Adaptive signal control method for malformed intersection based on LSTM-GNN

Abstract

The invention discloses a malformed intersection adaptive signal control method based on LSTM-GNN, and belongs to the technical field of intelligent traffic systems, and the method comprises the steps: obtaining the historical traffic data of a malformed intersection, and carrying out the cleaning and preprocessing operation; based on an LSTM model, inputting historical traffic data, and extracting key time characteristics of signal phase duration; establishing a graph structure between the lanes, and capturing spatial interaction characteristics between the lanes by adopting GNN; adding traffic flow density into the LSTM model, adding lane priority weight when GNN captures spatial interaction characteristics between lanes, and dynamically adjusting a control strategy of a signal lamp; the method is verified through comparison between the SUMO simulation platform and a fixed signal control scheme. The signal lamp phase duration can be dynamically adjusted, the traffic flow is optimized, the vehicle waiting time is shortened, and the passing efficiency is improved; an intelligent solution is provided for urban traffic management, and theoretical support is provided for relieving traffic jam and improving road traffic capacity.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation systems, and in particular to an adaptive signal control method for deformed intersections based on LSTM-GNN. Background Art

[0002] With the acceleration of urbanization, traffic congestion is becoming increasingly serious, especially at irregular intersections (such as Y-shaped, X-shaped, and double-T-shaped intersections). Traffic flow becomes more complex, traffic efficiency is low, and congestion and accidents are likely to occur. Traditional fixed signal control methods are difficult to adapt to the dynamic traffic demands of irregular intersections, resulting in long vehicle wait times and low traffic efficiency. While existing intelligent traffic signal control technologies have made some progress in improving traffic flow and reducing congestion, they still face some challenges.

[0003] For example, deep learning technology requires large amounts of high-quality data and long training times, and is prone to overfitting. Therefore, designing an adaptive signal control method that can adapt to the complex traffic flows at irregular intersections, reduce vehicle wait times, and improve traffic efficiency has become a pressing issue in the field of intelligent transportation systems. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides an adaptive signal control method for deformed intersections based on LSTM-GNN.

[0005] To implement the above technology, the specific steps are as follows: S1. Acquire historical traffic data of the deformed intersection, and perform cleaning and preprocessing operations on the acquired historical traffic data to obtain preprocessed historical traffic data; Historical traffic data includes: vehicle count, speed, waiting time, and lane occupancy; Cleaning and preprocessing operations include: removing outliers (vehicle speed is 0), filling missing values, and data standardization; In order to ensure feasibility and data consistency, the present invention needs to first set the test parameters and define the lane and phase numbers of the intersection; The test parameters include: intersection type, number of approach points, number of signal phases, processor, graphics card, operating system, simulation software, and simulation duration; Define the lane and phase numbers for the intersection based on the intersection type, number of approaches, and number of signal phases.

[0006] S2, based on the LSTM model, takes the pre-processed historical traffic data as input, learns the time-dependent characteristics of traffic flow, and extracts the key time features of signal phase duration; The key time feature of the signal phase duration is extracted by calculating the time correlation of traffic flow through the output of the LSTM hidden layer nodes; the hidden layer includes: input gate , Forget Gate , candidate memory , memory cells , output gate and hidden state output ; First, the hidden state of the previous moment and the current input (historical traffic data in this invention) are obtained through the input gate to determine how much information needs to be written into the memory cell. In this invention, the information that needs to be written into the memory cell includes: number of vehicles, speed, waiting time and lane occupancy rate; The forget gate uses the same input as the input gate and decides whether to forget outdated memory based on vehicle speed and waiting time. The candidate memory cell calculation uses the same input as the input gate and is used to generate new content. The updated content is obtained through memory cell updates. The influence of the current memory on the traffic light decision is obtained through the output gate calculation, and the output feature is obtained through the hidden state output; Input Gate , Forget Gate , candidate memory , memory cells , output gate and hidden state output The expression is as follows: ; ; ; ; ; ; Where, Represents the Sigmoid activation function; 、 、 and Represent the weight items of input gate, forget gate, candidate memory, and output gate respectively; 、 、 and Specifically represent the bias items of input gate, forget gate, candidate memory, and output gate; represents the input of the current time step, which in this invention represents the number of vehicles, vehicle speed, waiting time and lane occupancy rate; represents the hidden state of the previous time step; Hyperbolic tangent activation function; Represents the state of the memory cell at the previous time step; Through the above mechanism, LSTM can effectively capture the time dependency of traffic flow and extract the key time features of signal phase duration.

[0007] S3. Based on the road topology of the intersection, a graph structure between lanes is established, and GNN is used to capture the spatial interaction characteristics between lanes; The graph structure between lanes is established by representing each lane as a node and defining an adjacency matrix based on the intersection topology to represent the edge relationships between lanes. The expression for using GNN to capture the spatial interaction characteristics between lanes is as follows: Where, Indicates the l The feature matrix of the layer; represents the adjacency matrix; represents the degree matrix; Indicates the l The trainable weight matrix of the layer; Represents the activation function.

[0008] S4. Traffic flow density is incorporated into the LSTM model, and lane priority weights are added to the GNN to capture the spatial interaction characteristics between lanes. Combining the spatiotemporal features extracted by the LSTM model and the GNN, the optimal duration of each signal phase is calculated. Based on the real-time traffic flow at the deformed intersection, the traffic light control strategy is dynamically adjusted. To address the particularities of irregular intersections, the nonlinear mapping function of the LSTM-GNN model was improved. Specifically, the traffic flow density feature was introduced into the LSTM component to more accurately reflect the time dependence of traffic flow. Lane priority weights were added to the GNN component to reflect the varying importance of different lanes in the intersection. The expression of spatiotemporal features extracted by combining the LSTM model and GNN is as follows: Where, represents the real-time traffic flow input features, which in the present invention include vehicle volume, vehicle speed, waiting time, and lane occupancy; It represents the traffic flow density, which is calculated by the ratio of traffic volume to lane length; Indicates the lane priority weight, which is determined by weighting according to the lane's travel direction, connected road grade, and functional factors; represents the weight of traffic flow density; weight representing the lane priority weight; By introducing the above improvements, the model can more accurately capture the traffic flow characteristics of deformed intersections, thereby more effectively optimizing signal phase duration and improving traffic flow efficiency.

[0009] S5. Apply the obtained traffic light control strategy on the SUMO simulation platform and compare and analyze it with the fixed signal control scheme to verify the effectiveness of this method; Multiple rounds of experiments were conducted on the SUMO simulation platform to compare the traffic efficiency differences between the traffic light control strategy and the fixed signal control strategy. The evaluation indicators included: the cumulative number of passing vehicles, the average waiting time, and the number of signal phase adjustments.

[0010] Beneficial effects of the present invention By combining long short-term memory networks (LSTM) and graph neural networks (GNN), the present invention can dynamically adjust the phase duration of traffic lights, optimize traffic flow, reduce vehicle waiting time, and improve traffic efficiency; thereby providing an intelligent solution for urban traffic management and offering theoretical support for alleviating traffic congestion and improving road capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a flow chart of the steps of the present invention; Figure 2 It is a system framework diagram of the present invention; Figure 3 LSTM network structure diagram of the present invention; Figure 4 This is the GNN network structure diagram of the present invention; Figure 5 This is a flowchart of the time series data prediction of the LSTM-GNN network of the present invention; Figure 6 This is a schematic diagram of the intersection flow, lane numbering and lane function of the present invention; Figure 7 A comparison diagram of the phase duration frequency distribution in the embodiment of the present invention and the fixed signal control model; Figure 8 A comparison diagram of vehicle waiting time between an embodiment of the present invention and a fixed signal control model; Figure 9 This is a comparison diagram of the accumulated passing vehicles in the embodiment of the present invention and the fixed signal control model. DETAILED DESCRIPTION

[0012] The present invention is further described in detail below with reference to specific embodiments.

[0013] like Figure 1 and Figure 2 As shown, an adaptive signal control method for deformed intersections based on LSTM-GNN includes the following steps: S1. Acquire historical traffic data of the deformed intersection, and perform cleaning and preprocessing operations on the acquired historical traffic data to obtain preprocessed historical traffic data; Historical traffic data includes: vehicle count, speed, waiting time, and lane occupancy; Cleaning and preprocessing operations include: removing outliers (vehicle speed is 0), filling missing values, and data standardization; In order to ensure feasibility and data consistency, the present invention needs to first set the test parameters and define the lane and phase numbers of the intersection; The test parameters include: intersection type, number of approach points, number of signal phases, processor, graphics card, operating system, simulation software, and simulation duration; Define the lane and phase numbers of the intersection based on the intersection type, number of entrances, and number of signal phases; In this embodiment, the test parameters are set as shown in Table 1; Table 1 Test parameters In this embodiment, the intersection lane and phase numbers are shown in Table 2 and Figure 3 As shown; Table 2 Intersection lane and phase number table S2, such as Figure 3 As shown in the figure, based on the LSTM model, the pre-processed historical traffic data is used as input to learn the time-dependent characteristics of traffic flow and extract the key time features of signal phase duration; The key time feature of the signal phase duration is extracted by calculating the time correlation of traffic flow through the output of the LSTM hidden layer nodes; the hidden layer includes: input gate , Forget Gate , candidate memory , memory cells , output gate and hidden state output ; First, the hidden state of the previous moment and the current input (historical traffic data in this invention) are obtained through the input gate to determine how much information needs to be written into the memory cell. In this invention, the information that needs to be written into the memory cell includes: number of vehicles, speed, waiting time and lane occupancy rate; The forget gate uses the same input as the input gate and decides whether to forget outdated memory based on vehicle speed and waiting time. The candidate memory cell calculation uses the same input as the input gate and is used to generate new content. The updated content is obtained through memory cell updates. The influence of the current memory on the traffic light decision is obtained through the output gate calculation, and the output feature is obtained through the hidden state output; Input Gate , Forget Gate , candidate memory , memory cells , output gate and hidden state output The expression is as follows: ; ; ; ; ; ; Where, Represents the Sigmoid activation function; 、 、 and Represent the weight items of input gate, forget gate, candidate memory, and output gate respectively; 、 、 and Specifically represent the bias items of input gate, forget gate, candidate memory, and output gate; represents the input of the current time step, which in this invention represents the number of vehicles, vehicle speed, waiting time and lane occupancy rate; represents the hidden state of the previous time step; Hyperbolic tangent activation function; Represents the state of the memory cell at the previous time step; Through the above mechanism, LSTM can effectively capture the time dependency of traffic flow and extract the key time features of signal phase duration.

[0014] S3. Based on the road topology of the intersection, a graph structure between lanes is established, and GNN is used to capture the spatial interaction characteristics between lanes, such as Figure 4 As shown; The graph structure between lanes is established by representing each lane as a node and defining an adjacency matrix based on the intersection topology to represent the edge relationships between lanes. The expression for using GNN to capture the spatial interaction characteristics between lanes is as follows: Where, Indicates the l The feature matrix of the layer; represents the adjacency matrix; represents the degree matrix; Indicates the l The trainable weight matrix of the layer; Represents the activation function.

[0015] S4, such as Figure 5 As shown in the figure, traffic flow density is incorporated into the LSTM model, and lane priority weights are added when the GNN captures the spatial interaction characteristics between lanes. The LSTM model and the spatiotemporal features extracted by the GNN are combined to calculate the optimal duration of each signal phase. Based on the real-time traffic flow at the deformed intersection, the traffic light control strategy is dynamically adjusted. To address the particularities of irregular intersections, the nonlinear mapping function of the LSTM-GNN model was improved. Specifically, the traffic flow density feature was introduced into the LSTM component to more accurately reflect the time dependence of traffic flow. Lane priority weights were added to the GNN component to reflect the varying importance of different lanes in the intersection. The expression of spatiotemporal features extracted by combining the LSTM model and GNN is as follows: Where, represents the real-time traffic flow input features, which in the present invention include vehicle volume, vehicle speed, waiting time, and lane occupancy; It represents the traffic flow density, which is calculated by the ratio of traffic volume to lane length; Indicates the lane priority weight, which is determined by weighting according to the lane's travel direction, connected road grade, and functional factors; represents the weight of traffic flow density; weight representing the lane priority weight; By introducing the above improvements, the model can more accurately capture the traffic flow characteristics of deformed intersections, thereby more effectively optimizing signal phase duration and improving traffic flow efficiency.

[0016] S5. Apply the obtained traffic light control strategy on the SUMO simulation platform and compare and analyze it with the fixed signal control scheme to verify the effectiveness of this method; like Figure 6 As shown in the figure, multiple rounds of experiments were conducted on the SUMO simulation platform to compare the traffic efficiency differences between the traffic light control strategy and the fixed signal control strategy. The evaluation indicators include: the cumulative number of passing vehicles, the average waiting time and the number of signal phase adjustments; In this embodiment, the experimental setting is as follows: in the SUMO simulation environment, an "X"-shaped deformed intersection is set up, the number of entrances is 6, the number of signal phases is 15, and the simulation time is 1000 seconds. The signal control strategy of the present invention and the traditional fixed signal control strategy are respectively applied for comparative experiments; Comparison results: like Figure 7 As shown, the signal phase duration of the present invention is higher than that of the fixed signal control strategy, ensuring that the green light lasts longer during the lane peak hours, effectively improving traffic efficiency; like Figure 8 As shown in the figure, the average waiting time is: under the fixed signal control strategy, the average waiting time is within 10 to 20 seconds, and some even exceed 20 seconds; under the signal control strategy of the present invention, the average waiting time is 5 to 15 seconds, and the maximum does not exceed 20 seconds; like Figure 9 As shown in the figure, the cumulative number of passing vehicles: under the simulation time of the fixed signal control strategy, the cumulative number of passing vehicles is 45,000; under the simulation time of the signal control strategy of the present invention, the cumulative number of passing vehicles is 50,000, which is an increase of 10%; Through the comparison of these specific data, it can be clearly seen that the improved LSTM-GNN signal control strategy of the present invention has significant advantages in improving traffic efficiency and reducing vehicle waiting time.

[0017] It should be noted that the above are only preferred embodiments of the present application and do not limit the scope of patent protection of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly applied in other related technical fields, are also included in the scope of patent protection of the present application.

Claims

1. An adaptive signal control method for deformed intersections based on LSTM-GNN, characterized by: The following steps are involved: S1. Acquire historical traffic data of the deformed intersection, and perform cleaning and preprocessing operations on the acquired historical traffic data to obtain preprocessed historical traffic data; S2, based on the LSTM model, takes the pre-processed historical traffic data as input, learns the time-dependent characteristics of traffic flow, and extracts the key time features of signal phase duration; S3. Based on the road topology of the intersection, a graph structure between lanes is established, and GNN is used to capture the spatial interaction characteristics between lanes; S4. Traffic flow density is incorporated into the LSTM model, and lane priority weights are added to the GNN to capture the spatial interaction characteristics between lanes. Combining the spatiotemporal features extracted by the LSTM model and the GNN, the optimal duration of each signal phase is calculated. Based on the real-time traffic flow at the deformed intersection, the traffic light control strategy is dynamically adjusted to achieve adaptive signal control. S5. Apply the obtained traffic light control strategy on the SUMO simulation platform and compare and analyze it with the fixed signal control scheme to verify the effectiveness of this method; The evaluation indicators for comparative analysis with the fixed signal control scheme include: the cumulative number of passing vehicles, the average waiting time and the number of signal phase adjustments.

2. The adaptive signal control method for deformed intersections based on LSTM-GNN according to claim 1 is characterized in that: The method obtains historical traffic data of the deformed intersection and performs cleaning and preprocessing operations on the obtained historical traffic data to obtain the preprocessed historical traffic data, wherein the historical traffic data includes: number of vehicles, vehicle speed, waiting time and lane occupancy rate; the cleaning and preprocessing operations include: eliminating outliers, filling missing values, and data standardization.

3. The adaptive signal control method for deformed intersections based on LSTM-GNN according to claim 1 is characterized in that: The LSTM model uses pre-processed historical traffic data as input to learn the time-dependent characteristics of traffic flow and extract the key time characteristics of signal phase duration. The key time characteristics of signal phase duration are extracted in the following way: the time correlation of traffic flow is calculated through the output of the LSTM hidden layer nodes; wherein the hidden layer includes: input gate , Forget Gate , candidate memory , memory cells , output gate and hidden state output ; First, the hidden state of the previous moment and the current input are obtained through the input gate, which represents historical traffic data in this invention, so as to determine how much information needs to be written into the memory cell. In this invention, the information that needs to be written into the memory cell includes: number of vehicles, speed, waiting time and lane occupancy rate; The forget gate uses the same input as the input gate and decides whether to forget outdated memory based on vehicle speed and waiting time. The candidate memory cell calculation uses the same input as the input gate and is used to generate new content. The updated content is obtained through memory cell updates. Through the output gate calculation, the influence of the current memory on the traffic light decision is obtained, and the output feature is obtained through the hidden state output.

4. The adaptive signal control method for deformed intersections based on LSTM-GNN according to claim 3 is characterized in that: The input gate , Forget Gate , candidate memory , memory cells , output gate and hidden state output The expressions include: ； ； ； ； ； ； Where, Represents the Sigmoid activation function; 、 、 and Represent the weight items of input gate, forget gate, candidate memory, and output gate respectively; 、 、 and Specifically represent the bias items of input gate, forget gate, candidate memory, and output gate; represents the input of the current time step, which in this invention represents the number of vehicles, vehicle speed, waiting time and lane occupancy rate; represents the hidden state of the previous time step; Hyperbolic tangent activation function; Represents the state of the memory cell at the previous time step.

5. The adaptive signal control method for deformed intersections based on LSTM-GNN according to claim 1 is characterized in that: The method of establishing a graph structure between lanes based on the road topology of the intersection and using GNN to capture the spatial interaction characteristics between lanes is as follows: each lane is represented as a node, and an adjacency matrix is ​​defined according to the intersection topology to represent the edge relationship between lanes.

6. The adaptive signal control method for deformed intersections based on LSTM-GNN according to claim 1 or 5, characterized in that: The expression for using GNN to capture the spatial interaction characteristics between lanes is as follows: Where, Indicates the l The feature matrix of the layer; represents the adjacency matrix; represents the degree matrix; Indicates the l The trainable weight matrix of the layer; Represents the activation function.

7. The adaptive signal control method for deformed intersections based on LSTM-GNN according to claim 1 or 5, characterized in that: The traffic flow density is added to the LSTM model, and the lane priority weight is added when the GNN captures the spatial interaction characteristics between lanes. The LSTM model and the spatiotemporal features extracted by the GNN are combined to calculate the optimal duration of each signal phase. Based on the real-time traffic flow at the deformed intersection, the control strategy of the traffic light is dynamically adjusted. The expression is as follows: Where, represents the real-time traffic flow input features, which in the present invention include vehicle volume, vehicle speed, waiting time, and lane occupancy; It represents the traffic flow density, which is calculated by the ratio of traffic volume to lane length; Indicates the lane priority weight, which is determined by weighting according to the lane's travel direction, connected road grade, and functional factors; represents the weight of traffic flow density; Weight representing the lane priority weight.

Images