A Dynamic Traffic Signal Control Method, Device and Medium Based on Data Fusion

Abstract

The embodiment of the present application discloses a dynamic traffic signal control method, device and medium based on data fusion, belonging to the field of intelligent transportation, and solves the problem that existing traffic signal control is difficult to make an effective response when encountering sudden accidents. At the edge node, the first traffic data obtained by the road surveillance device and the second traffic data obtained by the positioning device of the floating vehicle are fused through the multi-head attention mechanism to obtain multi-source traffic feature data; after every first preset time interval, the cloud inputs the multi-source traffic feature data into the preset signal optimization model; the preset signal optimization model is updated and optimized through the PPO algorithm and the federated averaging method to generate a global traffic model; through knowledge distillation, the global traffic model is compressed, and the compressed edge traffic model is migrated to the corresponding edge node to output traffic signal strategies for the corresponding road section through the edge traffic model.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technologies, and in particular, to a dynamic traffic signal control method, device, and medium based on data fusion. Background Art

[0002] With the acceleration of the urbanization process, the problem of traffic congestion has become increasingly serious, bringing many challenges to people's daily travel, logistics transportation, and the sustainable development of cities. The core problems of traffic congestion include three points: rigid signal timing, that is, traditional timing strategies are difficult to adapt to dynamic traffic flows; data islands, that is, multi-source data such as cameras and GPS are not fully integrated; response delay, that is, the update speed of control strategies is slow.

[0003] Traditional signal light optimization methods adopt timing control strategies. According to historical traffic flow statistical data, a fixed-duration traffic light cycle is preset in advance. The arrival of vehicles is detected through inductive loops or radars, and the green light duration is dynamically adjusted. Or through on-site command by traffic police or manual modification of signal machine parameters, sudden congestion is temporarily dealt with.

[0004] However, traditional signal light optimization methods rely on a single sensor, with a limited data coverage range, and it is difficult to obtain multi-dimensional information such as real-time traffic flow density and vehicle type distribution across the entire road network. The data acquisition ability is weak. At some intersections, due to data missing, the signal timing does not match the traffic flow, which instead exacerbates congestion. In addition, due to the rigid control strategy, it is difficult to adapt to the spatio-temporal heterogeneity of traffic flows in case of sudden accidents, weather changes, etc., so that traffic signal control is difficult to make an effective response. Summary of the Invention

[0005] Embodiments of this application provide a dynamic traffic signal control method, device, and medium based on data fusion to solve the following technical problems: For traditional signal light optimization methods, the data coverage range is limited, the control strategy is rigid, and in case of sudden accidents, traffic signal control is difficult to make an effective response.

[0006] Embodiments of this application adopt the following technical solutions:

[0007] An embodiment of the present application provides a dynamic traffic signal control method based on data fusion. It includes, at the edge node, using the multi-head attention mechanism to assign weights to the first traffic data obtained by the road surveillance device and the second traffic data obtained by the positioning device of the floating vehicle, and performing data fusion based on the assigned weights to obtain multi-source traffic feature data; after every first preset time period, the cloud receives the multi-source traffic feature data uploaded by the edge node and inputs the multi-source traffic feature data into a preset signal optimization model; wherein, the preset signal optimization model is a multi-modal Transformer architecture based on DeepSeek-R1 and is constructed by integrating meta-learning; through the PPO algorithm and the federated averaging method, the preset signal optimization model is updated and optimized to generate a global traffic model; through knowledge distillation, the global traffic model is compressed, and the obtained edge traffic model after compression is migrated to the corresponding edge node, so as to output traffic signal strategies for the corresponding road section through the edge traffic model.

[0008] In the embodiment of the present application, the first traffic data of the road surveillance device and the second traffic data of the floating vehicle positioning device are fused through the multi-head attention mechanism, which can comprehensively obtain multi-dimensional information such as the real-time traffic flow density and vehicle type distribution of the entire road network, making up for the deficiency of the data coverage of traditional single sensors. Secondly, the embodiment of the present application uses the PPO algorithm and the federated averaging method to update and optimize the model, improve traffic fluency, integrate the data of multiple edge nodes, make the model training more comprehensive and accurate, avoid the influence of the data deviation of a single node, and the generated global traffic model is more universal and reliable. By compressing the global traffic model through knowledge distillation to obtain the edge traffic model and migrating it to the corresponding edge node, without losing too much performance, the model size is reduced, the computing and storage burden of the edge node is reduced, the model can run efficiently on the edge device, and the edge node can use local data and the edge traffic model to output traffic signal strategies in real time, reduce data transmission delay, quickly respond to local traffic changes, and improve the timeliness and accuracy of traffic control.

[0009] In one implementation of the present application, at the edge node, the first traffic data obtained by the road monitoring device and the second traffic data obtained by the positioning device of the floating vehicle are weighted by a multi-head attention mechanism, and data fusion is performed based on the allocated weights to obtain multi-source traffic feature data. Specifically, at the edge node, first feature extraction is respectively performed on the first traffic data and the second traffic data based on spatial features; and second feature extraction is respectively performed on the first traffic data and the second traffic data based on temporal features; the first feature and the second feature corresponding to the first traffic data, and the first feature and the second feature corresponding to the second traffic data are dynamically weighted by a multi-head attention mechanism; data fusion is performed based on the dynamic weights, the first feature and the second feature corresponding to the first traffic data, and the first feature and the second feature corresponding to the second traffic data to obtain multi-source traffic feature data.

[0010] In one implementation of the present application, before inputting the multi-source traffic feature data into the preset signal optimization model, the method further includes: taking minimizing the average intersection delay time as the core objective, and taking reducing the number of hard brakes and giving priority to bus vehicles passing through as auxiliary objectives, and constructing a reward function:

[0011] ;

[0012] where R is the reward function; w 1 is the first weight coefficient; w 2 is the second weight coefficient; w 3 is the third weight coefficient, ; D is the average intersection delay time; B is the number of hard brakes; P is the bus vehicle passing priority score; based on the traffic scenario corresponding to the edge node, the weight of the reward function is adjusted to optimize the preset signal optimization model based on the adjusted reward function.

[0013] In one implementation of the present application, the preset signal optimization model is updated and optimized by the PPO algorithm and the federated averaging method to generate a global traffic model. Specifically, based on the PPO algorithm, through an optimized policy, the long-term cumulative reward is maximized; where the objective function of the PPO algorithm is:

[0014] ;

[0015] where represents the probability ratio of the new and old policies; is the newly optimized policy; is the old optimized policy; represents the signal light control policy; represents the advantage function; represents the clipping threshold; clip is the clipping function; is the model parameter; for the agent at time step t according to the state s t executed action; for the agent at time step t state; Using the federated averaging algorithm, aggregate the model parameters corresponding to multiple edge nodes respectively to generate a global traffic model.

[0016] In an implementation manner of the present application, using the federated averaging algorithm, aggregate the model parameters corresponding to multiple edge nodes respectively to generate a global traffic model, which specifically includes: at each edge node, fine-tune the preset signal optimization model through local data to obtain fine-tuning parameters; upload the obtained fine-tuning parameters and the local data corresponding to each edge node to the cloud; at the cloud, perform weighted averaging on the data uploaded by multiple edge nodes to generate a global traffic model.

[0017] In an implementation manner of the present application, performing weighted averaging on the data uploaded by multiple edge nodes specifically includes: based on the function:

[0018] ;

[0019] ;

[0020] Based on the data volume corresponding to each edge node respectively, perform weighted averaging on the data uploaded by multiple edge nodes; where is the global traffic model; represents the number of edge nodes participating in federated learning; represents the node local quantity; N is the total data volume; is the edge node k corresponding model parameter.

[0021] In an implementation manner of the present application, through knowledge distillation, compress the global traffic model and migrate the compressed edge traffic model to the corresponding edge node, which specifically includes: based on the function:

[0022] ;

[0023] ;

[0024] Migrate the compressed edge traffic model to the corresponding edge node; where L is the loss function; is the cross - entropy loss, aligning the predicted data of the edge traffic model with the true label ; is the divergence loss, aligning the output probability of the edge traffic model with the global traffic model probability ; is the weight coefficient; is the logits output of the global traffic model; is the logits output of the edge traffic model; represents the temperature parameter.

[0025] In one implementation manner of the present application, the edge traffic model outputs traffic signal strategies for corresponding road sections, specifically including: each edge node inputs the obtained traffic information into the corresponding edge traffic model every second preset time period to output the traffic signal strategy corresponding to the current road section through the edge traffic model; and, when the average vehicle speed of the current road section is less than the preset vehicle speed threshold, or the vehicle queue length of the current road section is greater than the preset lane capacity threshold, a congestion response mechanism is triggered; the cloud determines the target vehicle speed and phase difference of the main road based on the global vehicle flow data, and delays the signal lights at the downstream intersections when the growth rate of the vehicle flow quantity at the upstream intersections of the main road is greater than the preset growth rate threshold; and, the cloud determines the vehicles driving towards the main road and sends recommended filtered vehicle speed information to the vehicles driving towards the main road through V2X communication.

[0026] An embodiment of the present application provides a dynamic traffic signal control device based on data fusion, including: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: at the edge node, perform weight allocation on the first traffic data obtained by the road surveillance device and the second traffic data obtained by the positioning device of the floating vehicle through the multi - head attention mechanism, and perform data fusion based on the allocated weights to obtain multi - source traffic feature data; every first preset time period, the cloud receives the multi - source traffic feature data uploaded by the edge node and inputs the multi - source traffic feature data into a preset signal optimization model; wherein, the preset signal optimization model is a multi - modal Transformer architecture based on DeepSeek - R1 and is constructed by integrating meta - learning; update and optimize the preset signal optimization model through the PPO algorithm and the federated averaging method to generate a global traffic model; compress the global traffic model through knowledge distillation, and migrate the compressed edge traffic model to the corresponding edge node to output traffic signal strategies for the corresponding road sections through the edge traffic model.

[0027] A non-volatile computer storage medium provided by an embodiment of the present application stores computer-executable instructions, and the computer-executable instructions are set to: at an edge node, perform weight allocation on first traffic data obtained by a road surveillance device and second traffic data obtained by a positioning device of a floating vehicle through a multi-head attention mechanism, and perform data fusion based on the allocated weights to obtain multi-source traffic feature data; after every first preset time period, the cloud receives the multi-source traffic feature data uploaded by the edge node and inputs the multi-source traffic feature data into a preset signal optimization model; wherein, the preset signal optimization model is a multi-modal Transformer architecture based on DeepSeek-R1 and is constructed by integrating meta-learning; update and optimize the preset signal optimization model through the PPO algorithm and the federated averaging method to generate a global traffic model; compress the global traffic model through knowledge distillation, and migrate the obtained edge traffic model after compression to the corresponding edge node, so as to output a traffic signal strategy for the corresponding road section through the edge traffic model.

[0028] The above at least one technical solution adopted by the embodiment of the present application can achieve the following beneficial effects: By integrating the first traffic data of the road surveillance device and the second traffic data of the floating vehicle positioning device through the multi-head attention mechanism, the embodiment of the present application can comprehensively obtain multi-dimensional information such as real-time traffic flow density and vehicle type distribution of the entire road network, making up for the defect of insufficient data coverage of traditional single sensors. Secondly, the embodiment of the present application uses the PPO algorithm and the federated averaging method to update and optimize the model, improve traffic fluency, integrate data of multiple edge nodes, make the model training more comprehensive and accurate, avoid the influence of single-node data deviation, and make the generated global traffic model more general and reliable. Compress the global traffic model through knowledge distillation to obtain an edge traffic model and migrate it to the corresponding edge node. Without losing too much performance, the model scale is reduced, the computing and storage burden of the edge node is reduced, the model can run efficiently on the edge device, and the edge node can use local data and the edge traffic model to output traffic signal strategies in real time, reduce data transmission delay, quickly respond to local traffic changes, and improve the timeliness and accuracy of traffic control. Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments recorded in the present application. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts. In the drawings:

[0030] Figure 1 It is a flowchart of a dynamic traffic signal control method based on data fusion provided by an embodiment of the present application;

[0031] Figure 2 Schematic structural diagram of the algorithm design principle of a DRL signal optimization model provided by an embodiment of the present application;

[0032] Figure 3 Schematic structural diagram of a dynamic regulation strategy mechanism provided by an embodiment of the present application;

[0033] Figure 4 Practical application sketch of a dynamic traffic signal optimization system under a vehicle-road-cloud collaboration framework provided by an embodiment of the present application;

[0034] Figure 5 Schematic structural diagram of a dynamic traffic signal control device based on data fusion provided by an embodiment of the present application.

[0035] Reference numerals:

[0036] 200: Dynamic traffic signal control device based on data fusion, 201: Processor, 202: Memory. Detailed implementation manners

[0037] An embodiment of the present application provides a dynamic traffic signal control method, device and medium based on data fusion.

[0038] In order to enable those skilled in the art to better understand the technical solutions in the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of this specification without creative efforts shall fall within the protection scope of the present application.

[0039] The technical solutions proposed in the embodiments of the present invention will be described in detail below with reference to the drawings.

[0040] Figure 1 Flowchart of a dynamic traffic signal control method based on data fusion provided by an embodiment of the present application, as Figure 1 shown, the process of the dynamic traffic signal control method based on data fusion includes the following steps:

[0041] Step 101: At the edge node, perform weight assignment on the first traffic data obtained by the road monitoring device and the second traffic data obtained by the positioning device of the floating vehicle through the multi-head attention mechanism, and perform data fusion based on the assigned weights to obtain multi-source traffic feature data.

[0042] In an implementation of the present application, at the edge node, first feature extraction is respectively performed on the first traffic data and the second traffic data based on spatial features. And, second feature extraction is respectively performed on the first traffic data and the second traffic data based on temporal features. The first features and second features corresponding to the first traffic data, and the first features and second features corresponding to the second traffic data are dynamically weighted by means of a multi-head attention mechanism. Data fusion is performed based on the dynamic weights, the first features and second features corresponding to the first traffic data, and the first features and second features corresponding to the second traffic data to obtain multi-source traffic feature data.

[0043] Specifically, first, multi-source data is obtained, including first traffic data, i.e., camera data, and second traffic data, i.e., GPS data. Specifically: The camera data is the vehicle queue length and vehicle type distribution, and the vehicle types include private cars, buses, bicycles, pedestrians, etc. The GPS data is the vehicle trajectory, speed, and location, including floating vehicles such as taxis, online car-hailing vehicles, and buses.

[0044] Secondly, the data is preprocessed and feature extraction is performed. On the one hand, during data preprocessing, spatio-temporal alignment and outlier filtering are required. Spatially, the camera and GPS data are unified into the same coordinate system, such as WGS84. Temporally, the data timestamps are ensured to be synchronized, such as sampling once every 5 seconds. When performing outlier filtering, stationary vehicles are excluded, such as GPS points in the parking lot, and abnormal traffic data caused by camera occlusion or failure is identified and excluded. On the other hand, feature extraction mainly divides data features in terms of time and space.

[0045] For spatial features, the camera data is the lane-level vehicle density, queue length, pedestrian, and bicycle flow, and the GPS data is the vehicle trajectory and speed distribution. In terms of temporal features, the camera data is the traffic flow change trend, such as the queue length growth rate, and the GPS data is the vehicle arrival rate and driving intention, such as turning left and going straight.

[0046] Further, attention weight calculation. The basis for weight allocation is data reliability, coverage, and real-time performance. First, the weight of the camera is high when the light is sufficient, and the weight of GPS is high in the tunnel. Second, the camera covers the intersection area, and GPS covers the entire road network. Third, the GPS data update frequency is high (once per second), and the camera update frequency is low (once every 5 seconds). The specific attention mechanism is designed to calculate the weight using multi-head attention (MHA):

[0047] ;

[0048] Among them, Q represents the query vector, that is, the current traffic state; K represents the key vector, that is, the camera and GPS features; VRepresents a value vector, i.e., camera and GPS data; Represents the feature dimension. T Is the transpose symbol of matrix K. In the dynamic adjustment of weights, when the camera detects a pedestrian or a bicycle, the weight of camera data is increased, and when the GPS data has a wide coverage area, the weight of GPS data is increased.

[0049] Furthermore, fuse camera data and GPS data. Specifically, by performing weighted summation on the features of camera and GPS data:

[0050] ;

[0051] Wherein, Represents the camera data feature, Represents the GPS data feature, Both represent dynamic weights. When a certain data source is missing, such as when the camera fails, historical data or another data source is used for interpolation.

[0052] Furthermore, use the Kalman filter algorithm to smooth the logarithm, eliminate noise, and output the result. The data result is to generate lane-level traffic states, i.e., traffic flow, queue length, vehicle speed, etc., for traffic signal timing optimization.

[0053] Edge collaborative computing plays a core role in real-time data processing and low-latency decision-making in intelligent transportation systems. By sinking computing power to edge nodes close to the data source, such as intersection signal controllers, the cloud transmission delay can be significantly reduced, and the system response speed can be improved.

[0054] Specifically, build an edge-cloud collaborative architecture through corresponding hardware devices, and upload the preprocessed data to the cloud for model training and updating.

[0055] Step 102: After every first preset time interval, the cloud receives the multi-source traffic feature data uploaded by the edge node, and inputs the multi-source traffic feature data into the preset signal optimization model.

[0056] In one implementation manner of the present application, based on the multi-source data fusion and edge computing collaboration, the embodiments of the present application construct a DRL signal optimization model through a multi-modal Transformer architecture and meta-learning to achieve dynamic optimization of traffic signals. Figure 2 Is a schematic structural diagram of the algorithm design principle of a DRL signal optimization model provided by the embodiments of the present application. The algorithm design principle of the model is as Figure 2 Shown. The core objective of this step is to output a control strategy through the algorithm of the DRL signal optimization model, dynamically optimize the traffic signal timing, minimize the average delay time at intersections, and simultaneously take into account auxiliary objectives such as energy consumption and bus priority.

[0057] In an implementation manner of this application, first, the fused multi-source feature data is input into the DRL model:

[0058] Core objective: Minimize the average intersection delay time (seconds):

[0059] ;

[0060] Among them, represents the actual passing time of the vehicle, which is calculated by the edge device through data;

[0061] represents the theoretical passing time under free-flow vehicle speed, which is pre-calculated by the cloud and sent down;

[0062] Auxiliary objectives: Reduce the number of hard brakes (reduce energy consumption) and give priority to ensuring the passage of bus vehicles. Detect hard brake events through GPS data, with an acceleration threshold < -2m / s 2 as the criterion, and calculate the number of hard brakes per minute B . Giving priority to ensuring the passage of bus vehicles means that when the bus is less than 200 meters away from the intersection, a priority signal is triggered P = 1, otherwise P = 0;

[0063] Construct the reward function:

[0064] ;

[0065] Among them, R is the reward function; w 1 is the first weight coefficient; w 2 is the second weight coefficient; w 3 is the third weight coefficient, ; D is the average intersection delay time; B is the number of hard brakes; P is the bus vehicle passage priority score;

[0066] Peak hours (congestion index > 7): ;

[0067] Off-peak hours: ;

[0068] Based on the traffic scenario corresponding to the edge node, adjust the weights of the reward function, and optimize the preset signal optimization model based on the adjusted reward function.

[0069] Furthermore, the multi-modal Transformer architecture based on DeepSeek-R1 in the embodiments of this application includes:

[0070] Input layer: Receives multi-modal feature data, i.e., lane-level traffic flow and historical delay time;

[0071] Transformer encoder: Uses the multi-head attention mechanism to capture spatio-temporal propagation patterns;

[0072] Output layer: Generates a signal timing plan, i.e., green light duration and phase switching sequence.

[0073] Meta-Learning adapts to different urban road networks:

[0074] Objective: To enable the model to quickly adapt to the characteristics of different urban road networks;

[0075] Task division: Divides the road network data of different cities into multiple tasks;

[0076] Meta-training: Trains the model on multiple tasks to generate initial parameters;

[0077] Meta-testing: Fine-tunes the model parameters by sending historical data to the cloud on a new urban road network to enable the model to quickly adapt to the local road network;

[0078] Continuous update: Dynamically updates the model according to real-time data by the edge device.

[0079] Step 103: Update and optimize the pre-set signal optimization model through the PPO algorithm and the federated averaging method to generate a global traffic model.

[0080] In one implementation of the present application, cloud centralized training: PPO algorithm.

[0081] Objective: By optimizing the policy , the probability distribution of signal control actions, to maximize the long-term cumulative reward.

[0082] Core formula: The objective function of the PPO algorithm includes policy gradient and importance sampling, and uses clipping to avoid excessive policy updates:

[0083] ;

[0084] Where represents the probability ratio of the new and old policies; is the newly optimized policy; is the old optimized policy; represents the signal control policy; represents the advantage function; represents the clipping threshold, usually set to 0.2; clip is the clipping function; is the model parameter; is the agent at time step t According to the state st Actions performed; For the agent at time step t state.

[0085] Furthermore, The advantage function is calculated using Generalized Advantage Estimation (GAE):

[0086] ; ;

[0087] where, is the discount factor 0.99, is the GAE smoothing coefficient 0.95, represents the state value estimated by the value network; is the time step t TD residual; represents the probability ratio of the old and new policies.

[0088] In one implementation of the present application, by maximizing the gradient ascent of , the policy parameters are updated.

[0089] In one implementation of the present application, at each edge node, the pre-set signal optimization model is fine-tuned with local data to obtain fine-tuning parameters. The obtained fine-tuning parameters and the local data corresponding to each edge node are uploaded to the cloud. At the cloud, the data uploaded by multiple edge nodes is weighted and averaged to generate a global traffic model.

[0090] Specifically, based on the function:

[0091] ;

[0092] ;

[0093] Based on the data volume corresponding to each edge node, the data uploaded by multiple edge nodes is weighted and averaged;

[0094] where, is the global traffic model; represents the number of edge nodes participating in federated learning; represents the node local quantity; N is the total amount of data; is the edge node k corresponding model parameters.

[0095] That is, edge - distributed fine - tuning is achieved through federated averaging. The goal is to aggregate the local model parameters of multiple edge nodes to generate a global model. Each edge node fine - tunes the model using local data to obtain parameters . Each node uploads and to the cloud, where they are averaged with weights according to the data volume to generate a new global model.

[0096] Step 104: Compress the global traffic model through knowledge distillation and migrate the resulting edge traffic model to the corresponding edge node to output traffic signal strategies for the corresponding road segments through the edge traffic model.

[0097] In an implementation manner of this application, model lightweighting is achieved through knowledge distillation.

[0098] The goal is to transfer the knowledge of the large cloud model to the small edge model.

[0099] Based on the function:

[0100] ;

[0101] ;

[0102] Migrate the resulting edge traffic model to the corresponding edge node;

[0103] Among them, L is the loss function; is the cross - entropy loss, aligning the predicted data of the edge traffic model with the true label ; is the divergence loss, aligning the output probability of the edge traffic model with the probability of the global traffic model; is the weight coefficient, usually taken as 0.1; is the logits output of the global traffic model; is the logits output of the edge traffic model; represents the temperature parameter.

[0104] Furthermore, model output softening:

[0105] ;

[0106] Among them, and are the logits outputs of the global traffic model and the edge traffic model respectively, represents the temperature parameter, usually taken as 3, softening the probability distribution to transfer more implicit knowledge.

[0107] Specifically, by compressing into a lightweight edge small model to adapt to edge devices.

[0108] In an implementation manner of the present application, each edge node inputs the obtained traffic information into the corresponding edge traffic model every second preset time period, so as to output the traffic signal strategy corresponding to the current road section through the edge traffic model. And, in the case where the average vehicle speed of the current road section is less than the preset vehicle speed threshold, or the vehicle queue length of the current road section is greater than the preset lane capacity threshold, a congestion response mechanism is triggered. The cloud determines the target vehicle speed and phase difference of the main road based on the global vehicle flow data, and delays the signal lights of the downstream intersection in the case where the growth rate of the vehicle flow quantity at the upstream intersection of the main road is greater than the preset growth rate threshold. And, the cloud determines the vehicles driving towards the main road, and sends recommended filtered vehicle speed information to the vehicles driving towards the main road through V2X communication.

[0109] Specifically, the regulation strategy is mainly divided into two aspects: hierarchical response and green wave coordination. Among them, the hierarchical response mechanism is divided into normal optimization and sudden congestion response. The edge device integrates data such as lane-level vehicle flow and queue length every 5 minutes, and generates a timing plan through a lightweight DRL model. When the real-time detected average vehicle speed of the lane < 15 km / h, or the queue length exceeds 80% of the lane capacity, a congestion response mechanism is triggered, and the current green light phase is extended by seconds or an emergency phase is inserted to forcibly evacuate the congested vehicle flow.

[0110] Furthermore, the cloud platform calculates the ideal vehicle speed and phase difference of the main road based on the global vehicle flow data. When the vehicle flow at the upstream intersection surges, the green light duration of the downstream intersection is extended to avoid vehicle flow accumulation. Recommend the green wave vehicle speed to the approaching vehicles through V2X communication, such as prompting the driver through the navigation APP that "driving at X km / h can continuously pass through Y green lights".

[0111] Specifically, the edge device obtains the vehicle speed, driving time, and license plate information corresponding to the vehicle, generates a vehicle speed driving sequence based on the vehicle speeds corresponding to different moments, and uploads the vehicle speed driving sequences corresponding to different vehicles to the cloud.

[0112] The cloud analyzes the obtained vehicle speed driving sequences, inputs the vehicle speed driving sequences into a preset vehicle speed prediction model to output the predicted vehicle speeds of each vehicle in the future time period, and determines the time when each vehicle reaches the stop line of the main road according to the position and predicted vehicle speed of each vehicle. Among them, the preset vehicle speed prediction model can be a neural network model, and the training process of this model is to use the historical vehicle speed driving sequence samples as inputs, use the historical predicted vehicle speed data corresponding to the input samples as output samples, and train the preset neural network model to obtain the preset vehicle speed prediction model.

[0113] Divide the main road into multiple sub-roads according to the number of intersections of the main road, and determine the predicted traffic flow corresponding to each sub-road in the future time period according to the time when each vehicle arrives at the stop line of the main road.

[0114] Furthermore, when the predicted traffic flow of the sub-road is greater than the preset traffic flow threshold, the cloud sends a green light delay instruction to the edge device related to the sub-road.

[0115] In addition, match the driving route of the vehicle with each sub-road. After the vehicle driving route matches the sub-road successfully, determine the positions of all traffic lights on the sub-road according to the geographical information of the sub-road. Form a set of these traffic lights to determine the set of traffic lights that the vehicle needs to pass through.

[0116] Determine the recommended green wave speed corresponding to the vehicle according to the regulation information of each traffic light in the traffic light set and the predicted vehicle speed of the vehicle in the future time period. Among them, the regulation information is information such as the cycle duration, green light start time, and green light duration of the traffic light. Specifically, calculate the time required for the vehicle to travel from the current position to the traffic light, which can be obtained by dividing the distance between the vehicle and the traffic light by the predicted vehicle speed of the vehicle. According to the cycle duration, green light start time, and green light duration of the traffic light, determine the state of the traffic light when the vehicle arrives. If the traffic light is in the green light state when the vehicle arrives, record the remaining green light time; if it is in the red light state, calculate the time when the next green light lights up. Within the safe driving speed range of the vehicle, traverse different vehicle speeds, and calculate the total time when the vehicle encounters a green light or the number of traffic lights passed by the green light when passing through the traffic light set at each vehicle speed. Select the vehicle speed that makes the number of traffic lights passed by the green light the most or the total green light time the longest as the recommended green wave speed.

[0117] It should be noted that when determining the recommended green wave speed, some actual situations need to be considered, such as road speed limits, traffic flow, vehicle performance, etc. If the recommended green wave speed exceeds the road speed limit or the actual drivable speed range of the vehicle, the recommended vehicle speed needs to be adjusted to ensure the safety and feasibility of vehicle driving.

[0118] Figure 3 It is a schematic structural diagram of a dynamic regulation strategy mechanism provided by an embodiment of the present application. As Figure 3 shown, first collect data such as the delay and number of hard brakes corresponding to the lane, upload the obtained data to the cloud to update the reward function weight, and send the optimized model to the edge device to execute signal control through the edge device.

[0119] Figure 4 It is a practical application diagram of a dynamic traffic signal optimization system under a vehicle-road-cloud collaboration framework provided by an embodiment of the present application. As Figure 4As shown in the figure, multiple edge devices respectively upload the acquired lane data to the cloud. The cloud optimizes the model through the PPO algorithm and the federated averaging algorithm, and then distributes the optimized model to the edge devices.

[0120] Furthermore, the embodiment of the present application also has a system anti-interference and robustness design.

[0121] Specifically, when data is missing, the multi-modal attention mechanism dynamically adjusts the weights to ensure the reliability of the fusion result. In addition, an adaptive pruning method is used to remove neurons with low contribution, reduce the computational load, realize model lightweight, and improve the anti-interference ability and robustness.

[0122] For example, to verify the technical effect of the invention, based on the actual road network parameters of an intersection in a certain city, a comparative experiment is carried out by simulating the morning peak traffic flow through the SUMO 1.14.1 simulation platform. Table 1 is a simulation result table of intersection signal timing provided by the embodiment of the present application. As shown in the experimental group parameters in Table 1, road network topology: 4-phase signal lights, 3 main roads. Morning peak traffic flow from 7:00 to 9:00: 1200 vehicles / hour, penetration rate 10%; vehicle types: private cars 70%, buses 10%, trucks 20%; DRL model output dynamic signal timing delay < 200ms. Comparative group parameters, traditional fixed-time control with a fixed cycle of 120 seconds, phases: straight 40s / left turn 30s / right turn 20s; induction control of the ground loop triggers a maximum green light extension of 15s. The simulation duration is 2 hours for the morning peak, including a warm-up time of 30 minutes and an effective data collection time of 1.5 hours. During the training and optimization process of the model, the PPO algorithm trains the global DRL model in the cloud, and federated learning aggregates 7 days of local data to fine-tune the parameters. The lightweight model is compressed to 9.8 million, and the edge inference delay is 45ms. As shown in Table 1, normal optimization: adjust the signal timing plan every 5 minutes, and the green wave coordinated control covers 3 upstream and downstream intersections. Emergency response: when the vehicle speed < 15km / h is detected, the green light is extended by 15 seconds in seconds or a bus priority phase is inserted.

[0123] Table 1

[0124]

[0125] Figure 5 It is a schematic structural diagram of a dynamic traffic signal control device based on data fusion provided by the embodiment of the present application. As Figure 5As shown in the figure, the dynamic traffic signal control device 200 based on data fusion includes: at least one processor 201; and a memory 202 communicatively connected to the at least one processor 201; wherein, the memory 202 stores instructions executable by the at least one processor 201, and when the instructions are executed by the at least one processor 201, the at least one processor 201 is enabled to: at the edge node, perform weight allocation on the first traffic data obtained by the road monitoring device and the second traffic data obtained by the positioning device of the floating vehicle through the multi-head attention mechanism, and perform data fusion based on the allocated weights to obtain multi-source traffic feature data; after every first preset time interval, the cloud receives the multi-source traffic feature data uploaded by the edge node and inputs the multi-source traffic feature data into a preset signal optimization model; wherein, the preset signal optimization model is a multi-modal Transformer architecture based on DeepSeek-R1 and is constructed by integrating meta-learning; update and optimize the preset signal optimization model through the PPO algorithm and the federated averaging method to generate a global traffic model; compress the global traffic model through knowledge distillation, and migrate the compressed edge traffic model to the corresponding edge node to output traffic signal strategies for the corresponding road section through the edge traffic model.

[0126] A non-volatile computer storage medium provided by an embodiment of the present application stores computer-executable instructions, and the computer-executable instructions are set to: at the edge node, perform weight allocation on the first traffic data obtained by the road monitoring device and the second traffic data obtained by the positioning device of the floating vehicle through the multi-head attention mechanism, and perform data fusion based on the allocated weights to obtain multi-source traffic feature data; after every first preset time interval, the cloud receives the multi-source traffic feature data uploaded by the edge node and inputs the multi-source traffic feature data into a preset signal optimization model; wherein, the preset signal optimization model is a multi-modal Transformer architecture based on DeepSeek-R1 and is constructed by integrating meta-learning; update and optimize the preset signal optimization model through the PPO algorithm and the federated averaging method to generate a global traffic model; compress the global traffic model through knowledge distillation, and migrate the compressed edge traffic model to the corresponding edge node to output traffic signal strategies for the corresponding road section through the edge traffic model.

[0127] The various embodiments in the present application are all described in a progressive manner. The same or similar parts among the various embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0128] The above are only the embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various modifications and changes can be made to the embodiments of the present application. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A dynamic traffic signal control method based on data fusion, characterized in that: The method comprises: At the edge node, the first traffic data obtained by the road monitoring device and the second traffic data obtained by the positioning device of the floating vehicle are weighted by a multi-head attention mechanism, and data fusion is performed based on the assigned weights to obtain multi-source traffic feature data; After each first preset time interval, the cloud receives the multi-source traffic feature data uploaded by the edge node, and inputs the multi-source traffic feature data into a preset signal optimization model; wherein the preset signal optimization model is a multimodal Transformer architecture based on DeepSeek-R1 and is constructed by integrating meta-learning; The preset signal optimization model is updated and optimized by using the PPO algorithm and the federated average method to generate a global traffic model; By means of knowledge distillation, the global traffic model is compressed, and the edge traffic model obtained after compression is migrated to the corresponding edge node, so as to output the traffic signal strategy for the corresponding road section through the edge traffic model; The preset signal optimization model is updated and optimized by the PPO algorithm and the federated average method to generate a global traffic model, specifically including: Based on the PPO algorithm, maximize the long-term cumulative rewards by optimizing the strategy; Among them, the objective function of the PPO algorithm is: ； in, represents the probability ratio of the new and old strategies; Optimize strategies for the new; Optimize strategies for legacy; Indicates the signal light control strategy; represents the advantage function; Indicates the clipping threshold; clip is the clipping function; are model parameters; For the agent at time step t According to the status s t the action performed; For the agent at time step t Status; The model parameters corresponding to the plurality of edge nodes are aggregated by using a federated averaging algorithm to generate the global traffic model.

2. A dynamic traffic signal control method based on data fusion according to claim 1, characterized in that: At the edge node, the first traffic data obtained by the road monitoring device and the second traffic data obtained by the positioning device of the floating vehicle are weighted by a multi-head attention mechanism, and data fusion is performed based on the weights assigned to obtain multi-source traffic feature data, specifically including: At the edge node, extracting first features from the first traffic data and the second traffic data based on spatial features; and, based on the time feature, extracting second features from the first traffic data and the second traffic data respectively; Dynamically weighting the first feature and the second feature corresponding to the first traffic data, and the first feature and the second feature corresponding to the second traffic data, by using a multi-head attention mechanism; Based on the dynamic weight, the first feature and the second feature corresponding to the first traffic data, and the first feature and the second feature corresponding to the second traffic data, data fusion is performed to obtain the multi-source traffic feature data.

3. The method for dynamic traffic signal control based on data fusion according to claim 1, characterized in that: Before inputting the multi-source traffic characteristic data into a preset signal optimization model, the method further comprises: Taking minimizing the average delay time at the intersection as the core goal, and reducing the number of emergency brakes and giving priority to public transportation vehicles as auxiliary goals, we construct a reward function: ； in, R is the reward function; w 1 is the first weight coefficient; w 2 is the second weight coefficient; w 3 is the third weight coefficient, ; D is the average delay time at the intersection; B is the number of emergency brakes; P Score the transit priority for public transport vehicles; Based on the traffic scenario corresponding to the edge node, the reward function is weighted to optimize the preset signal optimization model based on the adjusted reward function.

4. The method for dynamic traffic signal control based on data fusion according to claim 1, characterized in that: The method of using a federated average algorithm to aggregate the model parameters corresponding to the plurality of edge nodes to generate the global traffic model specifically includes: At each of the edge nodes, fine-tuning the preset signal optimization model using local data to obtain fine-tuning parameters; Uploading the obtained fine-tuning parameters and local data corresponding to each edge node to the cloud; At the cloud end, a weighted average is performed on the data uploaded by the plurality of edge nodes to generate the global traffic model.

5. A dynamic traffic signal control method based on data fusion according to claim 4, characterized in that: The weighted average of the data uploaded by the plurality of edge nodes specifically includes: Through the function: ； ； Based on the data volume corresponding to each edge node, weighted average is performed on the data uploaded by the plurality of edge nodes; in, is the global traffic model; Indicates the number of edge nodes participating in federated learning; Representation Node The number of local N is the total amount of data; For edge nodes k The corresponding model parameters.

6. The method for dynamic traffic signal control based on data fusion according to claim 1, characterized in that: The global traffic model is compressed through knowledge distillation, and the edge traffic model obtained after compression is migrated to the corresponding edge node, specifically including: Function-based: ； ； Migrate the compressed edge traffic model to the corresponding edge node; in, L is the loss function; For the cross entropy loss, align the marginal traffic model prediction data With the true label ; is the divergence loss, aligning the output probability of the edge traffic model and the global traffic model probability ; is the weight coefficient; Logits output of the global traffic model; Logits output for the edge traffic model; Represents the temperature parameter.

7. The method for dynamic traffic signal control based on data fusion according to claim 1, characterized in that: The outputting of the traffic signal strategy for the corresponding road section by using the edge traffic model specifically includes: Each of the edge nodes inputs the acquired traffic information into the corresponding edge traffic model every second preset time interval, so as to output a traffic signal strategy corresponding to the current road section through the edge traffic model; And, when the average vehicle speed of the current road section is less than a preset vehicle speed threshold, or the length of the vehicle queue of the current road section is greater than a preset lane capacity threshold, a congestion response mechanism is triggered; The cloud determines the target vehicle speed and phase difference of the main road based on the global traffic data, and when the growth rate of the traffic volume at the upstream intersection of the main road is greater than the preset growth rate threshold, the traffic lights at the downstream intersection are delayed; Furthermore, the cloud determines the vehicles heading toward the main road, and sends recommended filtered vehicle speed information to the vehicles heading toward the main road through V2X communication.

8. A dynamic traffic signal control device based on data fusion, characterized in that: The device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute the method according to any one of claims 1 to 7.

9. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions can execute the method according to any one of claims 1 to 7.

Images