving target after arriving at the approach lane are identified based on the driving intention of each moving target, and the turning requirements are marked in the lane-level arrival vehicle sequence.

[0090] The lane-level arrival vehicle sequences generated on each edge computing node are aggregated to the regional coordination control node. In the regional coordination control node, based on the road network spatiotemporal propagation map, the cross-intersection propagation delay calculation and timing correction are performed on the lane-level arrival vehicle sequences of multiple intersections to generate the regional traffic flow prediction information.

[0091] After lane-level traffic situation awareness information is generated, it needs to be processed efficiently in a distributed manner using the edge computing architecture of the 5G network. Each intersection in the road network corresponds to a processing unit deployed on a 5G edge computing node. The lane location information of each moving target carries a clear intersection affiliation identifier, which has been bound during the lane-level topology matching stage. According to the intersection affiliation, the lane-level traffic situation awareness information of each moving target is distributed to the edge computing node of its respective intersection. This allows each edge computing node to process only the moving target data related to its own intersection, avoiding the centralized uploading of the entire road network data to the regional coordination and control node, thereby significantly reducing the backhaul bandwidth pressure and processing latency.

[0092] At each edge computing node, for each moving target distributed to that node, three key attributes are extracted: lane position, driving intention, and speed. Lane position indicates the specific lane number where the moving target is currently located and its longitudinal coordinates within the lane; driving intention indicates the expected turning behavior of the moving target on the current road segment, including categories such as going straight, turning left, turning right, and making a U-turn. This information has already been obtained through trajectory curvature analysis and lane transition frequency matrix inference during the lane-level traffic situation perception stage; speed is calculated from the difference between displacement and timestamp between adjacent fused positioning results, reflecting the current real-time speed of the moving target.

[0093] Based on lane location and driving intent, the approach lane to which each moving target will arrive is determined. An intersection approach lane refers to the lane segment a vehicle travels before entering the stop line at an intersection; different driving intents correspond to different approach lane assignments. For example, a moving target intending to turn left typically belongs to the dedicated left-turn approach lane, while a vehicle intending to go straight corresponds to the straight-ahead lane approach lane. In the lane-level road network topology, the correspondence between each lane and approach lane is pre-established; by querying this topological relationship, the target approach lane for each moving target can be quickly determined.

[0094] After determining the target approach lane, calculate the estimated arrival time of each moving target at its corresponding approach lane. Let the spatial distance from the current lane position of the moving target to the stop line of the target approach lane be... The moving target's current movement speed is The current time is Then the estimated arrival time satisfy: Considering that speed is not constant during actual driving, when calculating spatial distance... When integrating, the path integral is performed along the centerline of the road network lanes, rather than the straight-line distance, to ensure that the distance metric is consistent with the actual travel path. For moving targets with low speeds or a deceleration trend, an acceleration correction term can be introduced to compensate for the estimated arrival time. However, in lightweight processing scenarios with edge computing nodes, a uniform speed assumption is usually adopted to ensure computational efficiency. Under conditions where the future time window is short (usually 30 to 120 seconds), the error introduced by the uniform speed assumption is within an acceptable range.

[0095] After obtaining the estimated arrival times of each moving target, the targets are arranged in ascending order according to their estimated arrival times within the future time window at each approach lane, forming a lane-level arrival vehicle sequence for each approach lane. This sequence is structured around a time axis and records future time windows. The list of moving targets expected to arrive at the entrance in sequence, among which... To predict the length of the time window, each record in the sequence contains a moving target identifier, an estimated arrival time, and a driving intention label. The driving intention label directly reflects the turning demand of the moving target after arriving at the approach lane. During the signal timing optimization stage, this turning demand will be used to determine the composition of the queued vehicles corresponding to each phase, thereby achieving refined allocation of phase green light durations.

[0096] During the sorting process, if the difference in the estimated arrival times of two moving targets is less than the time resolution threshold (usually 1 second), they are considered to have arrived simultaneously, recorded side-by-side in the sequence, and processed as arrival events within the same time slot in subsequent processing. The steering demand labeling uses an enumeration type, consistent with the intersection signal phase definition, ensuring that the lane-level arrival vehicle sequence can be directly used as input to the signal timing optimization model.

[0097] After each edge computing node generates the lane-level arrival vehicle sequence for each approach lane at its intersection, it uploads the sequence data to the regional coordination and control node via the 5G network. The regional coordination and control node aggregates the sequence data from edge computing nodes at all intersections within the road network. However, the prediction results for each intersection differ in terms of temporal reference and spatial correlation, requiring cross-intersection propagation delay calculation and timing correction.

[0098] The road network spatiotemporal propagation graph is the core data structure for realizing cross-intersection time series correction. This graph uses intersections as nodes and road segments as directed edges. Edge attributes include information such as segment length, number of lanes, speed limit, and historical average travel time. When the arrival sequence of vehicles at an intersection's approach lane contains vehicles from upstream intersections, there is a road segment propagation delay between the vehicle's departure time from the upstream intersection and its arrival time at the downstream intersection. Assume the vehicle departs from the upstream intersection... After leaving, drive along the road to the downstream intersection. The propagation delay is Then, in the regional coordination control node, the estimated arrival time of the relevant vehicle records in the upstream intersection prediction sequence is corrected to: ,in This is the estimated departure time of the vehicle at the upstream intersection. The propagation delay is determined by the historical travel time statistics of the corresponding road segments in the road network spatiotemporal propagation map and dynamically adjusted according to the traffic flow density of the current time period. By performing propagation delay calculations on all relevant road segments in the road network spatiotemporal propagation map one by one, the lane-level arrival vehicle sequences at each intersection are aligned on the time axis, eliminating the temporal inconsistency problem introduced by the independent prediction of each edge computing node.

[0099] After cross-intersection propagation delay calculation and timing correction, the lane-level arrival vehicle sequences at each intersection are integrated under a unified time reference to form regional traffic flow prediction information. This information comprehensively describes the vehicle arrival timing and turning demand distribution of each approach lane at multiple intersections within the road network within a future time window, providing high-precision and time-efficient input data support for the coordinated optimization of signal timing schemes at each intersection in the subsequent multi-agent game framework.

[0100] In one optional implementation, the regional coordination control node performs cross-intersection propagation delay calculation and timing correction on lane-level arrival vehicle sequences at multiple intersections based on the road network spatiotemporal propagation map to generate the regional traffic flow prediction information, including:

[0101] Each lane in the lane-level topology of the road network is abstracted as a node, and the connection relationship between lanes and the transfer time of vehicles between lanes are abstracted as directed edges. A spatiotemporal propagation graph of the road network is constructed in the regional coordination control node.

[0102] In the spatiotemporal propagation diagram, a multi-hop propagation path from the current lane position to the final target lane is constructed for each moving target. The lane transfer direction of each moving target at each intersection is determined according to the steering requirements. The theoretical time of passage of each moving target at each intermediate lane node without delay is calculated along the multi-hop propagation path.

[0103] The system obtains the traffic light status and the current number of vehicles in the queue at each intermediate lane node, and calculates the queuing delay time of each moving target at each intermediate lane node by combining the aforementioned no-delay theory. The queuing delay time is then added to the estimated arrival time of each moving target at the downstream intersection entrance.

[0104] The estimated arrival times after accumulating queuing delays are reordered according to time sequence for the lane-level arrival vehicle sequences at the downstream intersection entrances, and then propagated back to the corresponding downstream lane nodes in the spatiotemporal propagation graph to update the state of each node in the spatiotemporal propagation graph.

[0105] The lane-level arrival vehicle sequences of each node in the updated spatiotemporal propagation graph are extracted and organized according to the intersection affiliation to generate the regional traffic flow prediction information.

[0106] In the regional coordination and control node, it is necessary to perform cross-intersection temporal integration and propagation delay correction on the lane-level arrival vehicle sequences that converge from multiple edge computing nodes. To this end, based on the road network lane-level topology, each lane in the road network with independent travel direction and traffic attributes is abstracted as a node in the graph. Node attributes include the intersection number to which the lane belongs, the entrance / exit direction identifier, the lane type (straight, left turn, right turn, etc.), and the current number of vehicles in the queue. The connection relationship between lanes—that is, the next lane a vehicle can enter after exiting one lane—is abstracted as directed edges. The edge weight corresponds to the average travel time of the vehicle on that segment, i.e., the transfer time between lanes. Thus, a spatiotemporal propagation graph covering the entire road network is constructed in the regional coordination and control node. This graph can fully describe the spatial topology and time cost relationship of vehicle flow across intersections in the road network.

[0107] For each moving target in the spatiotemporal propagation graph, based on its current lane position and the turning requirement included in its driving intention, a multi-hop propagation path is determined in the graph, starting from the current lane node, passing through several intermediate lane nodes, and finally arriving at the target lane node. The determination process of the multi-hop propagation path comprehensively considers the constraints of lane connection relationships and turning requirements. For a moving target that needs to turn left at an intersection, it is only allowed to choose the corresponding left-turn lane outgoing edge from the directed edge at that intersection, thus ensuring the consistency between the path and the actual turning behavior. Along the determined multi-hop propagation path, the transfer time of each directed edge is accumulated sequentially to calculate the theoretical time of no delay for each moving target at each intermediate lane node. The theoretical time of no delay represents the earliest time when the moving target arrives at the intermediate lane node under ideal conditions where there are no traffic lights or queues. Its calculation starts from the sum of the current time and the travel time from the current position of the moving target to the first intermediate lane node. Then, the transfer time of each segment is accumulated hop by hop to obtain the theoretical time of no delay for each subsequent node.

[0108] When obtaining the traffic light status at each middle lane node, the current signal phase and timing parameters such as the remaining green light duration and red light duration are read in real time from the intersection signal controller via the 5G network. Simultaneously, the current number of vehicles queuing at that lane node is also read. The calculation of queuing delay time comprehensively considers two factors: firstly, the traffic light waiting delay. When the theoretically no-delay passage time falls within the red light phase interval, the moving target needs to wait until the start of the next green light; this waiting time is the traffic light waiting delay. Secondly, the queue dissipation delay. When there are queuing vehicles at the lane node, the moving target also needs to wait for the queuing vehicles to pass the stop line in sequence. The queue dissipation delay is related to the number of queuing vehicles and the average saturation flow rate per vehicle. Let the number of queuing vehicles at a certain middle lane node be... The average time for a single bicycle to dissipate its occupancy is The queue dissipation delay is... The total queuing delay at the node is obtained by adding the traffic light waiting delay to the queue dissipation delay. This total queuing delay is then added to the theoretical, delay-free arrival time of the moving target at that node to obtain the corrected actual estimated arrival time. This calculation is repeated for all intermediate lane nodes on the multi-hop path to finally obtain the corrected estimated arrival time of the moving target at the downstream intersection entrance, after adjusting for the accumulated queuing delay.

[0109] After calculating the revised estimated arrival times for each moving target, for each downstream intersection entrance lane, all moving targets expected to arrive at that entrance lane are arranged in ascending order of their revised estimated arrival times, thus regenerating the lane-level arrival vehicle sequence for that entrance lane. The reordered lane-level arrival vehicle sequence, compared to the original sequence reported by the edge computing nodes, incorporates the effects of cross-intersection propagation delay and queuing delay, more accurately reflecting the actual timing of vehicle arrivals at downstream intersections. After sorting, the revised estimated arrival times and corresponding moving target identification information are backpropagated to the corresponding downstream lane nodes in the spatiotemporal propagation graph, updating the lane-level arrival vehicle sequence and expected queuing state stored at that node. The purpose of backpropagation is to maintain consistency between the state of each node in the spatiotemporal propagation graph and the latest revised result, avoiding error accumulation caused by state lag in subsequent iterations. The update process is executed node by node from downstream to upstream according to the topological order of the nodes along the path, ensuring that the state updates of upstream nodes can perceive the feedback impact of downstream node queuing changes on propagation delay.

[0110] After the states of each node in the spatiotemporal propagation graph are updated, the lane-level arrival vehicle sequences of each lane node are extracted and organized according to their association with the intersection to which they belong. For multiple approach lane nodes belonging to the same intersection, their lane-level arrival vehicle sequences are aggregated into a complete set of approach lane arrival sequences for that intersection, along with the turning attributes and estimated arrival times of each lane. The set of approach lane arrival sequences for all intersections is then structured according to intersection numbers to form regional-level traffic flow prediction information covering all managed intersections within the road network. This regional-level traffic flow prediction information describes the vehicle arrival patterns of each intersection and each approach lane within a future time window in a time-series, lane-level format. This provides a refined input basis for optimizing the signal timing schemes for each intersection in the subsequent multi-agent game framework, enabling signal timing decisions to be based on traffic flow prediction results that truly reflect the cross-intersection propagation effect, thereby improving the accuracy and adaptability of the globally coordinated signal timing scheme.

[0111] In one optional implementation, based on the regional traffic flow prediction information, the signal timing problem at multiple intersections within the road network is modeled as a multi-agent game framework, including:

[0112] The lane-level arrival vehicle sequences at each intersection in the regional traffic flow prediction information are grouped according to the intersection identifier. The expected arrival time and turning demand of vehicles at each entrance lane of each intersection are extracted as the initial input of the agent's state space, and an independent agent is constructed for each intersection in the road network.

[0113] The combination of the signal phase sequence, the green light duration of each phase, and the phase switching order that each independent agent can choose is defined as the action space of the independent agent.

[0114] For each independent intelligent agent, a revenue indicator calculation rule is constructed. The revenue indicator calculation rule calculates the average delay time of all vehicles in the intersection controlled by the independent intelligent agent as the delay cost item based on the signal phase scheme and phase duration combination selected by the independent intelligent agent.

[0115] The vehicle overflow penalty for adjacent intersections is calculated based on the difference between the vehicle outflow volume of each exit lane of the independent intelligent agent controlled intersection and the vehicle carrying capacity of the corresponding entrance lane of the adjacent independent intelligent agent controlled intersection.

[0116] The delay cost item and the vehicle overflow penalty item at the adjacent intersection are weighted and combined according to a preset weight coefficient to obtain the revenue index of the independent intelligent agent, and a multi-agent game framework including the state space, action space and revenue index of each independent intelligent agent is established.

[0117] After generating regional-level traffic flow prediction information, the signal timing problem at multiple intersections within the road network needs to be incorporated into a unified optimization framework for coordinated solution. The lane-level arrival vehicle sequences at each intersection in the regional-level traffic flow prediction information are grouped according to intersection identifiers. Each intersection corresponds to a set of arrival vehicle sequence records, and each record in the sequence includes the vehicle's expected arrival time, its approach lane number, and its turning requirement (straight, left turn, or right turn). This extracted information is used as the initial input to the state space of the corresponding intersection's agent, constructing an independent agent for each intersection within the road network. These independent agents are logically independent but maintain coupling through an overflow penalty term in the payoff metric, thus forming the basic structure of a multi-agent game framework.

[0118] The state space of the independent agent consists of two parts: one is the initial input from regional-level traffic flow prediction information, including the vehicle arrival sequence of each approach lane within the prediction time window; the other is the dynamic state updated in real time by lane-level traffic situation perception information during the iteration process, including the current queue length of each approach lane, the current remaining capacity of each exit lane, and the current signal phase status of adjacent intersections. The action space of the independent agent is defined as the combination of selectable signal phase sequences, green light durations for each phase, and phase switching sequences. Specifically, the signal phase sequence refers to the arrangement order of conflicting phases at the intersection within a signal cycle, the green light duration refers to the green light duration allocated to each phase within the current signal cycle, and the phase switching sequence specifies the transition method between phases. The elements in the action space are jointly determined by the phase sequence, the green light duration vector for each phase, and the switching sequence. The combination of these three constitutes a complete signal timing scheme, which serves as the action that the independent agent can choose in each decision step.

[0119] When constructing revenue calculation rules for each independent agent, the delay cost is calculated using the signal phase scheme and phase duration combination selected by the independent agent as input. For each vehicle at the intersection, its delay time is defined as the difference between the actual time required to pass through the intersection under the current signal timing scheme and the ideal unobstructed passage time. The average delay time of all vehicles at the intersection under the current signal timing scheme is obtained by averaging the delay times of all vehicles at the intersection, denoted as . This is considered as a delay cost item. The magnitude of this value directly reflects the impact of the current signal timing scheme on the intersection's traffic efficiency. The larger the value, the more significant the obstruction of vehicle passage by the timing scheme.

[0120] The calculation of the vehicle overflow penalty at adjacent intersections requires considering both the vehicle outflow volume of each exit lane at the current intersection and the vehicle carrying capacity of the corresponding entrance lanes at adjacent intersections. Let the vehicle outflow volume of a certain exit lane within the prediction time window be... The vehicle carrying capacity of the adjacent intersection entrance lane is The difference between the two If the value is positive, it indicates that the number of vehicles exiting from the current intersection to the adjacent intersection exceeds the capacity of the adjacent intersection, resulting in an overflow risk. The overflow amounts in all exit lanes of the intersection are summed to obtain the total overflow penalty for that intersection. .when When the overflow penalty is not included in the corresponding direction, the overflow penalty is taken as the positive part. The number of physical lanes at the entrance of adjacent intersections, the maximum queuing capacity of a single lane, and the current remaining available capacity of adjacent intersections are all determined together, and are dynamically adjusted in each iteration as the status of adjacent intersections is updated.

[0121] Delay cost item Penalties for vehicles overflowing at adjacent intersections The independent agent's revenue index is obtained by weighting and combining the results according to preset weighting coefficients. , expressed as ,in As a weighting factor for delay costs, The overflow penalty weight coefficient is used. The negative sign of the profit indicator is because the optimization objective is to minimize delay and overflow. The minimization problem is transformed into the form of maximizing profit to adapt to the solution mechanism of the game framework. and The value is preset according to the actual operating needs of the road network. When the road network is in a high saturation state, it can be appropriately increased. The proportion of [something] is used to suppress spillover propagation, and can be increased when the overall load of the road network is low. The proportion of resources allocated to reducing vehicle delays should be prioritized.

[0122] Based on the aforementioned delay cost, overflow penalty, and revenue indicators, the state space, action space, and revenue indicators of all independent agents at intersections within the road network are integrated to establish a complete multi-agent game framework. Each independent agent simultaneously makes strategy choices within this framework. Each agent's payoff depends not only on its own chosen action but also on the actions of neighboring agents (reflected in the overflow penalty's dependence on the capacity of adjacent intersections). This coupling relationship ensures that signal timing decisions at each intersection are no longer isolated but are jointly optimized under global coordination constraints, providing a complete game modeling foundation for subsequent iterative solutions to the optimal strategy combination.

[0123] After the multi-agent game framework is established, the state space of each independent agent is updated in real time with lane-level traffic situation perception information in each iteration, ensuring that the traffic state information relied upon by each agent during the game solution process remains synchronized with the actual operating state of the road network. By continuously updating the state space and recalculating the payoff indicators of each agent during the iteration process, the optimal strategy combination for each intersection is gradually approximated, ultimately generating a globally coordinated signal timing scheme and achieving coordinated optimization of traffic signals at the road network level.

[0124] In one optional implementation, the optimal strategy combination for each intersection in the multi-agent game framework is iteratively calculated. During the iteration process, the state space of each agent is updated in real time using the lane-level traffic situation awareness information to obtain a globally coordinated signal timing scheme, including:

[0125] In the multi-agent game framework, a policy mapping mechanism is constructed for each agent. The policy mapping mechanism outputs the signal phase scheme and phase duration combination of the agent in the current round based on the agent's current state and the historical actions of neighboring agents. Each agent calculates the agent's reward index value based on the signal phase scheme and phase duration combination output by the policy mapping mechanism, and then backpropagates the reward index value to the policy mapping mechanism to adjust the mapping parameters of the policy mapping mechanism.

[0126] The real-time vehicle arrival rate and real-time queue length of each intersection entrance lane are extracted from the lane-level traffic situation perception information. The initial state of each agent is corrected based on the real-time vehicle arrival rate and real-time queue length. The corrected initial state is input into the strategy mapping mechanism of each agent to recalculate the signal phase scheme and phase duration combination. The output calculation of the strategy mapping mechanism and the back propagation process of the revenue index are iteratively executed until the revenue index values ​​of each agent converge.

[0127] The signal phase schemes and phase duration combinations output by the policy mapping mechanisms of each agent after convergence are merged according to the intersection identifiers to generate a globally coordinated signal timing scheme.

[0128] In the multi-agent game framework, each intersection corresponds to an independent agent. Each agent maps its current state and the historical actions of its neighboring agents to a signal phase scheme and phase duration combination for the current round through a policy mapping mechanism. The core of the policy mapping mechanism is a parameterized mapping function, whose input includes two parts: one part is the agent's own state vector, covering information such as the queue length of each approach lane, vehicle arrival rate, and current phase execution progress; the other part is the signal phase scheme and phase duration combination output by neighboring agents in the previous round, which is incorporated into the current agent's decision input as a coordination signal. The two parts of the input are concatenated and fed into the policy mapping mechanism to output the activation order of each signal phase in the current round and the corresponding green light duration allocation scheme.

[0129] The mapping parameters of the strategy mapping mechanism are adjusted through backpropagation of the payoff index. After obtaining the signal phase scheme and phase duration combination for the current round, each agent obtains the payoff index value according to the payoff index calculation method defined in claim 6. ,in Index the intersection by number. The gradient of the mapping parameters of the policy mapping mechanism is calculated, and the mapping parameters are updated according to the upward direction of the gradient. This allows the policy mapping mechanism to output signal phase schemes and phase duration combinations that result in higher payoff values ​​in subsequent rounds. In a multi-agent game framework, since the historical actions of neighboring agents change synchronously with the iteration process, each agent's policy mapping mechanism needs to treat the historical actions of neighboring agents as exogenous inputs when updating the mapping parameters, without calculating their gradients. This ensures that each agent optimizes independently within its own policy space, while implicit coordination is achieved through the transmission of historical actions from neighboring agents.

[0130] During the iteration process, lane-level traffic situation awareness information is continuously used to correct the initial state of each agent. Specifically, real-time vehicle arrival rates at each intersection's approach lanes are extracted from the lane-level traffic situation awareness information. With real-time queue length ,in Index the import lane number. Before the start of each iteration, [the following will be done]. and The state vector of the corresponding agent is written to replace the outdated state components from the previous round, thus correcting the initial state. The corrected initial state is then re-inputted into the policy mapping mechanism of each agent to drive the calculation of the signal phase scheme and phase duration combination for the new round. This mechanism ensures the dynamic coupling between the iterative process and the actual traffic conditions of the road network, avoiding policy mismatch problems caused by traffic flow fluctuations.

[0131] Real-time vehicle arrival rate The calculation relies on the number of new vehicles detected at each approach lane per unit time in lane-level traffic situational awareness information, and is obtained by smoothing historical arrival events through a sliding time window. Real-time queue length The system calculates the queue length in real time based on the number of vehicles waiting behind the stop line at each entrance lane. This queue length is then normalized by considering the physical capacity of each lane in the lane-level topology, ensuring comparability of entrance lanes with different capacities in the state vector. The normalized real-time queue length is shown below. satisfy ,in For the first The first intersection The maximum queuing capacity of each entrance lane.

[0132] After each iteration, the change in the profit index of each agent relative to the previous iteration is calculated. When all agents satisfy At that time, it was determined that the return indicators had converged, among which This is a preset convergence threshold. If it falls within the preset maximum iteration round... If the convergence condition is not met within the current round, the output of the policy mapping mechanism of each agent in the current round is used as the final result to prevent the iteration process from failing to terminate due to continuous fluctuations in traffic flow. Convergence threshold With the maximum number of iterations Pre-configured based on road network size and real-time requirements; this can be appropriately relaxed when the road network is large. or reduce To ensure the timely delivery of the solution.

[0133] After the revenue indicators converge, the strategy mapping mechanism of each agent outputs the signal phase scheme and phase duration combination for the current round, and merges them according to the intersection identifier. The merging process indexes the phase order list and corresponding green light duration sequence output by each agent with the intersection identifier as the key, forming a set of signal timing schemes covering all controlled intersections in the road network. For intersection groups in the road network that are physically adjacent and have strong phase coordination constraints, the green wave bandwidth constraints between adjacent intersections also need to be checked during the merging stage: if the phase offset of a pair of adjacent intersections exceeds the allowable range for green wave coordination, the phase start time of the intersection with the smaller offset is finely adjusted, with the adjustment range not exceeding 10% of the duration of a single signal cycle, in order to improve the green wave coordination effect of the trunk line without significantly losing the independent revenue of each intersection.

[0134] After consolidation, the globally coordinated signal timing scheme is distributed to the signal controllers at each intersection, indexed by intersection identifiers. The distributed information includes the activation sequence of each phase, the duration of green lights, yellow lights, and all-red lights, as well as the start time offset of each intersection's signal cycle. Upon receiving the timing scheme, each intersection's signal controller switches its current signal phase according to the scheme, completing the implementation of the globally coordinated signal timing scheme. When the next prediction time window arrives, lane-level traffic situational awareness information is updated, and the aforementioned iterative calculation process is retried, achieving rolling optimization of the signal timing scheme.

[0135] A second aspect of this invention provides an intelligent traffic signal optimization system integrating BeiDou-3 and 5G, comprising:

[0136] The positioning fusion unit is used to cross-verify the location information provided by BeiDou-3 with the communication connection status information provided by 5G network for each moving target in the road network. When the location information is found to have multipath interference or obstruction that causes accuracy degradation, the unit uses the signal arrival time difference and signal arrival angle difference in the communication connection status information to perform auxiliary positioning correction and generate a fused positioning result.

[0137] The lane perception unit is used to identify the lane position and driving intention of each moving target based on the fused positioning result and the road network lane-level topology, and obtain lane-level traffic situation perception information.

[0138] The traffic flow prediction unit is used to distribute the lane-level traffic situation perception information through the edge computing nodes of the 5G network, predict the lane-level arrival vehicle sequence of each approach lane in the future time window on the edge computing nodes corresponding to each intersection, and aggregate it to the regional coordination control node to generate regional traffic flow prediction information.

[0139] The signal timing unit is used to model the signal timing problem of multiple intersections in the road network as a multi-agent game framework based on the regional traffic flow prediction information, iteratively calculate the optimal strategy combination of each intersection in the multi-agent game framework, and use the lane-level traffic situation perception information to update the state space of each agent in real time during the iteration process to obtain a globally coordinated signal timing scheme.

[0140] A third aspect of the present invention provides an electronic device, comprising:

[0141] processor;

[0142] Memory used to store processor-executable instructions;

[0143] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0144] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0145] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent traffic signal optimization by fusing Beidou-3 and 5G, characterized in that, include: For each moving target within the road network, the location information provided by BeiDou-3 is cross-validated with the communication connection status information provided by the 5G network. When the location information is found to have multipath interference or obstruction that causes accuracy degradation, the signal arrival time difference and signal arrival angle difference in the communication connection status information are used for auxiliary positioning correction to generate a fused positioning result. Based on the fusion positioning results and the road network lane-level topology, the lane position and driving intention of each moving target are identified to obtain lane-level traffic situation awareness information; The lane-level traffic situation awareness information is distributed through the edge computing nodes of the 5G network. The lane-level arrival vehicle sequence of each approach lane in the future time window is predicted on the edge computing nodes corresponding to each intersection and then aggregated to the regional coordination and control node to generate regional traffic flow prediction information. Based on the regional traffic flow prediction information, the signal timing problem of multiple intersections in the road network is modeled as a multi-agent game framework. The optimal strategy combination of each intersection in the multi-agent game framework is iteratively calculated. During the iteration process, the state space of each agent is updated in real time using the lane-level traffic situation perception information to obtain a globally coordinated signal timing scheme.

2. The method of claim 1, wherein, When the location information is found to have multipath interference or occlusion causing accuracy degradation, auxiliary positioning correction is performed using the signal arrival time difference and signal arrival angle difference in the communication connection status information to generate a fused positioning result, including: Based on the signal arrival time difference between multiple 5G base stations and the mobile target in the communication connection status information, the distance difference between the mobile target and different 5G base stations is calculated. A hyperboloid is constructed with any two 5G base stations as the focus and the distance difference as the parameter. Multiple hyperboloids form a set of hyperbolic spatial positioning constraints. The arrival angle information of the uplink signal of the mobile target received by multiple 5G base stations is extracted, the signal arrival angle difference between different 5G base stations is calculated, and the arrival angle information is verified for consistency based on the spatial geometric relationship between the signal arrival angle difference and each 5G base station. Abnormal angle measurement values ​​that cause the signal arrival angle difference to not meet the spatial geometric consistency constraint are eliminated. Based on the arrival angle information after consistency verification, the ray direction of each 5G base station pointing to the moving target is determined. A spatial ray is constructed with the location of each 5G base station as the starting point and the ray direction as the direction vector. Multiple spatial rays form a set of ray intersection spatial positioning constraints. Calculate the intersection region in space between the hyperbolic spatial positioning constraint set and the ray intersection spatial positioning constraint set, and take the geometric center of the intersection region as the auxiliary positioning coordinates; The deviation between the location information and the auxiliary positioning coordinates is calculated. A correction weighting coefficient is determined based on the degree of accuracy degradation of the location information. The deviation is then weighted and compensated using the correction weighting coefficient and added to the location information to generate the fused positioning result.

3. The method of claim 1, wherein, Based on the fused localization results and the road network lane-level topology, the lane position and driving intention of each moving target are identified, resulting in lane-level traffic situation awareness information, including: The fused positioning results are projected onto the road network lane-level topology, and the lane position of each moving target is determined by calculating the vertical distance between the spatial coordinates of each moving target and the centerline of each lane. Extract the fusion positioning results of each moving target in a continuous time series, construct multiple accessible lane connection paths starting from the current lane position based on the lane-level topology of the road network, calculate the fitting deviation between the spatial trajectory formed by the continuous fusion positioning results and each accessible lane connection path, select the accessible lane connection path with the smallest fitting deviation as the driving path, determine the end lane of the driving path as the target lane, and identify the driving intention of each moving target. The lane position is associated with the driving intention, and the number of moving targets in each lane, the speed distribution of moving targets in each lane, and the lane changing behavior characteristics of moving targets are statistically analyzed to generate the lane-level traffic situation awareness information.

4. The method of claim 1, wherein, The lane-level traffic situation awareness information is processed in a distributed manner through the edge computing nodes of the 5G network. At the edge computing nodes corresponding to each intersection, the lane-level arrival vehicle sequences for each approach lane within a future time window are predicted and aggregated to the regional coordination and control node, generating regional-level traffic flow prediction information, including: The lane-level traffic situation awareness information is distributed to the edge computing nodes of the corresponding intersections according to the intersection affiliation of the lane position of each moving target. The lane position, driving intention and moving speed of each moving target are extracted at each edge computing node. The approach lane to which each moving target will arrive is determined based on the lane position and driving intention of each moving target. The estimated arrival time of each moving target at the corresponding approach lane is calculated based on the moving speed of each moving target and the spatial distance between the current lane position of each moving target and the corresponding approach lane. Each moving target is sorted according to its expected arrival time on each approach lane within a future time window, forming a lane-level arrival vehicle sequence for each approach lane within the future time window. During the sorting process, the turning requirements of each moving target after arriving at the approach lane are identified based on the driving intention of each moving target, and the turning requirements are marked in the lane-level arrival vehicle sequence. The lane-level arrival vehicle sequences generated on each edge computing node are aggregated to the regional coordination control node. In the regional coordination control node, based on the road network spatiotemporal propagation map, the cross-intersection propagation delay calculation and timing correction are performed on the lane-level arrival vehicle sequences of multiple intersections to generate the regional traffic flow prediction information.

5. The method according to claim 4, characterized in that, In the regional coordinated control node, based on the road network spatiotemporal propagation map, the cross-intersection propagation delay calculation and timing correction are performed on the lane-level arrival vehicle sequences at multiple intersections to generate the regional-level traffic flow prediction information, including: Each lane in the lane-level topology of the road network is abstracted as a node, and the connection relationship between lanes and the transfer time of vehicles between lanes are abstracted as directed edges. A spatiotemporal propagation graph of the road network is constructed in the regional coordination control node. In the spatiotemporal propagation diagram, a multi-hop propagation path from the current lane position to the final target lane is constructed for each moving target. The lane transfer direction of each moving target at each intersection is determined according to the steering requirements. The theoretical time of passage of each moving target at each intermediate lane node without delay is calculated along the multi-hop propagation path. The system obtains the traffic light status and the current number of vehicles in the queue at each intermediate lane node, and calculates the queuing delay time of each moving target at each intermediate lane node by combining the aforementioned no-delay theory. The queuing delay time is then added to the estimated arrival time of each moving target at the downstream intersection entrance. The estimated arrival times after accumulating queuing delays are reordered according to time sequence for the lane-level arrival vehicle sequences at the downstream intersection entrances, and then propagated back to the corresponding downstream lane nodes in the spatiotemporal propagation graph to update the state of each node in the spatiotemporal propagation graph. The lane-level arrival vehicle sequences of each node in the updated spatiotemporal propagation graph are extracted and organized according to the intersection affiliation to generate the regional traffic flow prediction information.

6. The method of claim 1, wherein, Based on the aforementioned regional traffic flow prediction information, the signal timing problem at multiple intersections within the road network is modeled as a multi-agent game framework, including: The lane-level arrival vehicle sequences at each intersection in the regional traffic flow prediction information are grouped according to the intersection identifier. The expected arrival time and turning demand of vehicles at each entrance lane of each intersection are extracted as the initial input of the agent's state space, and an independent agent is constructed for each intersection in the road network. The combination of the signal phase sequence, the green light duration of each phase, and the phase switching order that each independent agent can choose is defined as the action space of the independent agent. For each independent intelligent agent, a revenue indicator calculation rule is constructed. The revenue indicator calculation rule calculates the average delay time of all vehicles in the intersection controlled by the independent intelligent agent as the delay cost item based on the signal phase scheme and phase duration combination selected by the independent intelligent agent. The vehicle overflow penalty for adjacent intersections is calculated based on the difference between the vehicle outflow volume of each exit lane of the independent intelligent agent controlled intersection and the vehicle carrying capacity of the corresponding entrance lane of the adjacent independent intelligent agent controlled intersection. The delay cost item and the vehicle overflow penalty item at the adjacent intersection are weighted and combined according to a preset weight coefficient to obtain the revenue index of the independent intelligent agent, and a multi-agent game framework including the state space, action space and revenue index of each independent intelligent agent is established.

7. The method of claim 1, wherein, The optimal strategy combination for each intersection in the multi-agent game framework is calculated iteratively. During the iteration process, the state space of each agent is updated in real time using the lane-level traffic situation awareness information to obtain a globally coordinated signal timing scheme, including: In the multi-agent game framework, a policy mapping mechanism is constructed for each agent. The policy mapping mechanism outputs the signal phase scheme and phase duration combination of the agent in the current round based on the agent's current state and the historical actions of neighboring agents. Each agent calculates the agent's reward index value based on the signal phase scheme and phase duration combination output by the policy mapping mechanism, and then backpropagates the reward index value to the policy mapping mechanism to adjust the mapping parameters of the policy mapping mechanism. The real-time vehicle arrival rate and real-time queue length of each intersection entrance lane are extracted from the lane-level traffic situation perception information. The initial state of each agent is corrected based on the real-time vehicle arrival rate and real-time queue length. The corrected initial state is input into the strategy mapping mechanism of each agent to recalculate the signal phase scheme and phase duration combination. The output calculation of the strategy mapping mechanism and the back propagation process of the revenue index are iteratively executed until the revenue index values ​​of each agent converge. The signal phase schemes and phase duration combinations output by the policy mapping mechanisms of each agent after convergence are merged according to the intersection identifiers to generate a globally coordinated signal timing scheme.

8. The intelligent traffic signal optimization system fusing Beidou-3 and 5G, for implementing the method of any one of claims 1-7, characterized in that, include: The positioning fusion unit is used to cross-verify the location information provided by BeiDou-3 with the communication connection status information provided by 5G network for each moving target in the road network. When the location information is found to have multipath interference or obstruction that causes accuracy degradation, the unit uses the signal arrival time difference and signal arrival angle difference in the communication connection status information to perform auxiliary positioning correction and generate a fused positioning result. The lane perception unit is used to identify the lane position and driving intention of each moving target based on the fused positioning result and the road network lane-level topology, and obtain lane-level traffic situation perception information. The traffic flow prediction unit is used to distribute the lane-level traffic situation perception information through the edge computing nodes of the 5G network, predict the lane-level arrival vehicle sequence of each approach lane in the future time window on the edge computing nodes corresponding to each intersection, and aggregate it to the regional coordination control node to generate regional traffic flow prediction information. The signal timing unit is used to model the signal timing problem of multiple intersections in the road network as a multi-agent game framework based on the regional traffic flow prediction information, iteratively calculate the optimal strategy combination of each intersection in the multi-agent game framework, and use the lane-level traffic situation perception information to update the state space of each agent in real time during the iteration process to obtain a globally coordinated signal timing scheme.

9. An electronic device, comprising: include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon computer program instructions, wherein, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images