Urban traffic signal space-time adaptive flexible regulation and control method and system

Abstract

The invention discloses an urban traffic signal space-time adaptive flexible regulation and control method and system. The method comprises the steps of collecting traffic data such as the number of vehicles, the vehicle speed, the number of pedestrians and the pedestrian speed at an intersection in real time; performing cleaning, normalization and feature extraction on the acquired traffic data to generate standardized traffic parameters; based on the standardized traffic parameters, a signal lamp timing scheme is generated through a space-time diagram convolutional network and lightweight deep reinforcement learning hybrid algorithm; adjusting the traffic signal lamp in real time according to the signal lamp timing scheme; the method comprises the steps of storing traffic data and remotely monitored intersection states, integrating information of adjacent intersections to perform regional strategy optimization, integrating strategy data of each intersection through federal learning, and generating a global model to optimize a collaborative penalty coefficient; the method can quickly respond to the dynamic change of the traffic flow, optimizes the signal lamp timing scheme, and remarkably improves the data processing speed and decision-making efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent traffic management, and particularly relates to an urban traffic signal spatio-temporal adaptive flexible control method and system. Background Art

[0002] In recent years, with the acceleration of the urbanization process and the continuous growth of the motor vehicle ownership, the urban traffic flow has shown an explosive upward trend. As an important infrastructure for urban traffic management, traffic signal lights play a key role in regulating vehicle traffic and improving road utilization. With the rapid development of Internet of Things, big data and artificial intelligence technologies, intelligent transportation systems (ITS) are gradually becoming a hot research and application direction in the industry, providing new possibilities for the intelligent upgrade of traffic management.

[0003] Currently, the traffic signal control systems in many cities still mainly rely on fixed timing or simple timed adjustment. Such systems achieve traffic management through preset signal cycles and phase times, and are suitable for scenarios with relatively stable traffic flows. In addition, some research has begun to explore the use of data analysis and optimization algorithms to optimize signal control, but overall it is still in the experimental or small-scale application stage.

[0004] In the prior art, the fixed timing scheme cannot adapt to the dynamic changes of traffic flow, easily resulting in waste of green light time or excessive vehicle queuing. In addition, the existing methods have limitations in data processing and decision-making efficiency, and it is difficult to meet the real-time control requirements of large-scale urban road networks. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide an urban traffic signal spatio-temporal adaptive flexible control method that can adapt to the dynamic changes of traffic flow and improve data processing and decision-making efficiency; on the other hand, to provide an urban traffic signal spatio-temporal adaptive flexible control system.

[0006] Technical Solution: The flexible control method of the present invention includes the following steps:

[0007] (1) Real-time collect traffic data such as the number of vehicles, vehicle speed, number of pedestrians and pedestrian speed at intersections through radar sensors, laser speedometers, infrared sensors and cameras, realizing holographic perception of dynamic data of vehicles and pedestrians at intersections, covering not only traditional traffic flow detection, but also accurately capturing the pedestrian crossing demand, providing high-precision, multi-dimensional real-time traffic data support for subsequent intelligent decision-making, and effectively solving the problem of incomplete data coverage of traditional single sensors;

[0008] (2) Clean and normalize the collected traffic data to eliminate noise interference and unify the data scale. At the same time, extract key features (such as traffic density, speed mutation points, etc.) to generate standardized traffic parameters, significantly improve data quality, provide high signal-to-noise ratio input for subsequent algorithms, and avoid model prediction bias caused by data heterogeneity;

[0009] (3) Based on the standardized traffic parameters, a signal light timing plan is generated through a hybrid algorithm of spatiotemporal graph convolutional network and lightweight deep reinforcement learning. This not only accurately predicts the evolution trend of traffic flow at the intersection in the future period, but also generates a signal timing plan that takes into account both efficiency and fairness in real time. Compared with the fixed timing strategy, it can reduce the average waiting time of vehicles, and the algorithm calculation efficiency meets the real-time requirements.

[0010] (4) Traffic lights are adjusted in real time according to the signal light timing scheme to achieve dynamic switching of signal states and ensure that traffic flow quickly returns to a balanced state. This is particularly applicable to sudden congestion or special event scenarios, and avoids the problem of worsening congestion caused by adjustment lag in traditional control systems.

[0011] (5) Traffic data and remotely monitored intersection status are stored, and information from neighboring intersections is integrated to optimize regional strategies. The strategy data of each intersection is integrated through federated learning to generate a global model to optimize the collaborative penalty coefficient, suppress the spread of local congestion to surrounding intersections, and form regional traffic flow self-healing capabilities. At the same time, the distribution of the global model can achieve zero-cost cold start adaptation of newly connected intersections.

[0012] Preferably, the step 2 adopts a parallel computing architecture to complete the cleaning, normalization and feature extraction of traffic data within the time window T; the standardized traffic parameter is the standardized traffic feature matrix X t ∈R N×d , where N is the number of intersections and d is the feature dimension, which significantly improves the data processing speed and feature consistency. This design not only solves the problem of real-time processing of massive multi-source heterogeneous traffic data, but also ensures the standardization and comparability of input data for subsequent intelligent decision-making algorithms, providing a highly timely and reliable data foundation for spatiotemporal correlation analysis and dynamic optimization, thereby reducing the system's response delay in complex traffic scenarios.

[0013] Preferably, step 3 includes:

[0014] (31) Spatiotemporal traffic characteristics modeling

[0015] Modeling intersections and adjacent roads as a spatiotemporal graph structure in, is a node set, representing the target intersection and its adjacent N-1 intersections; ε is an edge set, representing the connection relationship and distance weight between intersections; X t ∈RN×d is the traffic feature matrix at time t, containing traffic flow, average speed, and pedestrian flow data at each intersection. The dimension d is the number of features;

[0016] (32) Spatiotemporal graph convolution prediction

[0017] Predict the traffic state within the next Δt time through the Spatiotemporal Graph Convolutional Network (ST-GCN):

[0018]

[0019] Among them, the inter-layer propagation formula of ST-GCN is:

[0020]

[0021] Among them, is the normalized adjacency matrix, calculated from the original adjacency matrix A and degree matrix D. The formula is H (l) is the node feature matrix of the l-th layer, with a dimension of N×d, where N is the number of intersections and d is the feature dimension. W (l) is the trainable weight matrix of the l-th layer, used for spatial graph convolution. T (l) is the temporal convolution kernel, and σ is the ReLU activation function;

[0022] (33) Lightweight deep reinforcement learning Light DRL dynamic optimization

[0023] Based on the prediction result of the traffic state Adopt the lightweight DRL policy network π θ to generate the signal timing action:

[0024] State space:

[0025] Action space: a t ∈ {Extend green light, Shorten red light, Switch phase}

[0026] Reward function:

[0027]

[0028] Among them, w i is the waiting time of the i-th vehicle, c i is the number of stops, Q throughput is the number of vehicles passing through per unit time, and α, λ are weight coefficients;

[0029] (34) Multi-objective collaborative optimization

[0030] Update the policy network parameters θ through the Proximal Policy Optimization (PPO) algorithm to maximize the cumulative reward:

[0031]

[0032] where π θ (a t |s t ) is the probability of the current policy network generating action a t , π old (a t |s t ) is the probability of the old policy network generating action a t , γ ∈ (0, 1) is the discount factor, T is the optimization period, ∈ is the Clip threshold, is the estimated value of the advantage function.

[0033] Precisely depict the spatio-temporal correlation between intersections through spatio-temporal graph structure modeling, and combine ST-GCN to realize the dynamic prediction of multi-dimensional traffic states, significantly improving the congestion prediction ability; build a "state-action-reward" closed-loop optimization system based on lightweight DRL, and generate the optimal timing plan in real time through a multi-objective reward function (taking into account waiting time, number of stops, and passing efficiency), and then dynamically adjust the policy parameters through the PPO algorithm to finally realize the spatio-temporal adaptive optimization of signal control. This solution improves the passing efficiency of intersections, reduces the calculation time compared with traditional methods, and combines prediction accuracy and decision-making real-time performance.

[0034] Preferably, the input data normalization formula of the spatio-temporal graph convolutional network ST-GCN is:

[0035]

[0036] where X is the original traffic feature matrix, including traffic flow, speed, and pedestrian flow data, μ X and σ X are the mean and standard deviation of historical data respectively.

[0037] Normalize the input data of ST-GCN based on the normalization formula of historical data mean and standard deviation, effectively eliminating the dimension difference and distribution deviation of traffic feature data at different intersections and different times, significantly improving the training stability and prediction accuracy of the spatio-temporal graph convolutional network. This normalization method enables the model to more accurately capture the spatio-temporal evolution law of traffic flow, and at the same time reduces the gradient anomaly problem caused by inconsistent data scales, providing a reliable input benchmark for subsequent reinforcement learning decision-making.

[0038] Preferably, the lightweight DRL policy network adopts a two-layer fully connected neural network, the hidden layer dimension is 64, the activation function is ReLU, and the output layer uses the Softmax function to generate the action probability distribution.

[0039] A lightweight DRL policy network is constructed by adopting a two-layer fully connected neural network with a hidden layer dimension of 64, combined with the efficient non-linear processing ability of the ReLU activation function and the probabilistic decision-making mechanism of the Softmax output layer. While ensuring the lightweight of the model, it realizes the rapid inference and optimal selection of signal timing actions; the network structure design not only meets the requirements of real-time traffic signal control for low latency, but also avoids sudden changes in the timing plan through the probability distribution output, making the signal switching smoother and more stable, thus improving the vehicle passing efficiency.

[0040] Preferably, the federated learning described in step 5 includes:

[0041] (51) Locally train the DRL policy network at each intersection to generate encrypted experience pool data

[0042] (52) The cloud aggregates the model parameters through the dynamic weighted average algorithm:

[0043]

[0044] Wherein, is the data volume of intersection i, and θ i is the local DRL policy network parameter of intersection i; is Gaussian noise, which meets the differential privacy requirements;

[0045] (53) Send the global model Θ global to each intersection for cold start optimization of newly connected intersections.

[0046] This solution not only realizes cross-intersection knowledge sharing and regional collaborative optimization while protecting the data privacy of each intersection, but also supports the rapid cold start adaptation of newly connected intersections. At the same time, the Gaussian noise injection mechanism effectively guarantees the security of traffic data, meeting the privacy protection requirements while improving the overall traffic control efficiency.

[0047] Preferably, the formula for optimizing the collaborative penalty coefficient described in step 5 is:

[0048]

[0049] Wherein, β is the penalty coefficient, β0 is the initial penalty coefficient, is the predicted traffic flow of intersection i at time t + 1, is the predicted traffic flow of intersection j at time t + 1, and τ is the temperature parameter, which controls the adjustment rate.

[0050] By introducing a dynamic penalty coefficient formula regulated based on temperature parameters, intelligent adaptive adjustment of the traffic congestion diffusion intensity is achieved. Through the collaborative effect of the initial penalty coefficient and temperature parameters, this formula can dynamically adjust the penalty intensity according to the real-time traffic situation, avoiding the over-regulation or under-regulation problems caused by traditional fixed penalty coefficients while ensuring the regional traffic collaborative optimization effect. This enables the system to more precisely suppress the spread of local congestion to surrounding intersections and improve the traffic coordination of the overall road network.

[0051] The flexible regulation system described in this invention includes:

[0052] A traffic data collection module, used to collect vehicle quantity, speed data, pedestrian quantity, and speed data at intersections in real time through radar sensors, laser speedometers, infrared sensors, and cameras.

[0053] A data processing module, used to clean, normalize, and extract features from the collected data to generate standardized traffic parameters.

[0054] An intelligent decision-making module, used to generate a signal timing plan based on the standardized traffic parameters through a hybrid algorithm of spatio-temporal graph convolution network and lightweight deep reinforcement learning.

[0055] An execution control module, used to adjust traffic signals in real time according to the signal timing plan.

[0056] A cloud platform module, used to store traffic data, remotely monitor intersection status, integrate information of adjacent intersections for regional strategy optimization, and integrate strategy data of each intersection through federated learning to generate a global model to optimize the collaborative penalty coefficient and suppress local congestion diffusion.

[0057] A computer-readable storage medium, on which a computer program is stored. The feature is that when the computer program is executed by a processor, it implements the urban traffic signal spatio-temporal adaptive flexible regulation method described above.

[0058] An electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor. The feature is that when the processor executes the program, it implements the urban traffic signal spatio-temporal adaptive flexible regulation method described above.

[0059] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: 1. It can quickly respond to the dynamic changes in traffic flow, optimize the signal timing plan, significantly improve the data processing speed and decision-making efficiency, and enhance the adaptability and robustness of the system in complex traffic environments; 2. It can accurately predict the traffic flow trend and adjust the traffic lights in real time, reduce the vehicle waiting time and the number of stops, improve the road traffic efficiency, and effectively relieve local congestion; 3. It realizes the collaborative optimization of regional traffic strategies, inhibits the spread of congestion, and at the same time reduces carbon emissions caused by traffic stagnation, contributing to the green and sustainable development of the city. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic structural diagram of the system of the present invention;

[0061] Figure 2 It is a flowchart of the method. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0063] As Figure 1 shown, the urban traffic signal spatio-temporal adaptive flexible control system of the present invention includes the following modules:

[0064] A traffic data collection module for collecting vehicle quantity and speed data through radar sensors and laser speedometers, and collecting pedestrian quantity and speed data through infrared sensors and cameras;

[0065] A data processing module for cleaning, normalizing, and feature extracting the data collected by the traffic data collection module to generate standardized traffic parameters;

[0066] An intelligent decision-making module for generating a signal timing plan through a hybrid algorithm of spatio-temporal graph convolutional network (ST-GCN) and lightweight deep reinforcement learning (Light DRL) based on the traffic parameters output by the data processing module;

[0067] An execution control module for adjusting the traffic lights in real time according to the timing plan generated by the intelligent decision-making module;

[0068] A cloud platform module for storing traffic data, remotely monitoring the intersection status, integrating adjacent intersection information for regional strategy optimization, and integrating the intersection strategy data through federated learning to generate a global model to optimize the collaborative penalty coefficient β and inhibit the spread of local congestion.

[0069] The data processing module adopts a parallel computing architecture to complete data cleaning and feature extraction within the time window T, and outputs a standardized traffic feature matrix X t ∈R N×d, where N is the number of intersections and d is the feature dimension.

[0070] The algorithm of the intelligent decision-making module includes the following steps:

[0071] (31) Spatiotemporal traffic feature modeling

[0072] Model the intersections and adjacent roads as a spatiotemporal graph structure

[0073]

[0074] Among them, is the set of nodes, representing the target intersection and its adjacent N - 1 intersections; ε is the set of edges, representing the connection relationship and distance weight between intersections; X t ∈R N×d is the traffic feature matrix at time t, containing traffic flow, average speed, and pedestrian flow data of each intersection, and the dimension d is the number of features.

[0075] (32) Spatiotemporal graph convolution prediction

[0076] Predict the traffic state within the next Δt time through the Spatiotemporal Graph Convolutional Network (ST-GCN):

[0077]

[0078] The input data normalization formula of the Spatiotemporal Graph Convolutional Network (ST-GCN) is:

[0079]

[0080] Among them, X is the original traffic feature matrix, containing traffic flow, speed, and pedestrian flow data, μ X and σ X are the mean and standard deviation of the historical data respectively.

[0081] The inter-layer propagation formula of ST-GCN is:

[0082]

[0083] Among them, is the normalized adjacency matrix, calculated from the original adjacency matrix A and degree matrix D, and the formula is H (l) is the node feature matrix of the l-th layer, with a dimension of N×d, where N is the number of intersections and d is the feature dimension, and W (l) is the trainable weight matrix of the l-th layer, used for spatial graph convolution, T (l) is the time convolution kernel, and σ is the ReLU activation function.

[0084] (33) Lightweight Deep Reinforcement Learning (Light DRL) dynamic optimization

[0085] Based on the prediction results Adopt a lightweight DRL policy network π θ Generate signal timing actions:

[0086] State space:

[0087] Action space: a t ∈ {Extend green light, Shorten red light, Switch phase}

[0088] Reward function:

[0089]

[0090] where w i is the waiting time (seconds) of the i-th vehicle, c i is the number of stops, Q throughput is the number of vehicles passing through per unit time, and α, λ are weight coefficients.

[0091] The lightweight DRL policy network uses a two-layer fully connected neural network, with a hidden layer dimension of 64, a ReLU activation function, and a Softmax function in the output layer to generate the action probability distribution.

[0092] (34) Multi-objective collaborative optimization

[0093] Update the policy network parameters θ through the Proximal Policy Optimization (PPO) algorithm to maximize the cumulative reward:

[0094]

[0095] where π θ (a t |s t ) is the probability of the current policy network generating action a t , π old (a t |s t ) is the probability of the old policy network generating action a t , γ ∈ (0, 1) is the discount factor, T is the optimization period, ∈ is the Clip threshold, is the estimated value of the advantage function.

[0096] Federated learning of the cloud platform module is achieved through the following steps:

[0097] (51) Locally train the DRL policy network at each intersection to generate encrypted experience pool data

[0098] (52) The cloud aggregates the model parameters through the dynamic weighted average algorithm:

[0099]

[0100] Among them, is the data volume of intersection i, and θ i are the local DRL policy network parameters of intersection i; is Gaussian noise, which meets the differential privacy requirements.

[0101] (53) The global model Θ global is sent to each intersection for cold start optimization of newly connected intersections.

[0102] As Figure 2 shown, the urban traffic signal spatio-temporal adaptive flexible regulation method described in the present invention includes the following steps:

[0103] Real-time collect traffic data such as the number of vehicles, vehicle speed, number of pedestrians, and pedestrian speed at intersections through radar sensors, laser speedometers, infrared sensors, and cameras;

[0104] Clean, normalize, and extract features from the collected traffic data to generate a standardized feature matrix X t ;

[0105] Based on the standardized traffic parameters, predict future traffic flow through ST-GCN and generate a real-time timing plan based on Light DRL;

[0106] (4) Adjust traffic signals in real time according to the signal timing plan;

[0107] (5) Store traffic data and the status of intersections monitored remotely, integrate information of neighboring intersections for regional strategy optimization, and integrate policy data of each intersection through federated learning to generate a global model to optimize the collaborative penalty coefficient β and suppress the spread of local congestion.

[0108] The dynamic adjustment formula of the collaborative penalty coefficient β is:

[0109]

[0110] Among them, β is the penalty coefficient, β0 is the initial penalty coefficient, is the predicted traffic flow of intersection i at time t + 1, is the predicted traffic flow of intersection j at time t + 1, and τ is the temperature parameter that controls the adjustment rate.

[0111] The present invention also discloses a computer device.

[0112] Specifically, the computer device can be a desktop computer, a laptop computer, a handheld computer, a cloud server, or other computer devices. The computer device can include, but is not limited to, a processor and a memory. Among them, the processor and the memory can be connected through a bus or other means. Among them, the processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, graphics processing units (GPUs), embedded neural network processors (NPUs), or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or combinations of the above types of chips.

[0113] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory. The memory can include a program storage area and a data storage area. Among them, the program storage area can store control units and application programs required for at least one function; the data storage area can store data created by the processor, etc. In addition, the memory can include high-speed random access memory and can also include non-transitory memory. In some embodiments, the memory optionally includes a memory remotely provided relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0114] The present invention also discloses a computer-readable storage medium.

[0115] Specifically, a computer-readable storage medium is used to store a computer program. When the computer program is executed by a processor, the methods in the above method embodiments are implemented. Those skilled in the art can understand that to implement all or part of the processes in the above method embodiments of the present application, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above types of memories.

Claims

1. An urban traffic signal spatio-temporal adaptive flexible regulation method, characterized in that, It includes the following steps: (1) Real-time collect traffic data such as the number of vehicles, vehicle speeds, the number of pedestrians, and pedestrian speeds at intersections through radar sensors, laser speedometers, infrared sensors, and cameras; (2) Clean, normalize, and extract features from the collected traffic data to generate standardized traffic parameters; (3) Based on the standardized traffic parameters, generate a signal timing plan through a hybrid algorithm of spatio-temporal graph convolutional network and lightweight deep reinforcement learning; (4) Adjust traffic signals in real time according to the signal timing plan; (5) Store traffic data and the status of remotely monitored intersections, integrate information of neighboring intersections for regional strategy optimization, and integrate strategy data of each intersection through federated learning to generate a global model to optimize the collaborative penalty coefficient.

2. The flexible regulation method according to claim 1, wherein The step 2 adopts a parallel computing architecture to complete the cleaning, normalization and feature extraction of traffic data within the time window T; the standardized traffic parameters are the standardized traffic feature matrix X t ∈R N×d , where N is the number of intersections and d is the feature dimension.

3. The flexible regulation method according to claim 1, wherein Step 3 includes: (31) Spatio-temporal traffic feature modeling Model the intersection and adjacent roads as a spatio-temporal graph structure Among them, is the set of nodes, representing the target intersection and its adjacent N-1 intersections; ε is the set of edges, representing the connection relationship and distance weight between intersections; X t ∈R N×d is the traffic feature matrix at time t, including traffic flow, average speed, and pedestrian flow data of each intersection. The dimension d is the number of features; (32) Spatio-temporal graph convolutional prediction Predict the traffic state within the next Δt time through the spatio-temporal graph convolutional network ST-GCN: Among them, the inter-layer propagation formula of ST-GCN is: Among them, is the normalized adjacency matrix, which is calculated from the original adjacency matrix A and the degree matrix D. The formula is H (l) is the node feature matrix of the l-th layer, with a dimension of N×d, where N is the number of intersections and d is the feature dimension. W (l) is the trainable weight matrix of the l-th layer, which is used for spatial graph convolution. T (l) is the time convolution kernel, and σ is the ReLU activation function; (33) Lightweight deep reinforcement learning Light DRL dynamic optimization Prediction results based on traffic conditions Adopt a lightweight DRL policy network π θ Generate signal timing actions: State space: Action space: a t ∈ {lengthen green light, shorten red light, switch phase} Reward function: Among them, w i is the waiting time of the i-th vehicle, c i is the number of parking times, Q throughput is the number of vehicles passing through per unit time, and α and λ are weight coefficients; (34) Multi-objective collaborative optimization Update the policy network parameter θ through the proximal policy optimization PPO algorithm to maximize the cumulative reward: where, π θ (a t |s t ) is the probability of the current policy network generating action a t , π old (a t |s t ) is the probability of the old policy network generating action a t , γ ∈ (0, 1) is the discount factor, T is the optimization period, ∈ is the Clip threshold, is the estimated value of the advantage function.

4. The flexible regulation method according to claim 3, characterized in that The input data normalization formula of the spatio-temporal graph convolutional network ST-GCN is: Among them, X is the original traffic feature matrix, including traffic flow, speed, and pedestrian flow data, μ X and σ X are the mean and standard deviation of historical data, respectively.

5. The flexible regulation method according to claim 3, wherein The lightweight DRL policy network adopts a two-layer fully connected neural network, the hidden layer dimension is 64, the activation function is ReLU, and the output layer uses the Softmax function to generate the action probability distribution.

6. The flexible regulation method according to claim 1, characterized in that, The federated learning described in step 5 includes: (51) Locally train the DRL policy network at each intersection to generate encrypted experience pool data (52) The cloud aggregates model parameters through the dynamic weighted average algorithm: Among them, is the data volume of intersection i, and θ i are the local DRL policy network parameters of intersection i; is Gaussian noise, satisfying the differential privacy requirement; (53) Distribute the global model Θ global to each intersection for cold start optimization of newly connected intersections.

7. The flexible regulation method according to claim 1, wherein The formula for optimizing the collaborative penalty coefficient described in step 5 is: Among them, β is the penalty coefficient, and β0 is the initial penalty coefficient. is the predicted traffic flow at intersection i at time t + 1. is the predicted traffic flow at intersection j at time t + 1, and τ is the temperature parameter that controls the adjustment rate.

8. An urban traffic signal spatio-temporal adaptive flexible control system, characterized in that, It includes: A traffic data collection module for real-time collecting the number of vehicles, speed data, the number of pedestrians, and speed data at intersections through radar sensors, laser speedometers, infrared sensors, and cameras; A data processing module for cleaning, normalizing, and extracting features from the collected data to generate standardized traffic parameters; An intelligent decision-making module for generating a signal timing plan based on the standardized traffic parameters through a hybrid algorithm of spatio-temporal graph convolutional network and lightweight deep reinforcement learning; An execution control module for adjusting traffic signals in real time according to the signal timing plan; A cloud platform module for storing traffic data, remotely monitoring the status of intersections, integrating information of neighboring intersections for regional strategy optimization, and integrating strategy data of each intersection through federated learning to generate a global model to optimize the collaborative penalty coefficient and inhibit the spread of local congestion.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the urban traffic signal spatio-temporal adaptive flexible regulation method described in any one of claims 1 to 7.

10. An electronic device, comprising a memory, a processor, and a program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the urban traffic signal spatio-temporal adaptive flexible regulation method described in any one of claims 1 to 7.

Images