A Spatiotemporal Adaptive Flexible Control Method and System for Urban Traffic Signals

Abstract

This invention discloses a spatiotemporal adaptive flexible control method and system for urban traffic signals. The method includes real-time collection of traffic data such as the number of vehicles, vehicle speed, number of pedestrians, and pedestrian speed at intersections; cleaning, normalizing, and extracting features from the collected traffic data to generate standardized traffic parameters; generating a traffic light timing scheme based on the standardized traffic parameters using a hybrid algorithm of spatiotemporal graph convolutional network and lightweight deep reinforcement learning; adjusting traffic lights in real-time according to the traffic light timing scheme; storing traffic data and remotely monitored intersection status, integrating information from neighboring intersections for regional strategy optimization, and integrating strategy data from various intersections through federated learning to generate a global model to optimize the collaborative penalty coefficient. This invention can quickly respond to dynamic changes in traffic flow, optimize traffic light timing schemes, and significantly improve data processing speed and decision-making efficiency.

Description

Technical Field

[0001] This invention belongs to the field of intelligent traffic management technology, and in particular relates to a spatiotemporal adaptive flexible control method and system for urban traffic signals. Background Technology

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

[0003] Currently, traffic signal control systems in many cities still primarily rely on fixed timing or simple timed adjustments. These systems manage traffic through preset signal cycles and phase times, making them suitable for scenarios with relatively stable traffic flow. Furthermore, some research has begun exploring the use of data analysis and optimization algorithms to improve signal control, but overall, this remains in the experimental or small-scale application stage.

[0004] In existing technologies, fixed timing schemes cannot adapt to dynamic changes in traffic flow, easily leading to wasted green light time or excessively long vehicle queues. Furthermore, existing methods have limitations in data processing and decision-making efficiency, making it difficult to meet the real-time control needs of large-scale urban road networks. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a spatiotemporally adaptive flexible control method for urban traffic signals that can adapt to dynamic changes in traffic flow and improve the efficiency of data processing and decision-making; on the other hand, it provides a spatiotemporally adaptive flexible control system for urban traffic signals.

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

[0007] (1) By using radar sensors, laser speed meters, infrared sensors and cameras to collect traffic data such as the number of vehicles, vehicle speed, number of pedestrians and pedestrian speed at intersections in real time, a holographic perception of the dynamic data of vehicles and pedestrians at intersections can be realized. This not only covers traditional traffic flow detection, but also accurately captures pedestrian crossing needs, providing high-precision, multi-dimensional real-time traffic data support for subsequent intelligent decision-making, effectively solving the problem of incomplete data coverage of traditional single sensors.

[0008] (2) The collected traffic data is cleaned and normalized to eliminate noise interference and unify the data scale. At the same time, key features (such as traffic density, speed change points, etc.) are extracted to generate standardized traffic parameters, which significantly improves the data quality, provides high signal-to-noise ratio input for subsequent algorithms, and avoids model prediction bias caused by data heterogeneity.

[0009] (3) Based on the standardized traffic parameters, a traffic light timing scheme is generated by a hybrid algorithm of spatiotemporal graph convolutional network and lightweight deep reinforcement learning. This not only accurately predicts the future traffic flow trend at intersections, but also generates a signal timing scheme that balances efficiency and fairness in real time. Compared with a fixed timing strategy, it can reduce the average waiting time of vehicles, and the algorithm's computational efficiency meets the real-time requirements.

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

[0011] (5) Store traffic data and remotely monitored intersection status, integrate information from neighboring intersections for regional strategy optimization, and integrate strategy data from each intersection through federated learning to generate a global model to optimize the collaborative penalty coefficient, suppress the spread of local congestion to surrounding intersections, form a regional traffic flow self-healing capability, and at the same time, the distribution of the global model can achieve zero-cost cold start adaptation for newly accessed intersections.

[0012] Preferably, step 2 employs a parallel computing architecture to complete the cleaning, normalization, and feature extraction of traffic data within a 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, this design significantly improves data processing speed and feature consistency. It not only solves the problem of real-time processing of massive multi-source heterogeneous traffic data, but also ensures the standardization and comparability of the input data for subsequent intelligent decision-making algorithms. It provides a highly timely and reliable data foundation for spatiotemporal correlation analysis and dynamic optimization, thereby reducing the system's response latency in complex traffic scenarios.

[0013] Preferably, step 3 includes:

[0014] (31) Spatiotemporal traffic feature modeling

[0015] Model the intersection and adjacent roads as a spatiotemporal graph structure. in, Let ε be the set of nodes, representing the target intersection and its N-1 neighboring intersections; let ε be the set of edges, representing the connection relationships and distance weights between intersections; X t ∈RN×d Let d be the traffic feature matrix at time t, containing traffic flow, average speed, and pedestrian flow data for each intersection, with dimension d representing the number of features.

[0016] (32) Spatiotemporal graph convolution prediction

[0017] Predicting traffic conditions in the future time interval Δt using the Spatiotemporal Graph Convolutional Network (ST-GCN):

[0018]

[0019] The interlayer propagation formula for ST-GCN is as follows:

[0020]

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

[0022] (33) Lightweight Deep Reinforcement Learning (Light DRL) Dynamic Optimization

[0023] Traffic condition-based prediction results The π network employs a lightweight DRL strategy. θ Generate traffic light timing actions:

[0024] State space:

[0025] Action space: a t ∈{Extend green light, shorten red light, switch phases}

[0026] Reward function:

[0027]

[0028] Among them, w i Let c be the waiting time for the i-th vehicle. i Q represents the number of stops. throughput α represents the number of vehicles passing through per unit time, and λ represents the weighting coefficients.

[0029] (34) Multi-objective collaborative optimization

[0030] The PPO algorithm is optimized using a near-end policy to update the policy network parameters θ, thereby maximizing the cumulative reward.

[0031]

[0032] Where, π θ (a t |s t Generate action a for the current policy network. t The probability, π old (a t |s t Generate action a for the old policy network. t The probability of γ∈(0,1) is given by the discount factor, T is the optimization period, and ∈ is the Clip threshold. This is the estimated value of the advantage function.

[0033] By accurately depicting the spatiotemporal correlations between intersections through spatiotemporal graph structure modeling, and combining it with ST-GCN to achieve dynamic prediction of multi-dimensional traffic conditions, the congestion prediction capability is significantly improved. A "state-action-reward" closed-loop optimization system is constructed based on a lightweight DRL, generating the optimal timing scheme in real time through a multi-objective reward function (considering waiting time, number of stops, and traffic efficiency). The PPO algorithm then dynamically adjusts the strategy parameters, ultimately achieving spatiotemporal adaptive optimization of signal control. This scheme improves intersection traffic efficiency while reducing computation time compared to traditional methods, combining predictive accuracy with real-time decision-making.

[0034] Preferably, the input data normalization formula for the spatiotemporal graph convolutional network ST-GCN is:

[0035]

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

[0037] The ST-GCN input data is standardized using a normalization formula based on the mean and standard deviation of historical data. This effectively eliminates the dimensional differences and distribution biases of traffic feature data at different intersections and time periods, significantly improving the training stability and prediction accuracy of the spatiotemporal graph convolutional network. This normalization method enables the model to more accurately capture the spatiotemporal evolution of traffic flow, while reducing gradient anomalies caused by inconsistent data scales, thus providing a reliable input benchmark for subsequent reinforcement learning decisions.

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

[0039] By constructing a lightweight DRL policy network using a two-layer fully connected neural network with a hidden layer dimension of 64, and combining the efficient nonlinear processing capability of the ReLU activation function with the probabilistic decision-making mechanism of the Softmax output layer, a rapid inference and optimization selection of signal timing actions is achieved while ensuring the model's lightweight nature. This network structure design not only meets the low latency requirements of real-time traffic signal control, but also avoids abrupt changes in timing schemes through probability distribution output, making signal switching smoother and more stable, thereby improving vehicle traffic efficiency.

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

[0041] (51) Train the DRL policy network locally at each intersection to generate encrypted experience pool data.

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

[0043]

[0044] in, Let θ be the amount of data at intersection i. i For the local DRL policy network parameters of intersection i; The noise is Gaussian to meet differential privacy requirements;

[0045] (53) Global model Θ global It is distributed to each intersection for cold start optimization of newly connected intersections.

[0046] This solution, while protecting the data privacy of each intersection, achieves cross-intersection knowledge sharing and regional collaborative optimization, supports rapid cold start adaptation of newly accessed intersections, and effectively ensures the security of traffic data through a Gaussian noise injection mechanism, thus improving the overall traffic management efficiency while meeting privacy protection requirements.

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

[0048]

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

[0050] By introducing a dynamic penalty coefficient formula based on temperature parameter control, intelligent adaptive adjustment of the intensity of traffic congestion spread is achieved. Through the synergistic effect of the initial penalty coefficient and temperature parameter, the formula can dynamically adjust the penalty intensity according to the real-time traffic situation. While ensuring the regional traffic coordination optimization effect, it avoids the over-control or under-control problems caused by the traditional fixed penalty coefficient, enabling the system to more accurately suppress the spread of local congestion to surrounding intersections and improve the overall traffic coordination of the road network.

[0051] The flexible control system of the present invention includes:

[0052] The traffic data acquisition module is used to collect real-time data on the number and speed of vehicles and pedestrians at intersections through radar sensors, laser speedometers, infrared sensors, and cameras.

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

[0054] The intelligent decision-making module is used to generate traffic light timing schemes based on the standardized traffic parameters by using a hybrid algorithm of spatiotemporal graph convolutional network and lightweight deep reinforcement learning.

[0055] The execution control module is used to adjust the traffic lights in real time according to the traffic light timing scheme;

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

[0057] A computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the spatiotemporal adaptive flexible control method for urban traffic signals.

[0058] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the spatiotemporal adaptive flexible control method for urban traffic signals.

[0059] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. It can quickly respond to the dynamic changes in traffic flow, optimize traffic light timing schemes, significantly improve data processing speed and decision-making efficiency, and enhance the system's adaptability and robustness in complex traffic environments; 2. It can accurately predict traffic flow trends and adjust traffic lights in real time, reduce vehicle waiting time and number of stops, improve road traffic efficiency, and effectively alleviate local congestion; 3. It can achieve coordinated optimization of regional traffic strategies, suppress the spread of congestion, and reduce carbon emissions caused by traffic stagnation, thus contributing to the green and sustainable development of cities. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0061] Figure 2 This is a flowchart of the method. Detailed Implementation

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

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

[0064] The traffic data acquisition module is used to collect vehicle quantity and speed data through radar sensors and laser speedometers, and to collect pedestrian quantity and speed data through infrared sensors and cameras;

[0065] The data processing module is used to clean, normalize, and extract features from the data collected by the traffic data acquisition module to generate standardized traffic parameters.

[0066] The intelligent decision-making module is used to generate traffic light timing schemes based on the traffic parameters output by the data processing module, using a hybrid algorithm of spatiotemporal graph convolutional network (ST-GCN) and lightweight deep reinforcement learning (Light DRL).

[0067] The execution control module is used to adjust the traffic lights in real time according to the timing scheme generated by the intelligent decision-making module;

[0068] The cloud platform module is used to store traffic data, remotely monitor intersection status, integrate information from neighboring intersections for regional strategy optimization, and integrate strategy data from various intersections through federated learning to generate a global model to optimize the collaborative penalty coefficient β and suppress the spread of local congestion.

[0069] The data processing module adopts a parallel computing architecture, completing data cleaning and feature extraction within a time window T, and outputting 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 for the intelligent decision-making module includes the following steps:

[0071] (31) Spatiotemporal traffic feature modeling

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

[0073]

[0074] in, Let ε be the set of nodes, representing the target intersection and its N-1 neighboring intersections; let ε be the set of edges, representing the connection relationships and distance weights between intersections; X t ∈R N×d Let d be the traffic feature matrix at time t, containing traffic flow, average speed, and pedestrian flow data at each intersection, with dimension d representing the number of features.

[0075] (32) Spatiotemporal graph convolution prediction

[0076] Predict traffic conditions in the future time interval Δt using a spatiotemporal graph convolutional network (ST-GCN):

[0077]

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

[0079]

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

[0081] The interlayer propagation formula for ST-GCN is:

[0082]

[0083] in, The normalized adjacency matrix is ​​calculated from the original adjacency matrix A and the degree matrix D, using the following formula: H (l) Let W be the node feature matrix of the l-th layer, with dimensions N×d, where N is the number of intersections, d is the feature dimension, and W is the feature matrix of the l-th layer. (l) T is the trainable weight matrix of the l-th layer, used for spatial graph convolution. (l) σ is the temporal 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 The π network employs a lightweight DRL strategy. θ Generate traffic light timing actions:

[0086] State space:

[0087] Action space: a t ∈{Extend green light, shorten red light, switch phases}

[0088] Reward function:

[0089]

[0090] Among them, w i Let c be the waiting time (in seconds) for the i-th vehicle. i Q represents the number of stops. throughput Let α represent the number of vehicles passing through per unit time, and λ represent weighting coefficients.

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

[0092] (34) Multi-objective collaborative optimization

[0093] The policy network parameters θ are updated using the Proximal Policy Optimization (PPO) algorithm to maximize the cumulative reward.

[0094]

[0095] Where, π θ (a t |s t Generate action a for the current policy network. t The probability, π old (a t |s t Generate action a for the old policy network. t The probability of γ∈(0,1) is given by the discount factor, T is the optimization period, and ∈ is the Clip threshold. This is the estimated value of the advantage function.

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

[0097] (51) Train the DRL policy network locally at each intersection to generate encrypted experience pool data.

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

[0099]

[0100] in, Let θ be the amount of data at intersection i. i For the local DRL policy network parameters of intersection i; It uses Gaussian noise to meet differential privacy requirements.

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

[0102] like Figure 2 As shown, the spatiotemporal adaptive flexible control method for urban traffic signals according to the present invention includes the following steps:

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

[0104] The collected traffic data is cleaned, normalized, and its features are extracted to generate a standardized feature matrix X. t ;

[0105] Based on the standardized traffic parameters, future traffic flow is predicted using ST-GCN. And generate real-time timing schemes based on Light DRL;

[0106] (4) Adjust the traffic lights in real time according to the traffic light timing scheme;

[0107] (5) Store traffic data and remotely monitored intersection status, integrate information from neighboring intersections for regional strategy optimization, and integrate strategy data from 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 for the collaborative penalty coefficient β is:

[0109]

[0110] Where β is the penalty coefficient, and β0 is the initial penalty coefficient. Let i be the predicted traffic flow at time t+1. Let τ be the predicted traffic flow at intersection j at time t+1, and τ be the temperature parameter, which controls the adjustment rate.

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

[0112] Specifically, the computer device can be a desktop computer, laptop computer, handheld computer, or cloud server. This computer device may include, but is not limited to, a processor and memory. The processor and memory can be connected via a bus or other means. 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 processing units (NPUs) or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0113] Memory, as a non-transitory computer-readable storage medium, 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 by running non-transitory software programs, instructions, and modules stored in memory. Memory may include a program storage area and a data storage area. The program storage area may store the control unit and the application program required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, memory may include high-speed random access memory and non-transitory memory. In some embodiments, memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

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

[0115] Specifically, the computer-readable storage medium is used to store a computer program, which, when executed by a processor, implements the methods described in the above-described method embodiments. Those skilled in the art will understand that implementing all or part of the processes in the methods described in the above-described embodiments of this application can be accomplished by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

Claims

1. A spatiotemporal adaptive flexible control method for urban traffic signals, characterized in that, Includes the following steps: (1) Real-time traffic data of the number of vehicles, vehicle speed, number of pedestrians and pedestrian speed at intersections are collected by radar sensors, laser speed meters, 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, a traffic light timing scheme is generated by a hybrid algorithm of spatiotemporal graph convolutional network and lightweight deep reinforcement learning; (31) Spatiotemporal traffic feature modeling: The intersection and adjacent roads are modeled as a spatiotemporal graph structure. ,in, Let N be a set of nodes, representing the target intersection and its N-1 neighboring intersections. Let be the set of edges, representing the connection relationships and distance weights between intersections; Let d be the traffic feature matrix at time t, containing traffic flow, average speed, and pedestrian flow data for each intersection, with dimension d representing the number of features. (32) Spatiotemporal graph convolutional prediction: Predict traffic conditions in the future time interval Δt using the spatiotemporal graph convolutional network ST-GCN. The interlayer propagation formula for ST-GCN is as follows: in, The normalized adjacency matrix is ​​calculated from the original adjacency matrix A and the degree matrix D, using the following formula: , For the first The node feature matrix of the layer has dimensions N×d, where N is the number of intersections and d is the feature dimension. For the first The trainable weight matrix of the layer is used for spatial graph convolution. For temporal convolution kernels, It is the ReLU activation function; (33) Lightweight Deep Reinforcement Learning (Light DRL) dynamically optimizes traffic state-based prediction results A lightweight DRL strategy network is adopted. Generate traffic light timing actions: State space: Action space: Reward function: in, Let be the waiting time for the i-th vehicle. For the number of stops, The number of vehicles passing through per unit time. , These are the weighting coefficients; (34) Multi-objective collaborative optimization updates policy network parameters through the PPO algorithm for near-end policy optimization. Maximize cumulative rewards: in, Generate actions for the current policy network The probability, Generate actions for the old policy network The probability, As a discount factor, To optimize the cycle, The Clip threshold, This is the estimated value of the dominance function; (4) Adjust the traffic lights in real time according to the traffic light timing scheme; (5) Store traffic data and remotely monitored intersection status, integrate information from neighboring intersections for regional strategy optimization, and integrate strategy data from each intersection through federated learning to generate a global model to optimize the collaborative penalty coefficient. The federated learning includes: Local training of DRL policy networks at each intersection generates encrypted experience pool data. ; The cloud aggregates model parameters using a dynamic weighted average algorithm: in, Let i be the amount of data at intersection i. For the local DRL policy network parameters of intersection i; The noise is Gaussian to meet differential privacy requirements; global model Distribute the data to each intersection for cold start optimization of newly connected intersections; The formula for the optimized collaborative penalty coefficient is: ;in, The penalty coefficient is... The initial penalty coefficient, Let i be the predicted traffic flow at time t+1. For the predicted traffic flow at intersection j at time t+1, The temperature parameter controls the adjustment rate.

2. The flexible control method according to claim 1, characterized in that, Step 2 employs a parallel computing architecture, within a time window. The system completes the cleaning, normalization, and feature extraction of traffic data; the standardized traffic parameters are a standardized traffic feature matrix. ,in Let d be the number of intersections and d be the feature dimension.

3. The flexible control method according to claim 1, characterized in that, The input data normalization formula for the spatiotemporal graph convolutional network ST-GCN is as follows: ;in, This is the original traffic feature matrix, containing vehicle flow, speed, and pedestrian flow data. and These are the mean and standard deviation of the historical data, respectively.

4. The flexible control method according to claim 1, characterized in that, The lightweight DRL policy network adopts a two-layer fully connected neural network with a hidden layer dimension of 64, an activation function of ReLU, and an output layer that uses the Softmax function to generate the action probability distribution.

5. A spatiotemporal adaptive flexible control system for urban traffic signals, used to implement the flexible control method described in claim 1, characterized in that, The system includes: The traffic data acquisition module is used to collect real-time data on the number and speed of vehicles and pedestrians at intersections through radar sensors, laser speedometers, infrared sensors, and cameras. The data processing module is used to clean, normalize, and extract features from the collected data to generate standardized traffic parameters. The intelligent decision-making module is used to generate traffic light timing schemes based on the standardized traffic parameters by using a hybrid algorithm of spatiotemporal graph convolutional network and lightweight deep reinforcement learning. The execution control module is used to adjust the traffic lights in real time according to the traffic light timing scheme; The cloud platform module is used to store traffic data, remotely monitor intersection status, integrate information from neighboring intersections for regional strategy optimization, and integrate strategy data from various intersections through federated learning to generate a global model to optimize the collaborative penalty coefficient and suppress the spread of local congestion.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the spatiotemporal adaptive flexible control method for urban traffic signals as described in any one of claims 1 to 4.

7. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the spatiotemporal adaptive flexible control method for urban traffic signals according to any one of claims 1 to 4.

Images