A highway traffic flow state adaptive control method using edge node perception

Abstract

The application discloses a highway traffic flow state adaptive control method using edge node perception and relates to the technical field of intelligent traffic control. The method comprises the following steps: S1, collecting traffic flow data of each intersection through a set sensor network and obtaining local utility values of each intersection; S2, determining marginal external costs according to traffic flow data of intelligent agents arranged at each intersection, combining the marginal external costs and the local utility values, and constructing a comprehensive decision function; and S3, setting a maximum comprehensive decision function corresponding to each intelligent agent according to the comprehensive decision function, and obtaining an optimal signal timing scheme of each intersection through a distributed optimization algorithm according to a set signal timing physical constraint condition. The method realizes autonomous collaborative control independent of a central server, guarantees intersection autonomy, and realizes overall optimization of regional traffic flow.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic control technology, specifically to an adaptive control method for highway traffic flow state using edge node sensing. Background Technology

[0002] In recent years, the development of Intelligent Transportation Systems (ITS) has provided new technological means for traffic flow optimization. Traffic control platforms based on centralized data processing can integrate multi-source traffic information to achieve regional collaborative control. However, such systems typically rely on a central server for data aggregation and analysis, resulting in problems such as high communication latency, heavy data processing load, and slow system response, making it difficult to meet the stringent real-time requirements of traffic control. Especially during peak traffic periods or network congestion, data transmission delays can lead to delayed control decisions, weakening the control effect.

[0003] Chinese invention patent CN110021168A discloses a hierarchical decision-making method for real-time intelligent traffic management under the Internet of Vehicles (IoV). This method utilizes local traffic environment information at intersections to make real-time local traffic control decisions. On a larger time scale, a cloud processing center observes the global traffic environment to improve overall traffic performance. Learning techniques are used to estimate key parameters of the decision-making system at each intersection, and these parameters are fed back to each intersection, allowing them to adjust their decision-making algorithms based on the feedback parameters, thus achieving collaborative intelligent traffic management. This invention proposes a hierarchical traffic management decision-making mechanism at different time scales, achieving intelligent traffic control that balances global optimality and real-time performance.

[0004] While the aforementioned intelligent traffic management methods can indirectly coordinate the objective function of intersections by relying on cloud-based weight adjustments, thereby achieving globally optimal and real-time intelligent traffic control, they can only adjust based on the signal controller's own data. They cannot coordinate the mutual influence between signal controllers, resulting in a one-sided and limited overall optimization. Therefore, they cannot achieve overall system optimization without sacrificing the autonomy of the node signal controllers. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive control method for highway traffic flow state using edge node sensing, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control method for highway traffic flow state using edge node sensing, comprising:

[0007] S1: Determine local utility: Collect traffic flow data at each intersection through the set sensor network and obtain the local utility value of each intersection;

[0008] S2: Determine the decision function: Based on the traffic flow data of the agents at each intersection, determine the marginal external cost, and combine the marginal external cost with the local utility value to construct a comprehensive decision function, including:

[0009] S2.1: Determine marginal external costs: Based on traffic flow data between different agents, determine the cost prediction value for each agent through the marginal external cost model;

[0010] S2.2: Constructing a comprehensive decision function: Using the local utility value and the cost prediction value, a comprehensive decision function is constructed;

[0011] S3: Determine the signal timing scheme: Based on the comprehensive decision function, set the maximum comprehensive decision function corresponding to each intelligent agent, and obtain the optimal signal timing scheme for each intersection through a distributed optimization algorithm according to the set physical constraints of signal timing.

[0012] Furthermore, the local utility value of each intersection is obtained, including:

[0013] S1.1: Data Acquisition: A sensor network is set up using millimeter-wave radar, geomagnetic coils, and cameras to collect radar data, coil data, and video data. Preprocessed traffic flow data is obtained through data cleaning and standardization.

[0014] S1.2: Utility Calculation: The preprocessed traffic flow data is used as the input to the prediction model, and the corresponding delay reduction potential score is obtained as the output. Based on the delay reduction potential score, the local utility value is determined.

[0015] Furthermore, the cost prediction for each agent is determined, including:

[0016] S2.1.1: Acquire interactive data: Based on the preprocessed traffic flow data, determine the congestion index, received traffic flow index, and phase state discretization at the downstream intersection of each agent;

[0017] S2.1.2: Cost Calculation: Discretize the congestion index, the receiving traffic flow index, and the phase state as input to the marginal external cost model, and output the corresponding cost prediction value.

[0018] Furthermore, the congestion index, received traffic flow index, and phase state discretization at each downstream intersection of the intelligent agent are determined, including:

[0019] S2.1.1.1: Determine the congestion index: Determine the number of vehicles passing through in the current cycle through coil data, determine the maximum number of vehicles passing through through video data, and obtain the real-time saturation based on the number of vehicles passing through in the current cycle and the maximum number of vehicles passing through. At the same time, the real-time saturation is processed for privacy to determine the congestion index.

[0020] S2.1.1.2: Determine the receiving traffic flow index: Based on the average headway, measured headway, and effective number of lanes at the current intersection, determine the ideal traffic capacity and the current actual traffic capacity, and determine the traffic capacity index by the ratio between the ideal traffic capacity and the current actual traffic capacity;

[0021] S2.1.1.3: Determine phase state discretization: Divide the signal period of the traffic lights by setting the basic time window and set the corresponding discretization label.

[0022] Furthermore, based on the average headway, effective lane number, and vehicle type correction coefficient of the current intersection, the ideal traffic capacity is set, and based on the measured headway, effective lane number, and total factor correction coefficient of the current intersection, the current actual traffic capacity is set, whereby the total factor correction coefficient includes vehicle type correction coefficient, weather correction coefficient, and road correction coefficient.

[0023] Furthermore, the total factor correction factor is set using the vehicle model correction factor, weather correction factor, and road correction factor, including:

[0024] S2.1.1.2.1: Set vehicle model coefficient: Divide the vehicle models according to the vehicle length, and set the corresponding vehicle model weight according to the divided vehicle models. At the same time, combine the vehicle model weight with the number of vehicles to determine the number of each vehicle model, and determine the vehicle model correction coefficient according to the ratio between the number of large and medium vehicle models and the total number of vehicle models.

[0025] S2.1.1.2.2: Set weather coefficients: By obtaining precipitation intensity data and visibility data, set precipitation weights and visibility weights, and use the precipitation weights and visibility weights as inputs to the LSTM network model, and output the corresponding weather correction coefficients;

[0026] S2.1.1.2.3: Setting the road coefficient: Obtain road slope data using a laser rangefinder, and set the road correction coefficient based on the road slope data;

[0027] S2.1.1.2.4: Determine the total factor coefficient: Combine the vehicle type correction coefficient, weather correction coefficient, and road correction coefficient to determine the total factor correction coefficient.

[0028] Furthermore, vehicle model weights are set according to vehicle type: the weight for small cars is set to 1, the weight for medium-sized cars is set to 1.3, and the weight for large cars is set to 2.

[0029] The precipitation weight is set according to the precipitation intensity, specifically: the weight is set to 0.9 when the precipitation intensity is 0-2.5 mm / h, the weight is set to 0.85 when the precipitation intensity is 2.5-10 mm / h, and the weight is set to 0.8 when the precipitation intensity is greater than 10 mm / h.

[0030] The visibility weights are set based on the visibility data, specifically: the weight is set to 0.8 when the visibility range is 200-500m, the weight is set to 0.75 when the visibility range is 100-200m, and the weight is set to 0.7 when the visibility range is less than 100m.

[0031] Furthermore, based on the current number of lanes and headway at the intersection, a basic traffic capacity is set, and the basic traffic capacity is combined with the set traffic weight to determine the maximum number of vehicles that can pass.

[0032] Based on the actual influencing factors at the current intersection, the traffic weights are set. These actual influencing factors include weather, vehicle type ratio, and road construction. The traffic weight corresponding to weather is set to 0.85, the traffic weight corresponding to vehicle type ratio is set to 0.92, and the traffic weight corresponding to road construction is set to 0.7.

[0033] Furthermore, based on pedestrian crossing time and safety buffer time, the minimum green light time in the signal timing physical constraints is set, and based on the minimum green light time, common signal cycle, and phase switching loss time, the maximum green light time in the signal timing physical constraints is set.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] Firstly, this invention quantifies the impact of local decisions on congestion at adjacent intersections by setting a marginal external cost model, and achieves collaborative optimization between intersections through a comprehensive decision function. This ensures the autonomy of intersections while avoiding the decline in global performance caused by local optimization.

[0036] Secondly, this invention uses distributed processing of multimodal sensor data at edge nodes, which not only avoids the communication delay problem of traditional centralized systems and enables rapid perception of traffic flow changes, but also improves the response speed to sudden congestion or fluctuations in traffic demand by dynamically adjusting signal timing.

[0037] Thirdly, this invention combines lightweight gradient boosting trees with deep reinforcement learning models, and can dynamically adjust traffic capacity based on real-time traffic data and external environmental factors, thereby improving the accuracy and adaptability of signal timing. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the adaptive control method for highway traffic flow state in this invention.

[0039] Figure 2 This is a schematic diagram of the process for obtaining local utility values ​​in this invention;

[0040] Figure 3 This is a comparison chart of the signal timing optimization effects in this invention;

[0041] Figure 4 This is a comparison chart of resource consumption in this invention. Detailed Implementation

[0042] 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.

[0043] Existing intelligent traffic management methods can only adjust based on the signal controller's own data, failing to coordinate the mutual influence between signal controllers. This results in a one-sided and limited overall optimization, making it impossible to achieve system-wide optimization without sacrificing the autonomy of individual signal controllers. In contrast, this application's technical solution deploys millimeter-wave radar, geomagnetic coils, and cameras at intersections to form a multimodal sensor network. This network collects traffic flow data in real time and calculates corresponding local utility values. Simultaneously, through collaborative communication between edge agents and a marginal external cost model, local utility values ​​and external costs are fused to form a comprehensive decision function. Under set physical constraints, a distributed optimization algorithm solves for the optimal signal timing scheme at each intersection. This achieves autonomous collaborative control without relying on a central server, ensuring intersection autonomy while optimizing regional traffic flow.

[0044] Example 1

[0045] refer to Figures 1-4 This embodiment provides an adaptive control method for highway traffic flow state using edge node sensing. The adaptive control method for highway traffic flow state includes the following steps:

[0046] Step S1: Determine local utility. This involves collecting real-time traffic flow data from the sensor network connected to each edge agent at each intersection, and then using a pre-defined prediction model to determine the corresponding local utility value based on this data. Specifically:

[0047] Step S1.1: Data Acquisition. A sensor network is set up using millimeter-wave radar, geomagnetic coils, and cameras to form a multimodal perception layer. Specifically, millimeter-wave radar, geomagnetic coils, and cameras are installed at each intersection. The millimeter-wave radar detects vehicle speeds within a 200-meter range, the geomagnetic coils calculate lane-level traffic flow, and the cameras identify queue lengths. In other words, the corresponding radar data, coil data, and video data are acquired.

[0048] Furthermore, the acquired radar data, coil data, and video data undergo data cleaning and standardization. Specifically, sliding window mid-range filtering is used to eliminate false radar detections. Simultaneously, the cleaned traffic flow data (i.e., radar data, coil data, and video data) is subjected to Min-Max normalization to obtain preprocessed traffic flow data. It is worth noting that the data cleaning and standardization processes, i.e., Min-Max normalization, in this embodiment are conventional data processing methods, and therefore will not be specifically described in this embodiment.

[0049] Step S1.2: Utility Calculation. This involves using the preprocessed traffic flow data obtained from the agents at each intersection as described in Step S1.1, and then using this preprocessed traffic flow data as input to a pre-defined prediction model, such as a lightweight gradient boosting tree model, to output the corresponding delay reduction potential score.

[0050] Furthermore, based on the obtained delay reduction potential score, the corresponding local utility value is obtained through Sigmoid function transformation, specifically:

[0051]

[0052] in: Let be the local utility value of the i-th intersection. It is a natural constant. The potential score is lowered to account for delays.

[0053] During the implementation process, the obtained delay reduction potential score was 68, and the corresponding local utility value was -0.224.

[0054] Step S2: Determine the decision function. This involves determining the marginal external cost based on the communication data between the agents at each intersection, and then combining this determined marginal external cost with the local utility value obtained in step S1.2 to determine the corresponding comprehensive decision function. Specifically:

[0055] Step S2.1: Determine the marginal external cost. This involves determining the corresponding cost prediction value based on the communication interaction data obtained between different agents and the constructed marginal external cost model. Specifically:

[0056] Step S2.1.1: Obtain interactive data. This involves determining the congestion index, received traffic flow index, and phase state discretization at the downstream intersections of each agent based on the preprocessed traffic flow data obtained from the agents at each intersection. Specifically:

[0057] Step S2.1.1.1: Determine the congestion index. This involves determining the number of vehicles passing through in the current cycle using loop detector data, determining the maximum number of vehicles that can pass through using video data, and combining the determined number of vehicles passing through in the current cycle with the maximum number of vehicles that can pass through to obtain the corresponding real-time saturation level. Specifically:

[0058]

[0059] in: Let j be the current traffic saturation level at the j-th intersection. This represents the number of vehicles that actually passed through during the current signal cycle. This represents the maximum number of vehicles that can pass through.

[0060] In this embodiment, the corresponding basic traffic capacity is determined based on the headway and the number of lanes at the current intersection, specifically as follows:

[0061]

[0062] in: Basic number of passes, For the headway, This refers to the number of lanes.

[0063] Furthermore, based on the actual influencing factors at the current intersection, corresponding traffic weights are set, and these weights are combined with the basic traffic capacity to determine the maximum number of vehicles that can pass. In this embodiment, the actual influencing factors at the current intersection include weather (i.e., rainy or snowy days), the proportion of large vehicles (i.e., large trucks), and road construction. Specifically, when the actual influencing factor is weather (i.e., rainy or snowy days), the corresponding traffic weight is set to 0.85. When the actual influencing factor is that the proportion of large vehicles (i.e., large trucks) exceeds 30%, the corresponding traffic weight is set to 0.92. When the actual influencing factor is road construction, the corresponding traffic weight is set to 0.7.

[0064] Furthermore, based on the current traffic saturation obtained at the intersection, this sensitization is processed to obtain the corresponding congestion index, specifically:

[0065]

[0066] in: The saturation of the output after noise. Let j be the current traffic saturation level at the j-th intersection. It is Gaussian noise with a mean of 0 and a standard deviation of 0.05.

[0067] Step S2.1.1.2: Determine the received traffic flow index. This involves determining the ideal traffic capacity based on the average headway and number of effective lanes at the current intersection, and simultaneously determining the current actual traffic capacity based on the measured headway and number of effective lanes at the current intersection. The corresponding traffic capacity index is then obtained by combining the determined ideal and actual traffic capacities. Specifically:

[0068]

[0069] in: This is a traffic capacity index. For ideal traffic capacity, This represents the current actual traffic capacity.

[0070] In this embodiment, the ideal traffic capacity is determined based on the average headway and the number of effective lanes at the current intersection, specifically:

[0071]

[0072] in: For ideal traffic capacity, For the headway, For the number of lanes, This is a correction factor for the vehicle model.

[0073] Furthermore, based on the measured headway and effective number of lanes at the current intersection, the corresponding current actual traffic capacity is determined, specifically as follows:

[0074]

[0075] in: Based on the current actual traffic capacity, To measure the headway, For the number of effective lanes, This is the correction coefficient for the k-th factor.

[0076] Step S2.1.1.3: Determine the phase state discretization. That is, divide the signal period of the traffic lights into segments using the set base time window, and set a corresponding discretization label for each segmented time interval.

[0077] Step S2.1.2: Cost Calculation. The noise-reduced release saturation obtained in step S2.1.1.1, the traffic capacity index obtained in step S2.1.1.2, and the discretized label obtained in step S2.1.1.3 are used as inputs to the marginal external cost model (such as a deep reinforcement learning collaborative model), and the corresponding cost prediction value is obtained as output.

[0078] Step S2.2: Construct the integrated decision function. This involves constructing the corresponding integrated decision function based on the local utility value obtained in step S1.2 and the cost prediction value obtained in step S2.1.2. Specifically:

[0079]

[0080] in: Let be the comprehensive decision function value for the i-th intersection. Let be the local utility value of the i-th intersection. Let be the cost impact weight of the i-th intersection on the j-th neighboring intersection. This represents the additional congestion cost caused by the i-th intersection to the j-th adjacent intersection.

[0081] In the specific implementation process, when the local utility value is -0.35 and the cost prediction value is 1.89, the corresponding comprehensive decision function value is -1.54.

[0082] Step S3: Determine the signal timing scheme. Specifically, based on the comprehensive decision function value obtained in step S2.2, construct the maximum comprehensive decision function for each agent, as follows:

[0083]

[0084] in: Let be the comprehensive decision function value for the i-th intersection. Let be the local utility value of the i-th intersection. Let be the cost impact weight of the i-th intersection on the j-th neighboring intersection. Let i be the additional congestion cost caused by the i-th intersection to the j-th adjacent intersection. Let be the controllable decision variable for the i-th intersection. Let be the set of decision variables for all intersections except the i-th intersection. Let i be the set of adjacent intersections of the i-th intersection. Let j be the other intersection besides the i-th intersection.

[0085] Furthermore, based on the physical constraints corresponding to signal timing, corresponding constraint conditions are set, specifically as follows:

[0086]

[0087] in: Let green light duration be the decision variable for the i-th intersection. Let be the minimum green light time at the i-th intersection. Let i be the maximum green light time at the i-th intersection. Let be the phase offset at the i-th intersection. Let be the phase offset at the j-th intersection. For the maximum permissible phase difference, This is the common signal cycle.

[0088] In this embodiment, the minimum green light time is set according to the pedestrian crossing time and the set safety buffer time, specifically as follows:

[0089]

[0090] in: Let be the minimum green light time at the i-th intersection. For pedestrian crossing time, This is a safety buffer period.

[0091] Furthermore, based on the set common signal cycle and minimum green light time, combined with the set phase switching loss time, the corresponding maximum green light time is determined, specifically as follows:

[0092]

[0093] in: Let i be the maximum green light time at the i-th intersection. For the common signal period, Let be the minimum green light time at the j-th intersection. The time lost due to phase switching , This is the index for the intersection.

[0094] Furthermore, based on the distance between two adjacent intersections and the corresponding speed settings, the corresponding travel time can be obtained, which is the maximum permissible phase difference.

[0095] In other words, by setting physical constraints and maximizing the comprehensive decision function corresponding to the agents at each intersection, the optimal signal timing scheme for each intersection can be obtained by solving the problem through a distributed optimization algorithm.

[0096] refer to Figure 3 It is known that the delay during peak hours is significantly increased by traditional timing control methods, while this solution dynamically adjusts the green light duration and phase difference by determining the final signal timing scheme, thereby reducing the delay during peak hours by 30%-40%.

[0097] refer to Figure 4 It is known that the latency at edge nodes is less than 100 seconds, and bandwidth is only required for communication between adjacent nodes. In contrast, the latency at centralized cloud nodes is greater than 500 milliseconds, and bandwidth requirements are high due to network congestion. Therefore, this solution can reduce communication load through edge computing, making it suitable for large-scale road network deployments.

[0098] Example 2

[0099] This embodiment provides an adaptive control method for highway traffic flow state using edge node perception. The specific implementation method is the same as in Embodiment 1, except that in determining the ideal / current actual traffic capacity, the corresponding correction coefficients include weather coefficients, vehicle type coefficients, and road coefficients. Furthermore, the weather coefficient, vehicle type coefficient, and road coefficient are combined to determine the corresponding total factor correction coefficient. The invention will be illustrated below with specific examples of this embodiment.

[0100] In this embodiment, the weather coefficient, vehicle type coefficient, and road coefficient are combined to determine the corresponding total factor correction coefficient, as follows:

[0101] Step S2.1.1.2.1: Set vehicle type coefficients. This involves setting vehicle type correction coefficients based on the proportion of large vehicles at the current intersection. Specifically, based on the vehicle types at the current intersection, the weight for small vehicles is set to 1, the weight for medium-sized vehicles to 1.3, and the weight for large vehicles to 2. Simultaneously, the number of vehicles of each type is combined with the corresponding set vehicle type weights to determine the corresponding number of vehicle types. Then, based on the ratio between the determined number of large and medium-sized vehicles and the total number of vehicle types, the corresponding vehicle type correction coefficient is determined, specifically as follows:

[0102]

[0103] in: For vehicle model correction factor, For the number of large vehicles, For the number of medium-sized vehicles, For the number of small cars, Weighting for mid-size cars, Weighting for large vehicles Weighting for small cars.

[0104] In this embodiment, vehicles with a length of no more than 5m are designated as small cars, vehicles with a length between 5 and 7m are designated as medium-sized cars, and vehicles with a length greater than 7m are designated as large cars. In the specific implementation, there are 142 small cars, 35 medium-sized cars, and 12 large cars. These are then combined with their respective weights, resulting in the following vehicle equivalents: 142 small cars, 45.5 medium-sized cars, and 24 large cars. This means the corresponding vehicle type correction coefficient is 0.934.

[0105] Step S2.1.1.2.2: Set the weather coefficient. This involves obtaining the corresponding precipitation intensity and visibility data through the National Meteorological Administration's API interface, and setting the corresponding precipitation weight and visibility weight based on the obtained precipitation intensity and visibility data.

[0106] In this embodiment, when the precipitation intensity is 0-2.5 mm / h, the corresponding weight is set to 0.9; when the precipitation intensity is 2.5-10 mm / h, the corresponding weight is set to 0.85; and when the precipitation intensity is greater than 10 mm / h, the corresponding weight is set to 0.8.

[0107] To elaborate further, when the visibility range is 200-500m, the corresponding weight is set to 0.8; when the visibility range is 100-200m, the corresponding weight is set to 0.75; and when the visibility range is less than 100m, the corresponding weight is set to 0.7.

[0108] Specifically, the acquired precipitation intensity and visibility data are used as inputs to the LSTM network model, and the corresponding weather correction coefficients are output. In the actual implementation, when the current environment is heavy rain and dense fog, the weight coefficient of 0.8 corresponding to heavy rain and the weight coefficient of 0.7 corresponding to dense fog are combined, that is, the corresponding weather correction coefficient is 0.56.

[0109] Step S2.1.1.2.3: Set the road coefficient. This involves setting up a laser rangefinder 1.5m from the lane centerline, and establishing measurement points at 50-meter intervals to obtain the corresponding road slope data. Specifically, based on the obtained road slope, set the corresponding road correction coefficient, as follows:

[0110]

[0111] in: This is the road correction factor. This represents the road slope value.

[0112] Step S2.1.1.2.4: Determine the total factor coefficient. That is, based on the vehicle model correction coefficient, weather correction coefficient, and road correction coefficient obtained in steps S2.1.1.2.1-S2.1.1.2.3, combine these coefficients to determine the corresponding total factor correction coefficient, specifically:

[0113]

[0114] in: This is the total factor correction coefficient. This is a weather correction factor. For vehicle model correction factor, This is the road correction factor.

[0115] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.

Claims

1. An adaptive control method for highway traffic flow state using edge node sensing, characterized in that, Including: S1: Determine local utility: Collect traffic flow data at each intersection through the set sensor network and obtain the local utility value of each intersection; S2: Determine the decision function: Based on the traffic flow data of the agents at each intersection, determine the marginal external cost, and combine the marginal external cost with the local utility value to construct a comprehensive decision function, including: S2.1: Determine Marginal External Costs: Based on traffic flow data between different agents, determine the predicted cost for each agent using the marginal external cost model, including: S2.1.1: Acquiring Interaction Data: Based on the preprocessed traffic flow data, determine the congestion index, received traffic flow index, and phase state discretization at the downstream intersections of each agent, including: S2.1.1.1: Determine the congestion index: Determine the number of vehicles passing through in the current cycle through coil data, determine the maximum number of vehicles passing through through video data, and obtain the real-time saturation based on the number of vehicles passing through in the current cycle and the maximum number of vehicles passing through. At the same time, the real-time saturation is processed for privacy to determine the congestion index. S2.1.1.2: Determine the receiving traffic flow index: Based on the average headway, measured headway, and effective number of lanes at the current intersection, determine the ideal traffic capacity and the current actual traffic capacity, and determine the traffic capacity index by the ratio between the ideal traffic capacity and the current actual traffic capacity; S2.1.1.3: Determine phase state discretization: Divide the signal period of the traffic light by setting the basic time window and set the corresponding discretization label; S2.1.2: Cost Calculation: Discretize the congestion index, the receiving traffic flow index, and the phase state as input to the marginal external cost model, and output the corresponding cost prediction value. S2.2: Constructing a comprehensive decision function: Using the local utility value and the cost prediction value, a comprehensive decision function is constructed; S3: Determine the signal timing scheme: Based on the comprehensive decision function, set the maximum comprehensive decision function corresponding to each intelligent agent, and obtain the optimal signal timing scheme for each intersection through a distributed optimization algorithm according to the set physical constraints of signal timing.

2. The adaptive control method for highway traffic flow state using edge node sensing according to claim 1, characterized in that, The local utility value of each intersection is obtained, including: S1.1: Data Acquisition: A sensor network is set up using millimeter-wave radar, geomagnetic coils, and cameras to collect radar data, coil data, and video data. Preprocessed traffic flow data is obtained through data cleaning and standardization. S1.2: Utility Calculation: The preprocessed traffic flow data is used as the input to the prediction model, and the corresponding delay reduction potential score is obtained as the output. Based on the delay reduction potential score, the local utility value is determined.

3. The adaptive control method for highway traffic flow state using edge node sensing according to claim 1, characterized in that, The ideal traffic capacity is set based on the average headway, number of effective lanes, and vehicle type correction coefficient of the current intersection. The actual traffic capacity is set based on the measured headway, number of effective lanes, and total factor correction coefficient of the current intersection. The total factor correction coefficient includes vehicle type correction coefficient, weather correction coefficient, and road correction coefficient.

4. The adaptive control method for highway traffic flow state using edge node sensing according to claim 3, characterized in that, The total factor correction factor is set using the vehicle model correction factor, weather correction factor, and road correction factor, including: S2.1.1.2.1: Set vehicle model coefficient: Divide the vehicle models according to the vehicle length, and set the corresponding vehicle model weight according to the divided vehicle models. At the same time, combine the vehicle model weight with the number of vehicles to determine the number of each vehicle model, and determine the vehicle model correction coefficient according to the ratio between the number of large and medium vehicle models and the total number of vehicle models. S2.1.1.2.2: Set weather coefficients: By obtaining precipitation intensity data and visibility data, set precipitation weights and visibility weights, and use the precipitation weights and visibility weights as inputs to the LSTM network model, and output the corresponding weather correction coefficients; S2.1.1.2.3: Setting the road coefficient: Obtain road slope data using a laser rangefinder, and set the road correction coefficient based on the road slope data; S2.1.1.2.4: Determine the total factor coefficient: Combine the vehicle type correction coefficient, weather correction coefficient, and road correction coefficient to determine the total factor correction coefficient.

5. The adaptive control method for highway traffic flow state using edge node sensing according to claim 4, characterized in that, The vehicle model weights are set according to the vehicle model, specifically: the weight for small cars is set to 1, the weight for medium-sized cars is set to 1.3, and the weight for large cars is set to 2. The precipitation weight is set according to the precipitation intensity, specifically: the weight is set to 0.9 when the precipitation intensity is 0-2.5 mm / h, the weight is set to 0.85 when the precipitation intensity is 2.5-10 mm / h, and the weight is set to 0.8 when the precipitation intensity is greater than 10 mm / h. The visibility weights are set based on the visibility data, specifically: the weight is set to 0.8 when the visibility range is 200-500m, the weight is set to 0.75 when the visibility range is 100-200m, and the weight is set to 0.7 when the visibility range is less than 100m.

6. The adaptive control method for highway traffic flow state using edge node sensing according to claim 1, characterized in that, Based on the current number of lanes and headway at the intersection, a basic traffic capacity is set, and the basic traffic capacity is combined with the set traffic weight to determine the maximum number of vehicles that can pass. Based on the actual influencing factors at the current intersection, the traffic weights are set. These actual influencing factors include weather, vehicle type ratio, and road construction. The traffic weight corresponding to weather is set to 0.85, the traffic weight corresponding to vehicle type ratio is set to 0.92, and the traffic weight corresponding to road construction is set to 0.

7.

7. The adaptive control method for highway traffic flow state using edge node sensing according to claim 1, characterized in that, The minimum green light time in the signal timing physical constraints is set based on pedestrian crossing time and safety buffer time. At the same time, the maximum green light time in the signal timing physical constraints is set based on the minimum green light time, the common signal cycle, and the phase switching loss time.

Images