Traffic signal optimization method and system based on artificial intelligence

Abstract

The invention discloses a traffic signal optimization method and system based on artificial intelligence, and the method comprises the steps: collecting vehicle microscopic behavior data, and obtaining a driving intention prediction result; according to the driving intention prediction result, constructing a dynamic vehicle motion track; and completing the optimization of the intersection traffic signal by using the constructed dynamic vehicle motion trail. The traffic signal response speed and accuracy can be effectively improved, the traffic jam is reduced, and the traffic efficiency is improved. By adopting the convolutional neural network and the long and short term memory network, the system can automatically extract and analyze the driving behavior characteristics and mine the driving intention information, thereby optimizing the duration and phase sequence of the signal lamp and ensuring the smoothness of the traffic flow. Besides, the dynamic planning method based on the Markov process model enables the signal optimization decision to be more scientific and reasonable, can adapt to the change of traffic flow in real time, and further improves the intelligent level of traffic management.

Description

Technical Field

[0001] The present invention relates to the field of traffic signal optimization, and in particular to a traffic signal optimization method and system based on artificial intelligence. Background Art

[0002] With the acceleration of urbanization, urban traffic congestion is becoming increasingly serious, especially during peak hours. The optimization and scheduling of traffic lights has become the key to alleviating traffic pressure. Traditional traffic signal control methods often rely on fixed schedules or simple induction control. These methods cannot flexibly respond to complex traffic flow changes, resulting in low efficiency of traffic light control and inability to fully utilize road resources. In addition, due to the lack of in-depth understanding and prediction of driving behavior, it is difficult for existing systems to achieve dynamic optimization of traffic lights. Therefore, it is particularly important to develop a method that can accurately predict driving intentions and optimize traffic signals accordingly. This can not only improve road use efficiency and reduce traffic delays, but also reduce energy consumption and environmental pollution, which is of great significance to improving the level of urban traffic management. It is in this context that the present invention proposes a traffic signal optimization method and system based on artificial intelligence, which aims to achieve intelligent control of traffic lights through advanced data processing and machine learning technologies to cope with the growing demand for traffic management. Summary of the invention

[0003] In order to solve the technical problems in the above background, the present invention dynamically constructs the vehicle motion trajectory according to the driving intention prediction results, and adopts a reinforcement learning algorithm to model the vehicle's decision-making behavior under different road conditions, and then determines the vehicle's actual turning tendency at the intersection for optimal scheduling of traffic lights.

[0004] To achieve the above object, the present invention provides a traffic signal optimization method based on artificial intelligence, the steps comprising:

[0005] Collect vehicle micro-behavior data to obtain driving intention prediction results;

[0006] constructing a dynamic vehicle motion trajectory according to the driving intention prediction result;

[0007] The constructed dynamic vehicle motion trajectory is used to optimize the traffic signals at the intersection.

[0008] Preferably, the vehicle behavior micro data includes: vehicle speed, acceleration and steering angle; and time series data is formed based on the vehicle behavior micro data.

[0009] Preferably, the method for obtaining the driving intention prediction result includes: inputting the micro-behavior data into a pre-constructed convolutional neural network model to extract key features characterizing different driving behavior patterns; inputting the extracted key features into a pre-constructed long short-term memory network model, selectively memorizing and forgetting the features through a gating unit, mining the driving intention information contained in the time series data, and obtaining the driving intention prediction result.

[0010] Preferably, based on the dynamic vehicle motion trajectory, the congestion situation at the intersection is predicted; when the prediction result is congestion, a dynamic programming model is constructed based on the Markov process model according to the traffic flow data to complete the optimization of the traffic signal at the intersection.

[0011] The present invention also provides a traffic signal optimization system based on artificial intelligence, the system is used in the above method, including: a collection module, a construction module and an optimization module;

[0012] The acquisition module is used to collect vehicle micro-behavior data and obtain driving intention prediction results;

[0013] The construction module is used to construct a dynamic vehicle motion trajectory according to the driving intention prediction result;

[0014] The prediction module is used to optimize the traffic signal at the intersection by utilizing the constructed dynamic vehicle motion trajectory.

[0015] Preferably, the vehicle behavior micro data includes: vehicle speed, acceleration and steering angle; and time series data is formed based on the vehicle behavior micro data.

[0016] Preferably, the workflow of the acquisition module includes: inputting the micro-behavior data into a pre-built convolutional neural network model to extract key features characterizing different driving behavior patterns; inputting the extracted key features into a pre-built long short-term memory network model, selectively memorizing and forgetting the features through a gating unit, mining the driving intention information contained in the time series data, and obtaining the driving intention prediction result.

[0017] Preferably, the workflow of the optimization module includes: predicting intersection congestion based on the dynamic vehicle motion trajectory; when the prediction result is congestion, constructing a dynamic programming model based on the Markov process model according to the traffic flow data to complete the optimization of intersection traffic signals.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The present invention can effectively improve the response speed and accuracy of traffic signals, reduce traffic congestion, and improve traffic efficiency. By adopting convolutional neural networks and long short-term memory networks, the system can automatically extract and analyze driving behavior characteristics, mine driving intention information, and then optimize the duration and phase sequence of traffic lights to ensure smooth traffic flow. In addition, the dynamic programming method based on the Markov process model makes signal optimization decisions more scientific and reasonable, and can adapt to changes in traffic flow in real time, further improving the level of intelligent traffic management. Overall, the present invention has significant practical value and broad application prospects in the field of traffic signal control. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 The figure is a schematic diagram of a method flow of an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Embodiment 1

[0025] like Figure 1 FIG. 1 is a schematic diagram of the method flow of this embodiment, and the steps include:

[0026] S1. Collect vehicle micro-behavior data and obtain driving intention prediction results.

[0027] Collect the vehicle's micro-behavior data, including vehicle speed, acceleration, steering angle and other information, to form time series data; input the acquired micro-behavior data into a pre-built convolutional neural network model, and extract key features that characterize different driving behavior patterns through operations such as convolution and pooling; fuse them with the extracted key features; input the fused features into a pre-built long and short-term memory network model, and selectively memorize and forget the features through the gating unit to mine the driving intention information contained in the time series data; judge the driving intention of the target vehicle in the future based on the output of the long and short-term memory network. If the output value is greater than the preset threshold, the intention is judged to be established, otherwise it is judged to be not established.

[0028] Specifically, the micro-behavior data of the target vehicle is obtained and formed into time series data. Through the sensors on the vehicle, such as speed sensors, acceleration sensors, steering wheel angle sensors, etc., data is collected every 1 second. After continuous collection for 10 seconds, a series of data containing 100 time points is formed. These data can reflect the changes in the vehicle's motion state in a short period of time. The reason for forming time series data is that driving behavior is a continuous process. Only by considering the changes in the time dimension can the driving intention be understood more accurately. The obtained micro-behavior data is input into the pre-built convolutional neural network (CNN) model. CNN can extract key features that characterize different driving behavior patterns through operations such as convolution and pooling. Vehicle behaviors such as rapid acceleration, rapid deceleration, and sharp turns will generate large activation values ​​on certain specific convolution kernels of CNN. Assuming that there is a convolution kernel that specifically recognizes the behavior of sharp turns, when the steering angle data of the vehicle changes drastically in a short period of time, the output value of this convolution kernel will be very high. In this way, CNN can extract features representing sharp turns from the original micro-behavior data. The purpose of using CNN is to automatically learn and extract features to avoid the tediousness and inefficiency of manual design.

[0029] S2. Constructing a dynamic vehicle motion trajectory based on the driving intention prediction result.

[0030] A reinforcement learning algorithm is used to model the decision-making behavior of vehicles under different road conditions, and then judge the actual turning tendency of the vehicle at the intersection for the optimal scheduling of traffic lights.

[0031] Before modeling, the collected data must first be cleaned to remove noise and outliers caused by sensor errors or abnormal operations; in this way, the accuracy and reliability of subsequent analysis can be ensured.

[0032] Next, a long short-term memory neural network (LSTM) is used to construct a driving intention prediction model. LSTM is particularly suitable for processing and predicting important events in time series data because it can learn long-term dependent information. LSTM selectively memorizes and forgets features through gated units, and can mine driving intention information contained in time series data. For example, the continuous acceleration, lane merging, and approaching the roadside of a vehicle may indicate the driver's intention to overtake. LSTM can remember these continuous behaviors and reflect them in the final output. LSTM is used because it can effectively process time series data and capture temporal dependencies. According to the output of LSTM, the driving intention of the target vehicle in the future is judged. In this embodiment, if the output value of LSTM is greater than the preset threshold value 8, it is judged that the driver has the intention to overtake; otherwise, it is judged that there is no intention to overtake, and the selection of the threshold needs to be adjusted according to the actual situation.

[0033] S3. Use the constructed dynamic vehicle motion trajectory to optimize the traffic signal at the intersection.

[0034] When the dynamic vehicle motion trajectory predicts that a certain intersection may experience traffic congestion, the signal light duration and phase sequence are dynamically optimized through the Markov decision process according to the distribution of vehicle driving intentions at each entrance to the congested section, extending the green light time in the diversion direction and shortening the red light time in the congested direction, thereby guiding traffic evacuation and alleviating backlogs.

[0035] According to the traffic flow data, a Markov process model is established, and each entrance is regarded as a state. For each entrance, real-time traffic data such as vehicle flow and queue length are obtained. According to the real-time traffic data, the current Markov state of each entrance is determined. The signal light duration and phase sequence are used as decision variables to build a dynamic programming model. In the dynamic programming model, the state transition probability matrix and the immediate reward function are defined. The value iteration or policy iteration algorithm is used to solve the dynamic programming model and obtain the optimal control strategy. The optimal control strategy is applied to the signal light control, and the phase duration and phase sequence of each phase are dynamically adjusted to optimize the traffic flow.

[0036] Specifically, a Markov process model is established based on traffic flow data, and each entrance lane is used as a state. Assume that there are four entrance lanes at a city intersection, marked as A, B, C, and D. Real-time traffic data such as vehicle flow and queue length of each entrance lane are collected through geomagnetic sensors, cameras and other equipment. Assume that the vehicle flow of entrance lane A during peak hours is 500 vehicles per hour and the queue length is 50 meters; the vehicle flow of entrance lane B is 300 vehicles per hour and the queue length is 30 meters; the data of entrance lanes C and D are 400 vehicles per hour and 40 meters, 200 vehicles per hour and 20 meters, respectively. According to real-time traffic data, the Markov state of each entrance lane is determined. Assume that the vehicle flow and queue length are divided into three levels: low, medium, and high. Entrance lane A is in a high-flow high-queue state, B is in a medium-flow medium-queue state, C is in a high-flow medium-queue state, and D is in a low-flow low-queue state. These states correspond to different state nodes in the Markov process. The signal light duration and phase sequence are used as decision variables to construct a dynamic programming model. In the model, the state transition probability matrix and the immediate reward function are defined. The state transition probability matrix describes the probability of transitioning from one state to another. For example, when the entrance channel A is in a high-flow and high-queue state, after one signal cycle, there is a 70% probability that it is still in this state, a 20% probability that it will transition to a medium-flow and medium-queue state, and a 10% probability that it will transition to a low-flow and low-queue state. The immediate reward function reflects the optimization goal in each state, such as reducing the queue length and improving traffic efficiency. Assume that in a high-flow and high-queue state, the reward value is 10 for every 1-meter reduction in queue length; in a medium-flow and medium-queue state, the reward value is 5 for every 1-meter reduction in queue length. Use value iteration or policy iteration algorithm to solve the dynamic programming model and obtain the optimal control strategy. The value iteration algorithm continuously updates the state value function until it converges to the optimal value.

[0037] Specifically, in the initial state, the value function of each state is 0. Through multiple iterations, the optimal value function of entrance lane A under high traffic and high queue conditions may converge to 100, indicating the maximum cumulative reward that can be obtained by adopting the optimal strategy under this state. The strategy iteration algorithm continuously improves the strategy until the optimal strategy is found. Assuming that the initial strategy is to have equal green light time for each entrance lane, through iteration, it is found that extending the green light time of entrance lane A and shortening the green light time of entrance lane D can significantly reduce the queue length of A and improve the overall traffic efficiency. The optimal control strategy is applied to signal light control, dynamically adjusting the duration and sequence of each phase to optimize traffic flow.

[0038] Embodiment 2

[0039] This embodiment also provides an artificial intelligence-based traffic signal optimization system, including: an acquisition module, a construction module and an optimization module; the acquisition module is used to collect vehicle micro-behavior data and obtain driving intention prediction results; the construction module is used to construct dynamic vehicle motion trajectories according to the driving intention prediction results; the prediction module is used to use the constructed dynamic vehicle motion trajectories to complete the optimization of intersection traffic signals.

[0040] The following will describe in detail how the present invention solves the technical problems in practical work in conjunction with this embodiment.

[0041] First, the acquisition module is used to collect the vehicle's micro-behavior data, including vehicle speed, acceleration, steering angle and other information, to form time series data; the acquired micro-behavior data is input into a pre-built convolutional neural network model, and the key features representing different driving behavior patterns are extracted through operations such as convolution and pooling; the key features are fused with the extracted key features; the fused features are input into a pre-built long and short-term memory network model, and the features are selectively memorized and forgotten through the gating unit to mine the driving intention information contained in the time series data; according to the output of the long and short-term memory network, the driving intention of the target vehicle in the future period is judged. If the output value is greater than the preset threshold, the intention is judged to be established, otherwise it is judged to be not established.

[0042] Specifically, the micro-behavior data of the target vehicle is obtained and formed into time series data. Through the sensors on the vehicle, such as speed sensors, acceleration sensors, steering wheel angle sensors, etc., data is collected every 1 second. After continuous collection for 10 seconds, a series of data containing 100 time points is formed. These data can reflect the changes in the vehicle's motion state in a short period of time. The reason for forming time series data is that driving behavior is a continuous process. Only by considering the changes in the time dimension can the driving intention be understood more accurately. The obtained micro-behavior data is input into the pre-built convolutional neural network (CNN) model. CNN can extract key features that characterize different driving behavior patterns through operations such as convolution and pooling. Vehicle behaviors such as rapid acceleration, rapid deceleration, and sharp turns will generate large activation values ​​on certain specific convolution kernels of CNN. Assuming that there is a convolution kernel that specifically recognizes the behavior of sharp turns, when the steering angle data of the vehicle changes drastically in a short period of time, the output value of this convolution kernel will be very high. In this way, CNN can extract features representing sharp turns from the original micro-behavior data. The purpose of using CNN is to automatically learn and extract features to avoid the tediousness and inefficiency of manual design.

[0043] The construction module then constructs a dynamic vehicle motion trajectory based on the driving intention prediction result.

[0044] A reinforcement learning algorithm is used to model the decision-making behavior of vehicles under different road conditions, and then judge the actual turning tendency of the vehicle at the intersection for the optimal scheduling of traffic lights.

[0045] Before modeling, the collected data must first be cleaned to remove noise and outliers caused by sensor errors or abnormal operations; in this way, the accuracy and reliability of subsequent analysis can be ensured.

[0046] Next, a long short-term memory neural network (LSTM) is used to construct a driving intention prediction model. LSTM is particularly suitable for processing and predicting important events in time series data because it can learn long-term dependent information. LSTM selectively memorizes and forgets features through gated units, and can mine driving intention information contained in time series data. For example, the continuous acceleration, lane merging, and approaching the roadside of a vehicle may indicate the driver's intention to overtake. LSTM can remember these continuous behaviors and reflect them in the final output. LSTM is used because it can effectively process time series data and capture temporal dependencies. According to the output of LSTM, the driving intention of the target vehicle in the future is judged. In this embodiment, if the output value of LSTM is greater than the preset threshold value 8, it is judged that the driver has the intention to overtake; otherwise, it is judged that there is no intention to overtake, and the selection of the threshold needs to be adjusted according to the actual situation.

[0047] Finally, the optimization module uses the constructed dynamic vehicle motion trajectory to optimize the traffic signals at the intersection.

[0048] When the dynamic vehicle motion trajectory predicts that a certain intersection may experience traffic congestion, the signal light duration and phase sequence are dynamically optimized through the Markov decision process according to the distribution of vehicle driving intentions at each entrance to the congested section, extending the green light time in the diversion direction and shortening the red light time in the congested direction, thereby guiding traffic evacuation and alleviating backlogs.

[0049] According to the traffic flow data, a Markov process model is established, and each entrance is regarded as a state. For each entrance, real-time traffic data such as vehicle flow and queue length are obtained. According to the real-time traffic data, the current Markov state of each entrance is determined. The signal light duration and phase sequence are used as decision variables to build a dynamic programming model. In the dynamic programming model, the state transition probability matrix and the immediate reward function are defined. The value iteration or policy iteration algorithm is used to solve the dynamic programming model and obtain the optimal control strategy. The optimal control strategy is applied to the signal light control, and the phase duration and phase sequence of each phase are dynamically adjusted to optimize the traffic flow.

[0050] Specifically, a Markov process model is established based on traffic flow data, and each entrance lane is used as a state. Assume that there are four entrance lanes at a city intersection, marked as A, B, C, and D. Real-time traffic data such as vehicle flow and queue length of each entrance lane are collected through geomagnetic sensors, cameras and other equipment. Assume that the vehicle flow of entrance lane A during peak hours is 500 vehicles per hour and the queue length is 50 meters; the vehicle flow of entrance lane B is 300 vehicles per hour and the queue length is 30 meters; the data of entrance lanes C and D are 400 vehicles per hour and 40 meters, 200 vehicles per hour and 20 meters, respectively. According to real-time traffic data, the Markov state of each entrance lane is determined. Assume that the vehicle flow and queue length are divided into three levels: low, medium, and high. Entrance lane A is in a high-flow high-queue state, B is in a medium-flow medium-queue state, C is in a high-flow medium-queue state, and D is in a low-flow low-queue state. These states correspond to different state nodes in the Markov process. The signal light duration and phase sequence are used as decision variables to construct a dynamic programming model. In the model, the state transition probability matrix and the immediate reward function are defined. The state transition probability matrix describes the probability of transitioning from one state to another. For example, when the entrance channel A is in a high-flow and high-queue state, after one signal cycle, there is a 70% probability that it is still in this state, a 20% probability that it will transition to a medium-flow and medium-queue state, and a 10% probability that it will transition to a low-flow and low-queue state. The immediate reward function reflects the optimization goal in each state, such as reducing the queue length and improving traffic efficiency. Assume that in a high-flow and high-queue state, the reward value is 10 for every 1-meter reduction in queue length; in a medium-flow and medium-queue state, the reward value is 5 for every 1-meter reduction in queue length. Use value iteration or policy iteration algorithm to solve the dynamic programming model and obtain the optimal control strategy. The value iteration algorithm continuously updates the state value function until it converges to the optimal value.

[0051] Specifically, in the initial state, the value function of each state is 0. Through multiple iterations, the optimal value function of entrance lane A under high traffic and high queue conditions may converge to 100, indicating the maximum cumulative reward that can be obtained by adopting the optimal strategy under this state. The strategy iteration algorithm continuously improves the strategy until the optimal strategy is found. Assuming that the initial strategy is to have equal green light time for each entrance lane, through iteration, it is found that extending the green light time of entrance lane A and shortening the green light time of entrance lane D can significantly reduce the queue length of A and improve the overall traffic efficiency. The optimal control strategy is applied to signal light control, dynamically adjusting the duration and sequence of each phase to optimize traffic flow.

[0052] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A traffic signal optimization method based on artificial intelligence, characterized in that the steps include: Collect vehicle micro-behavior data to obtain driving intention prediction results; constructing a dynamic vehicle motion trajectory according to the driving intention prediction result; The constructed dynamic vehicle motion trajectory is used to optimize the traffic signals at the intersection.

2. The traffic signal optimization method based on artificial intelligence according to claim 1 is characterized in that: The vehicle behavior micro data includes: vehicle speed, acceleration and steering angle; time series data is formed based on the vehicle behavior micro data.

3. The traffic signal optimization method based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the driving intention prediction result includes: inputting the micro-behavior data into a pre-built convolutional neural network model to extract key features characterizing different driving behavior patterns; inputting the extracted key features into a pre-built long short-term memory network model, selectively memorizing and forgetting the features through a gating unit, mining the driving intention information contained in the time series data, and obtaining the driving intention prediction result.

4. The traffic signal optimization method based on artificial intelligence according to claim 1 is characterized in that: Based on the dynamic vehicle motion trajectory, the congestion situation at the intersection is predicted; when the prediction result is congestion, a dynamic programming model is constructed based on the Markov process model according to the traffic flow data to complete the optimization of the traffic signal at the intersection.

5. A traffic signal optimization system based on artificial intelligence, the system is used to implement the method described in any one of claims 1 to 4, characterized in that: include: Acquisition module, construction module and optimization module; The acquisition module is used to collect vehicle micro-behavior data and obtain driving intention prediction results; The construction module is used to construct a dynamic vehicle motion trajectory according to the driving intention prediction result; The prediction module is used to optimize the traffic signal at the intersection by utilizing the constructed dynamic vehicle motion trajectory.

6. The traffic signal optimization system based on artificial intelligence according to claim 5 is characterized in that: The vehicle behavior micro data includes: vehicle speed, acceleration and steering angle; time series data is formed based on the vehicle behavior micro data.

7. The traffic signal optimization system based on artificial intelligence according to claim 5, characterized in that: The workflow of the acquisition module includes: inputting the micro-behavior data into a pre-built convolutional neural network model to extract key features that characterize different driving behavior patterns; inputting the extracted key features into a pre-built long short-term memory network model, selectively memorizing and forgetting the features through a gating unit, mining the driving intention information contained in the time series data, and obtaining the driving intention prediction result.

8. The traffic signal optimization system based on artificial intelligence according to claim 5, characterized in that: The workflow of the optimization module includes: predicting the congestion situation at the intersection based on the dynamic vehicle motion trajectory; when the prediction result is congestion, constructing a dynamic programming model based on the Markov process model according to the traffic flow data to complete the optimization of the traffic signal at the intersection.

Images