A Stability Control Method for Intelligent Transportation

Abstract

The present application provides an intelligent transportation stability control method, including: obtaining a user's travel time preference, location preference and sensitivity to traffic control measures, and constructing a user attribute model to describe the user's acceptance of different traffic control measures; generating candidate combination schemes of control measures according to the user attribute model and the traffic prediction model, screening the candidate combination schemes, and obtaining an optimal control combination strategy; converting the optimal control combination strategy into a traffic control instruction to dynamically regulate and guide the traffic flow; at the same time, based on the user's feedback on each control measure, calculating the matching degree between the user behavior and the traffic control, and identifying the user behavior as a risky behavior if the matching degree is lower than a preset threshold; analyzing the effects and risks of historical control measures, extracting the correlation between the control measures and the user behavior and traffic flow, and establishing an optimization iteration unit for the control measures.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an intelligent transportation stability control method. Background Art

[0002] The implementation of traffic control measures faces the difficult problem of balancing the travel experience of multiple types of users and the stability of the traffic system. The acceptance and feedback of different types of users on control measures vary greatly. Private car owners, bus passengers and truck drivers have different concerns about travel efficiency, operating time adjustments and traffic restrictions and speed limits. The impact of emergencies such as traffic accidents and bad weather on factors such as traffic flow and road conditions changes rapidly, further increasing the uncertainty of traffic control. In a dynamic and changing traffic environment, the system needs to perceive the travel needs of various types of users and the real-time status of the traffic system in real time, and perform real-time calculation and optimization based on massive heterogeneous data to seek the optimal control combination. The dynamic changes in user needs and traffic system status, as well as the uncertainty of emergencies, make the optimization of the control combination face complex constraints and optimization goals. The execution of the optimization results needs to be closely integrated with the actual traffic control equipment and adapt to the differences in control strategies in different regions and times, which further increases the difficulty of system design and implementation. Summary of the invention

[0003] The present invention provides an intelligent transportation stability control method, which mainly includes:

[0004] Obtain users’ travel time preferences, location preferences, and sensitivity to traffic control measures, and build a user attribute model to describe users’ acceptance of different traffic control measures;

[0005] Obtain real-time traffic flow data and road condition data of the road network, combine historical traffic data and historical abnormal road conditions, train a traffic prediction model, and determine the dynamic change trend of traffic flow data and road condition data;

[0006] Generate candidate combinations of control measures based on the user attribute model and traffic prediction model, screen the candidate combinations, and obtain the optimal control combination strategy;

[0007] Convert the optimal control combination strategy into traffic control instructions to dynamically regulate and guide traffic flow;

[0008] In the process of dynamically regulating and guiding traffic flow, real-time monitoring of traffic flow, vehicle speed, and queue length of the road network during the execution of traffic control instructions is carried out to evaluate the actual control effect and dynamically adjust traffic control instructions based on the actual evaluation results;

[0009] At the same time, based on the user's feedback on various control measures, the matching degree between the user behavior and the traffic control is calculated. If the matching degree is lower than the preset threshold, the user behavior is identified as risky behavior;

[0010] Construct virtual simulation scenarios corresponding to traffic control, simulate traffic flow and user behavior under different control measures based on identified risk behaviors, and evaluate the effects and risks of various control measures in virtual simulation scenarios;

[0011] Analyze the effects and risks of historical control measures, extract the correlation between control measures and user behavior and traffic flow, and establish an optimized iteration unit for control measures.

[0012] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0013] The present invention discloses a method for intelligent transportation stability control, which aims to solve the challenge of balancing user experience and system stability in traffic control; by obtaining the user's travel preferences and sensitivity to traffic control measures, combined with the acquired real-time traffic status data, the optimal traffic control strategy is generated, and executed through the intelligent traffic control system to achieve refined and personalized traffic management; by integrating the user's travel preferences, sensitivity analysis of control measures, and real-time traffic status data, the optimal traffic control strategy is generated and executed to achieve refined and personalized traffic management; a user attribute model is constructed, combined with traffic prediction, road condition changes are predicted, and targeted control measures are generated, and at the same time, a multi-objective optimization algorithm is used to select the optimal strategy to ensure that the control measures both meet user needs and maintain the stability of the traffic system. The real-time monitoring and dynamic adjustment mechanism ensures the real-time and effectiveness of control, improves traffic efficiency and safety, enhances user satisfaction, and ultimately achieves traffic flow optimization, congestion reduction, and significant improvement in travel experience, opening up a new path for smart city traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flow chart of an intelligent transportation stability control method of the present invention;

[0015] Figure 2 A schematic diagram of an intelligent transportation stability control method of the present invention; DETAILED DESCRIPTION

[0016] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0017] like Figure 1 -2. In this embodiment, a method for controlling the stability of intelligent transportation may specifically include:

[0018] Step S101, obtaining the user's travel time preference, location preference and sensitivity to traffic control measures, and constructing a user attribute model to describe the user's acceptance of different traffic control measures.

[0019] Obtain the user's historical travel data, classify the user's travel time according to the historical travel data, and obtain the user's time preference type; perform spatial clustering on the destinations in the historical travel data to obtain the user's location preference type. Calculate the user's traffic control sensitivity index based on the user's compliance rate and complaint rate for different traffic control measures. Construct a user attribute model, use the time preference type, the location preference type and the traffic control sensitivity index as input variables, and obtain the user's acceptance score and feedback classification for different traffic control measures through training on historical data. For new traffic control measures, extract the corresponding acceptance score and feedback classification, combine the current traffic mode selection, travel frequency and environmental factors, determine the weight of each factor, and calculate the comprehensive acceptance index. According to the comprehensive acceptance index, combined with the road condition information sensitivity, economic cost and travel purpose classification, predict the user's specific response to the new traffic control measures.

[0020] Specifically, the user's historical travel data is obtained, and the user's travel time is classified using the K-means algorithm to obtain the user's time preference type. At the same time, the user's historical destination is spatially clustered using the DBSCAN algorithm to obtain the user's location preference type. The user's traffic control sensitivity index is calculated based on the user's compliance rate and complaint rate for different traffic control measures. The user attribute model is constructed using the C4.5 decision tree, and the user's time preference type, location preference type and traffic control sensitivity index are used as input variables. The information gain ratio is used as the feature selection criterion. Through training on historical data, the user's acceptance score and feedback classification for different traffic control measures are obtained. For new traffic control measures, the corresponding acceptance score and feedback classification are extracted from the user attribute model, combined with the current choice of transportation mode, travel frequency and environmental factors. The analytic hierarchy process is used to determine the weight of each factor, and the weighted average method is used to calculate the comprehensive acceptance index. According to the comprehensive acceptance index, combined with the sensitivity of road condition information, economic cost considerations and travel purpose classification, the Mamdani fuzzy reasoning method is used to predict the user's specific response to the new traffic control measures. An if-then rule base was established to perform three stages of fuzzification, reasoning and defuzzification, thereby providing a basis for decision-making for traffic management departments. When obtaining historical travel data of users, travel records of users in the past 6 months were collected, including departure time, destination and traffic control measures encountered. The K-means algorithm was used to cluster travel time, and the K value was set to 5, and five time preferences were obtained: morning peak, daytime, evening peak, night and early morning. Spatial clustering was performed using the DBSCAN algorithm, with the neighborhood radius set to 500 meters and the minimum number of points set to 5, to identify user preferences for places such as workplaces, shopping centers and leisure places that users often visit. The traffic control sensitivity index was obtained by calculating the compliance rate and complaint rate of users when encountering traffic control. When the compliance rate was 90% and the complaint rate was 5%, the sensitivity index was 0.85. When the C4.5 decision tree constructed the user attribute model, the information gain ratio was used as the feature selection criterion, the minimum number of leaf node samples was set to 50, and the maximum tree depth was set to 5. The input variables include time preference type, location preference type and traffic control sensitivity index, and the output is the user's acceptance score (1-10 points) and feedback classification for different traffic control measures, including positive, neutral and negative. Through training on 10,000 historical data, a user attribute model with an accuracy of 85% was obtained. In the evaluation of new traffic control measures, the acceptance score of 7 points and the feedback category of neutral were extracted from the user attribute model. Combined with the current choice of transportation mode, travel frequency and environmental factors, such as choosing a private car as the transportation mode, a travel frequency of 5 times a week and an air quality index of 120.The weights were determined using the analytic hierarchy process, with the weight of acceptance score being 0.4, the weight of feedback category being 0.3, the weight of transportation mode selection being 0.1, the weight of travel frequency being 0.1, and the weight of environmental factors being 0.1. The comprehensive acceptance index was calculated using the weighted average method to be 0.72. Finally, the Mamdani fuzzy reasoning method was used to predict the specific reactions of users to the new traffic control measures. The input variables were the comprehensive acceptance index of 0.72, the sensitivity of road condition information of 0.8, the economic cost being considered as medium, and the purpose of travel being commuting. A fuzzy rule base was established, containing 20 if-then rules. After three stages of fuzzification, reasoning, and defuzzification, it was concluded that the predicted reaction of users to the new traffic control measures was "basically accepted, but there may be slight complaints", which provided a specific decision-making reference for the traffic management department.

[0021] Step S102, obtaining real-time traffic flow data and road condition data of the road network, combining historical traffic flow data and historical abnormal road conditions, training a traffic prediction model, and determining the dynamic change trend of traffic flow data and road condition data.

[0022] The vehicle quantity information collected by the traffic monitoring camera, the driving speed data recorded by the vehicle-mounted GPS device and the lane occupancy data detected by the roadside sensor are obtained; a traffic status data set in a unified format is generated according to the vehicle quantity information, the driving speed data and the lane occupancy data; a traffic prediction model is constructed based on the traffic status data set in combination with the road network topology information and traffic signal control parameters; real-time traffic flow data and road condition data are input into the traffic prediction model, and the traffic flow change trend and road condition change trend within a preset time in the future are obtained through a sliding time window method.

[0023] Specifically, real-time traffic flow data of the road network, including the number of vehicles, driving speed and vehicle type distribution, are obtained from traffic monitoring cameras and vehicle-mounted GPS devices. At the same time, real-time road condition data, including lane occupancy, average headway and traffic accident information, are collected from roadside sensors. Data from different sources are processed using a data fusion algorithm, and a traffic status data set in a unified format is generated through steps such as denoising, standardization and time alignment. The ARIMA model is used to process historical traffic data and historical abnormal road condition records to extract periodic patterns and long-term trends. A traffic prediction model based on the LSTM network is constructed by combining the road network topology and traffic signal control parameters. The data set is divided into a training set, a validation set and a test set with a ratio of 6:2:2. The LSTM network adopts a three-layer structure, each layer contains 128 neurons, and the Adam optimizer is used for model training, with the learning rate set to 0.001. The real-time traffic flow data and road condition data are input into the trained traffic prediction model, and the traffic flow change trend and road condition change trend in the next hour are predicted by the sliding time window method. The sliding time window size is set to 1 hour, and the step size is 5 minutes. Weather factors are added as additional input features to the traffic prediction model, including temperature, precipitation and visibility. According to the predicted traffic flow change trend and road condition change trend, the average speed, congestion index and traffic capacity of the road network are calculated. The congestion index is calculated by the ratio of the actual road speed to the free flow speed. The preset threshold is determined by statistical analysis of historical data. If the congestion index exceeds 0.8, or the average speed is less than 50% of the design speed, the stability level of the intelligent traffic control system is judged to be low. By comparing the calculation results with the preset threshold, the stability level of the intelligent traffic control system is obtained, which provides a basis for traffic management decisions. In actual applications, real-time traffic flow data is obtained from 100 traffic monitoring cameras and 10,000 vehicle-mounted GPS devices every 5 minutes. At the same time, real-time road condition data is collected from 200 roadside sensors. The data fusion algorithm uses Kalman filtering to time align and denoise data from different sources. For example, during the morning rush hour from 8:00 to 9:00, the average speed of a main road dropped from 40 km / h to 20 km / h, and the traffic volume increased from 800 vehicles / hour to 1500 vehicles / hour. The ARIMA model uses historical data from the past 30 days and sets the parameters (p, d, q) to (2, 1, 2) to extract the periodic pattern and long-term growth trend of the morning and evening rush hours on weekdays. The LSTM network training uses 60,000 historical records, 20,000 for the validation set and 20,000 for the test set. During the training process, the batch size is set to 64 and the number of iterations is set to 100, and finally a prediction accuracy of 90% is achieved on the test set. In the sliding time window method, each prediction uses the data from the past hour to predict the situation in the next hour, with a step size of 5 minutes. Real-time meteorological data is used as additional input, such as temperature 20℃, precipitation 0mm, and visibility 10km.The prediction results show that in the next hour, the average speed of a certain section of road will drop from 35km / h to 25km / h, and the traffic volume will increase from 1,200 vehicles / hour to 1,600 vehicles / hour. In the calculation of the congestion index, assuming that the free flow speed of the section is 60km / h and the current actual average speed is 30km / h, the congestion index is 0.5. Through statistical analysis of the data of the past 90 days, the congestion index threshold is set to 0.7, and the average speed threshold is 60% of the design speed. According to the prediction results, the congestion index of this section will reach 0.75 after 1 hour, exceeding the preset threshold. The stability level of the intelligent traffic control system is judged to be low, and it is recommended to initiate traffic control measures.

[0024] Step S103, generating candidate combination schemes of control measures according to the user attribute model and the traffic prediction model, screening the candidate combination schemes, and obtaining the optimal control combination strategy.

[0025] According to user acceptance data, traffic flow prediction results and real-time traffic status data, candidate combination schemes of control measures are generated. The real-time traffic status data adjusts the candidate scheme generation process by affecting the decision tree splitting condition; the candidate combination schemes are encoded using a genetic algorithm, each gene represents a control measure parameter, and a fitness function is established. The fitness function adopts a weighted summation form, and the weights are pre-set according to user acceptance, travel time and location preferences, and traffic stability level; the candidate solutions are screened to obtain a set of non-dominated solutions, and the congestion of each solution is calculated. The congestion calculation is based on the normalized value of each objective function; the solution with the largest congestion is selected from the non-dominated solution set as the optimal control combination strategy, and the optimal control combination strategy is matched with the real-time traffic status data to determine whether the preset trigger condition is met. If the trigger condition is met, the control strategy information is sent to the relevant traffic management equipment.

[0026] Specifically, based on the user acceptance data in the user attribute model and the traffic flow prediction results in the traffic prediction model, combined with the real-time traffic status data, the C4.5 decision tree algorithm is used to generate candidate combinations of control measures. The real-time traffic status data adjusts the generation process of candidate solutions by affecting the splitting conditions of the decision tree. Specifically, it includes measures such as traffic restriction, diversion, and traffic light adjustment, and each measure is given different weights according to the real-time traffic conditions. The candidate combination scheme is encoded using a genetic algorithm, and each gene represents the parameters of a control measure. The fitness function is designed, and user acceptance, travel time and location preferences, and traffic stability level are used as optimization objectives. The fitness function adopts the form of weighted summation, and the weights are pre-set according to the importance of each objective. The offspring is generated by a single-point crossover operation, and the crossover probability is set to 0.8. The uniform mutation operation is used, and the mutation probability is set to 0.1 to maintain population diversity. The generated candidate solutions are screened using a fast non-dominated sorting algorithm to obtain a non-dominated solution set. The congestion of each solution is calculated, and the congestion calculation is based on the normalized value of each objective function. The solution with the largest congestion is selected from the non-dominated solution set as the optimal control combination strategy. The optimal strategy contains specific control measures parameters, such as the scope of restricted areas, diversion path selection, and signal light cycle adjustment value. The optimal control combination strategy is matched with real-time traffic status data, and the control measures are determined according to the preset trigger conditions. The trigger conditions include thresholds such as traffic flow exceeding 80% of the road capacity and average vehicle speed being lower than 60% of the design speed. If the trigger conditions are met, the control strategy information is sent to the relevant traffic management equipment through the traffic management communication network. The receiving device automatically adjusts the restricted signs, variable information board display content, and traffic signal timing scheme according to the received strategy information. In practical applications, user acceptance data is first extracted from the user attribute model, such as the acceptance of restriction measures is 0.7 and the acceptance of diversion measures is 0.8. The traffic prediction model predicts that the traffic flow on a main road will increase from 2,000 vehicles / hour to 3,000 vehicles / hour in the next 2 hours. Real-time traffic status data shows that the current average speed is 30km / h. The C4.5 decision tree algorithm uses these data as input to generate candidate combination schemes. For example, when the traffic volume exceeds 2,500 vehicles per hour, it is recommended to implement diversion measures; when the average speed is less than 25 km / h, it is recommended to adjust the timing of the signal lights. The genetic algorithm encodes these solutions, and each gene represents the parameters of a measure, such as the diversion ratio or the green light time of the signal light. In the fitness function, the user acceptance weight is 0.4, the travel time and location preference weight is 0.3, and the traffic stability level weight is 0.3. After 100 generations of evolution, the population size is 200, and a set of non-dominated solutions is finally obtained. The fast non-dominated sorting algorithm sorts this set of solutions and calculates the congestion. The solution with the largest congestion is selected as the optimal strategy, such as implementing 30% vehicle diversion on the main road and extending the green light time of the signal light on the parallel road by 10 seconds.The intelligent traffic control system continuously monitors the real-time traffic status and triggers the preset conditions when it detects that the traffic flow on the main road reaches 2,800 vehicles per hour. It then sends instructions to related equipment through the traffic management communication network, automatically adjusts the variable information board to display diversion information, and updates the timing plan of traffic lights along the route.

[0027] Step S104, converting the optimal control combination strategy into traffic control instructions to dynamically regulate and guide traffic flow.

[0028] Generate a standardized traffic control instruction set based on the optimal control combination strategy, the instruction set includes signal light timing adjustment, lane allocation, and speed limit setting parameters; if the strategy includes speed limit measures, generate speed limit setting instructions. Verify the traffic control instructions, the verification includes generating polynomial selection, data grouping, division operation, and remainder comparison; transmit the verified instructions to the intelligent traffic control system through a secure encrypted channel. After receiving the traffic control instructions, obtain the current road condition parameters, the road condition parameters include traffic flow and average speed; perform real-time control of traffic lights, the control includes fuzzification, reasoning, and defuzzification. After receiving the traffic control information, the on-board terminal obtains the current location and destination information of the vehicle and calculates the optimal path.

[0029] Specifically, the optimal control combination strategy is converted into a standardized traffic control instruction set. The rule engine defines the conversion rules, such as "if the strategy includes speed limit measures, then generate speed limit value setting instructions." The converted instruction set includes specific parameters such as signal light timing adjustment, lane allocation, and speed limit value setting. The instructions are prioritized according to the execution timing to form an ordered instruction sequence. The traffic control instructions are transmitted to the intelligent traffic control system through a secure encrypted channel. The cyclic redundancy check (CRC) algorithm is used to ensure the integrity and accuracy of instruction transmission. The CRC check process includes generating polynomial selection, data grouping, division operation, and remainder comparison. At the same time, an instruction execution queue is established to sort and manage instructions according to priority and timestamp. After receiving the instructions, the intelligent traffic control system uses a fuzzy control algorithm to control the traffic lights in real time. The fuzzy control algorithm includes three steps: fuzzification, reasoning, and defuzzification. The input variables are current road condition parameters, such as traffic volume and average speed, and the output variables are green light duration and phase difference. The fuzzy rule base is set according to traffic engineering experience, such as "if the traffic volume is large and the average speed is low, extend the green light time." The traffic light control parameters are obtained through fuzzy reasoning to achieve the optimal allocation of traffic flow. Traffic control information is pushed to the vehicle terminal and roadside information board based on the vehicle-road collaboration method. The DSRC (dedicated short-range communication) protocol is used for data transmission to ensure low latency and high reliability. After receiving the information, the vehicle terminal uses the navigation algorithm to calculate the optimal path in combination with the current location and destination of the vehicle. At the same time, the vehicle driving data, including speed, acceleration and fuel consumption, are collected through the OBD interface. These data are transmitted back to the traffic control center through the wireless network to evaluate the control effect and optimize the subsequent strategy. In actual application, the rule engine first receives the optimal control combination strategy, such as "the speed limit on the main road is 60km / h, and the lane allocation of the secondary road is adjusted to 2 in and 3 out". 100 conversion rules are preset in the rule engine, and the strategy is converted into specific instructions through rule matching. For example, the speed limit strategy is converted into the instruction of "set electronic speed limit sign, value 60". The converted instruction set is sorted by execution timing, with a priority from 1 to 10 and a time interval of 30 seconds. The secure encryption channel uses the AES-256 encryption algorithm with a key length of 256 bits. The CRC check uses CRC-32 polynomial, and the data packet size is 1KB. The instruction execution queue uses the priority queue data structure, and the queue length is capped at 1000 instructions. In the fuzzy control algorithm, the fuzzy set of the input variable "traffic flow" is {small, medium, large}, and the fuzzy set of "average speed" is {slow, medium, fast}. The fuzzy set of the output variable "green light duration" is {short, medium, long}, and the fuzzy set of "phase difference" is {small, medium, large}. The fuzzy rule base contains 25 rules. For example, when the traffic flow is "large" and the average speed is "slow", the green light duration is adjusted to "long" and the phase difference is adjusted to "large".Defuzzification uses the centroid method, and the final output green light duration is dynamically adjusted between 30-120 seconds, and the phase difference varies between 0-30 seconds. In vehicle-road collaboration, the DSRC communication frequency band is 5.9GHz, the communication range is 200 meters, and the data transmission rate is 6Mbps. The on-board terminal receives traffic control information once a second, including road conditions and control measures within a surrounding range of 500 meters. The navigation algorithm uses an improved A* algorithm, taking into account real-time road conditions and control information, and updates the optimal path every 5 seconds. The OBD interface collects vehicle data every 1 second, including instantaneous speed, with an accuracy of 0.1km / h, acceleration, with an accuracy of 0.01m / s2, and fuel consumption rate, with an accuracy of 0.1L / 100km. These data are packaged and sent to the traffic control center every 30 seconds through the 4G network for real-time evaluation of control effects and strategy optimization.

[0030] Step S105, in the process of dynamically regulating and guiding the traffic flow, real-time monitoring of the traffic flow, vehicle speed, and queue length of the road network during the execution of the traffic control instructions, evaluating the actual control effect, and dynamically adjusting the traffic control instructions according to the actual evaluation results.

[0031] Obtain traffic flow, vehicle speed and queue length data collected by road induction coils and video surveillance equipment, and pre-process the original data; determine whether the value exceeds the preset range, and if so, remove the abnormal value; smooth the data after removing the abnormal value to obtain pre-processed data; fuse the pre-processed data; obtain the optimal state estimate through an iterative update process, and the optimal state estimate includes the average vehicle speed, traffic density and queue length; calculate the management and control effect evaluation index according to the optimal state estimate; use fuzzy comprehensive evaluation method to obtain a comprehensive score; establish a factor set U, a comment set V and a weight set W, and the weight set W is determined by hierarchical analysis method; generate an adjustment strategy according to the comprehensive score and the preset threshold, obtain the rules from the rule library, and use forward chain reasoning to obtain the adjustment strategy; generate a control instruction according to the adjustment strategy, and send the control instruction to the traffic control device for execution.

[0032] Specifically, the traffic flow, vehicle speed and queue length data are collected in real time through road induction coils and video surveillance equipment, and the sliding time window method is used to preprocess the raw data. The sliding time window size is set to 5 minutes, and the sliding step is 1 minute. The criterion for removing outliers is that the value exceeds the range of 3 standard deviations of the average value of the previous 10 time windows. The data after removing outliers is smoothed by moving average. The preprocessed data is fused using the Kalman filter algorithm to generate the traffic state estimation value of each road section. The state equation defines the state variables as the average vehicle speed, traffic density and queue length, and the observation equation defines the observation variables as the measured data of the induction coil and video surveillance equipment. Through the iterative update process, the optimal state estimation value is obtained. These estimates are used for subsequent management and control effect evaluation. The management and control effect evaluation indicators are calculated based on the traffic state estimation value, including the travel time index (Travel Time Index, TTI), queue length ratio (Queue Length Ratio, QLR), and level of service (Level of Service, LOS). The travel time index is calculated as the ratio of actual travel time to free flow time. TTI>1 indicates congestion, and the larger the value, the more severe the congestion. The queue length ratio is the ratio of actual queue length to road length. The service level is divided into six levels from A to F according to the Highway Capacity Manual (HCM) standard, where A represents the best service level and F represents the worst service level. The fuzzy comprehensive evaluation method is used to obtain the comprehensive score, and the factor set U = {u1, u2, u3} is established, corresponding to the three evaluation indicators, and the comment set V = {v1, v2, v3, v4, v5}

[0033] Corresponding to the five evaluation levels, the weight set W is determined by the hierarchical analysis method. Based on the comprehensive score and the preset threshold, the rule reasoning engine is used to dynamically generate the adjustment strategy. The preset threshold is determined by statistical analysis of historical data. For example, the adjustment is triggered when the comprehensive score is lower than 0.6. The rule base contains rules such as "If the comprehensive score is lower than the threshold and the traffic volume is greater than 80% of the capacity, the green light time is extended." The reasoning process adopts the forward chain reasoning method. Starting from the known facts, the adjustment strategy is obtained through rule matching and execution. According to the reasoning results, new control instructions are generated, such as adjusting signal timing, updating guidance information or modifying the restricted range, and the instructions are sent to the traffic control equipment for execution. In actual applications, 100 induction coils and 50 video surveillance devices are arranged on the main roads of a city, and data is collected once a second. In the sliding time window method, the window size of 5 minutes contains 300 data points, and 60 data points are slid forward each time (1 minute). For example, in the 5-minute window from 7:30 to 7:35 during the morning rush hour, the average speed of a certain road section dropped from 40 km / h to 25 km / h, the traffic volume increased from 1,200 vehicles / hour to 1,800 vehicles / hour, and the queue length increased from 50 meters to 150 meters. The outlier judgment found that the instantaneous speed of 70 km / h at 7:32 was beyond the normal range and was eliminated. In the Kalman filter algorithm, the state vector x = [v, ρ, l]T, where v is the average speed, ρ is the traffic density, and l is the queue length. The observation vector z = [vm, qm, lm]T contains the measured speed, flow, and queue length. The instantaneous speed measurement vm is the average speed measurement of the vehicle at a certain moment, in kilometers per hour (km / h). The speed measurement values ​​vary at different time points, and outliers, such as 70km / h at 7:32, are removed to improve data quality; the traffic density measurement qm refers to the number of vehicles passing a fixed point on the road per unit time, usually in vehicles / hour (veh / hr). The measured value of traffic density increased from 1200 vehicles / hour to 1800 vehicles / hour, indicating a significant increase in traffic volume. The queue length measurement lm refers to the length of the queue of vehicles waiting to pass through a certain area, such as an intersection or bottleneck, usually in meters (m). The measured value of the queue length increased from 50 meters to 150 meters, indicating that the congestion has intensified. Through iterative calculation, the optimal estimate at 7:35 was obtained, with an average speed of 26km / h, a traffic density of 90 vehicles / km, and a queue length of 145 meters. The evaluation index calculation results show that TTI is 1.8, QLR is 0.29, and LOS is E-level. In the fuzzy comprehensive evaluation, the weight vector W = [0.4, 0.3, 0.3], and the comprehensive score is 0.55. The rule inference engine detects that the comprehensive score is lower than the preset threshold of 0.6, triggering the adjustment strategy. The matching rule is "If the comprehensive score is less than 0.6 and the traffic density is greater than 80 vehicles / km, then increase the green light time of adjacent parallel paths by 20 seconds."Generate new control instructions and send them to the signal controllers at relevant intersections through the traffic control network to achieve dynamic adjustment.

[0034] Step S106, based on the user's feedback on each control measure, the matching degree between the user behavior and the traffic control is calculated. If the matching degree is lower than a preset threshold, the user behavior is identified as a risky behavior.

[0035] Acquire traffic behavior data carrying a user identification number; send a feature extraction instruction based on the traffic behavior data, the feature extraction instruction is used to extract a keyword vector and a sentiment score from user feedback; receive a matching degree calculation request with a user identification number and a feature vector, the feature vector is constructed by combining the keyword vector, the sentiment score and the user behavior data; determine whether the feature vector matches the traffic control measures, including: using a support vector machine algorithm to calculate the matching degree score; if the matching degree score is lower than a preset threshold, it is determined to be a risky behavior.

[0036] Specifically, the user's behavior data during the traffic control process is collected once a second through the vehicle terminal and mobile application, including the speed accurate to 0.1km / h, the driving path with GPS positioning accuracy within 5 meters, and the acceleration change accurate to 0.01m / s2. At the same time, the corresponding control measures are recorded, which are divided into four categories: speed limit, diversion, signal light adjustment, and lane control, and the implementation time is recorded accurately to seconds. SSL encryption is used to transmit data and stored in a distributed database. The Word2Vec word vector model is used to extract keywords from the user feedback text, and the sentiment analysis algorithm based on LSTM is used to obtain the sentiment tendency score. The keyword vector, sentiment score and user behavior data are combined to construct a 30-dimensional feature vector. The features include average speed, maximum acceleration, path deviation, feedback text length, and the proportion of positive sentiment words. Each feature is normalized to obtain standardized user feedback data. The support vector machine algorithm is used to calculate the matching degree between user behavior and traffic control. The radial basis function (RBF) is selected as the kernel function, and the optimal parameters C=10 and gamma=0.1 are determined by the grid search method. 80% of the historical data is used for training, and the remaining 20% ​​is used for verification. The user behavior feature vector and the control measure feature vector are used as input, and a matching score between 0 and 1 is output. The preset threshold is determined to be 0.7 through ROC curve analysis. If it is lower than this value, it is considered a risky behavior. For the identified risky behavior, personalized guidance optimization measures are generated based on the C4.5 decision tree algorithm. The decision tree uses the user's historical behavior pattern and current traffic conditions as features, uses the information gain ratio as the division criterion, and sets the minimum number of leaf node samples to 50, which prompts the decision tree algorithm to generate a wider but shallower tree. The pessimistic pruning method estimates the error rate of the child node after segmentation. If the estimated error rate of the child node after segmentation is higher than the error rate when not segmented, then no segmentation will be performed to prevent overfitting. The decision result selects the most suitable guidance plan from the preset measure library, which contains 20 specific measures such as text reminders, voice broadcasts, route recommendations, and reward points. The selected guidance plan is pushed to the user in real time through the vehicle-mounted device or mobile application. In actual applications, an intelligent traffic control system deployed 10,000 vehicle terminals and a mobile application with 500,000 users. During the morning rush hour from 7:30 to 9:30, data was collected every second, and it was recorded that the speed of user A dropped from 40km / h to 20km / h. The GPS trajectory showed that it deviated from the recommended route by 500 meters, and the maximum deceleration reached -2.5m / s2. At the same time, the intelligent traffic control system recorded that a speed limit of 50km / h was implemented on this section of road. User A sent feedback through the application as "traffic congestion, optimization is recommended." The Word2Vec model converts words such as "traffic conditions", "congestion", and "optimization" into 300-dimensional word vectors, and the LSTM sentiment analysis algorithm outputs a sentiment score of -0.6, indicating a relatively negative sentiment.These data are combined into a 30-dimensional feature vector, where the average vehicle speed is 30km / h, the maximum deceleration is 2.5m / s2, the path deviation is 0.1, the feedback text length is 9, and the proportion of negative sentiment words is 0.67. The support vector machine algorithm is configured with parameters of C=10 and gamma=0.1, and the matching degree between user behavior and control measures is calculated to be 0.65, which is lower than the threshold of 0.7 and is determined to be a risky behavior. The C4.5 decision tree algorithm is based on the historical data of user A, such as 3 similar behaviors in the past week and the current road conditions, and its congestion index is 0.8. The combination of "voice broadcast + route recommendation" is selected from the measure library. The intelligent traffic control system immediately plays voice prompts including "the road ahead is congested, it is recommended to turn right and detour" through the on-board equipment, and pushes a new recommended route on the application, which is expected to save 15 minutes of travel time and be completed within 2 seconds after the risky behavior is detected.

[0037] Step S107, construct a virtual simulation scenario corresponding to traffic control, simulate traffic flow and user behavior under different control measures based on the identified risk behaviors, and evaluate the effects and risks of each control measure in the virtual simulation scenario.

[0038] Construct a virtual traffic network model based on GIS data including road geometry, intersection layout, and traffic facility locations. Simulate traffic flow operation and determine vehicle acceleration, deceleration, and lane change operations based on preset behavioral rules; if the vehicle position is updated, calculate new speed and acceleration data. Use a multi-agent approach to simulate user response behavior, where each agent represents a driver; adjust the parameter settings of the virtual simulation scene based on the identified risk behaviors. Calculate average travel time, congestion index, and accident risk assessment indicators to compare the effects of different control measures.

[0039] Specifically, a virtual traffic network is constructed based on geographic information system data, including road geometry, intersection layout, and traffic facility locations. Real-time traffic flow and signal timing data are obtained every 5 minutes, and the real-time data is fused with static geographic data through spatiotemporal interpolation methods to generate a simulation scene that matches the actual road conditions. The fusion process uses the Kriging interpolation algorithm to ensure the spatial continuity of the data. The cellular automaton algorithm is used to simulate traffic flow operation, and behavioral rules such as vehicle acceleration, deceleration, and lane change are set. The cell size is set to 7.5 meters, the time step is 1 second, and the update rules include the four steps of acceleration, deceleration, random slowing, and position update of the NaSch model. Random disturbances are generated based on the probability distribution obtained from historical data statistics, and random numbers that obey the normal distribution are generated using the Box-Muller transformation to simulate the operating characteristics of different types of vehicles. The vehicle position, speed, and acceleration data are output at each time step. Multi-agent technology is used to simulate the user's response behavior to control measures. Each agent represents a driver, and the attributes include driving experience, risk preference, and compliance with control. Decision rules for different types of drivers are set based on the user attribute model, including path selection, speed adjustment, and compliance with control information. Agents interact through a message passing mechanism to simulate vehicle following and avoidance behaviors. The decision-making process is updated in real time in combination with the output data of traffic flow simulation. Multiple simulation experiments are carried out using the Monte Carlo method, and the number of iterations is set to 1000, or the iteration is stopped when the coefficient of variation of the evaluation index is less than 0.05. Evaluation indicators such as average travel time, congestion index, and accident risk are calculated. The average travel time is obtained by averaging the time from the starting point to the end point of the vehicle; the congestion index is defined as the ratio of the actual travel time to the free flow time; the accident risk is calculated based on the time distance and spacing between vehicles. The TTC (TimeToCollision) indicator. According to the identified risk behavior type and severity, the risk factor of each risk behavior is calculated according to the preset rules, and a list of risk behavior risk factors is obtained. By analyzing the traffic flow and road attributes of each section in the virtual simulation scene, the basic accident probability of each section is obtained, and a list of basic accident probabilities of the sections is obtained. Based on the risk behavior hazard coefficient list and the road section basic accident probability list, a weighted calculation method is used to combine the two to determine the comprehensive accident probability of each road section and obtain a comprehensive accident probability list of the road section. By comparing the comprehensive accident probability of the road section with the preset safety threshold, the safety level of each road section is judged to obtain the road section safety level assessment result. According to the road section safety level assessment result, the traffic capacity and traffic flow of the corresponding road section in the virtual simulation scene are dynamically adjusted to obtain the adjusted virtual simulation scene parameter settings. The adjusted virtual simulation scene parameter settings are applied to the virtual simulation.According to the results of virtual simulation analysis, risk behaviors such as speeding, illegal lane changes, and violation of traffic lights are identified, and relevant parameters in the virtual scene are adjusted, such as increasing the probability of speeding to 20% and reducing the traffic capacity of illegal sections to 1,200 vehicles / hour. According to the "Road Traffic Accident Classification" standard issued by the Ministry of Transport, the identified risk behaviors are classified, and the hazard coefficients of various behaviors are calculated using the hierarchical analysis method to obtain a list of risk behavior hazard coefficients. Through virtual simulation analysis of traffic flow and road attributes of each section, such as the number of lanes and speed limit, combined with the logistic regression model, the basic accident probability of each section is obtained, and a list of basic accident probabilities of the sections is formed. The weighted calculation method is used to comprehensively consider the risk coefficient of risk behavior and the basic accident probability of the section, determine the comprehensive accident probability of each section, and obtain a list of comprehensive accident probabilities of the section. Set the safety threshold to 10. -4, compare the comprehensive accident probability of each section, if it is higher than the threshold, it is determined to be an accident-prone section, and the section safety level assessment result is formed. According to the assessment results, dynamically adjust the parameter settings of the corresponding section in the virtual scene, such as reducing the traffic capacity of the accident-prone section, increasing traffic flow, and re-simulating the virtual scene. The one-way analysis of variance method is used to compare the effect differences under different control measures, and the significance level is set to 0.05. By perturbing the key parameters by ±10%, sensitivity analysis is performed to identify potential risk factors. In a certain intelligent traffic control system, the virtual simulation scene covers an area of ​​100 square kilometers, including a 500-kilometer road network and 200 signal-controlled intersections. The intelligent traffic control system obtains real-time traffic flow data from 1,000 detectors every 5 minutes, and updates the signal timing scheme of 200 intersections at the same time. The Kriging interpolation algorithm uses an exponential variation function, and the correlation distance is set to 500 meters to interpolate discrete data points into a continuous traffic flow distribution. In the cellular automaton simulation, a total of 66,667 cells (500 km / 7.5 m) were divided to simulate 10,000 vehicles of different types. The maximum speed in the NaSch model was set to 5 cells / second (135 km / h), and the probability of deceleration was 0.1. The Box-Muller transformation generated normally distributed random numbers with a mean of 0 and a standard deviation of 0.5 to simulate the random fluctuations of vehicle acceleration. In the multi-agent system, each agent has 10 attributes, such as driving experience (0-50 years), risk preference (continuous value of 0-1), etc. The decision rule contains 50 if-then statements, such as "If the distance to the front vehicle is less than 10 meters and the relative speed is greater than 5 km / h, then decelerate by 2 m / s2." The Monte Carlo simulation was run 1,000 times, simulating 1 hour of traffic conditions each time. Among the evaluation indicators, the average travel time increased from 15 minutes to 25 minutes, the congestion index increased from 1.2 to 1.8, and the number of events with TTC less than 3 seconds increased from 100 to 300 per hour. The results of variance analysis showed that the speed limit measures had the most significant impact on travel time (p value < 0.01). Sensitivity analysis found that driver compliance had the greatest impact on the congestion index, and a change of ±10% resulted in a change of ±15% in the congestion index; the entire simulation process was executed in parallel on a high-performance computing cluster, and a single complete simulation took about 10 minutes.

[0040] Step S108, analyze the effects and risks of historical control measures, extract the correlation between control measures and user behavior and traffic flow, and establish an optimization iteration unit for control measures.

[0041] Acquire historical control measures data, the historical control measures data including control type, implementation time, location, duration, traffic flow change, and user feedback; use Apriori algorithm to analyze the correlation between control measures and user behavior and traffic flow in the historical control measures data according to preset minimum support and minimum confidence, and obtain strong association rules; convert the strong association rules into feature vectors as inputs of long short-term memory networks; use the long short-term memory network to predict the impact of different control measures on traffic flow and user behavior, and obtain prediction results; according to the prediction results, construct an optimization iteration unit for control measures based on genetic algorithm, and the optimization iteration unit adopts binary coding; if the optimization iteration unit generates a new control plan, write the new control plan into the optimization plan table of the traffic control knowledge base.

[0042] Specifically, a distributed database is built to store historical control measures data, including control type, implementation time, location, duration, traffic flow changes, user feedback and other information. Talend open source ETL tool is used to extract, transform and load relevant data from multi-source data systems. Regular expressions are used to clean data, remove outliers and duplicates, and the Z-Score method is used to standardize data to ensure the consistency of data from different sources. The Apriori algorithm is used to analyze the correlation between control measures and user behavior and traffic flow, setting the minimum support to 0.05 and the minimum confidence to 0.7. Strong association rules are extracted through three steps: generating candidate item sets, pruning and generating frequent item sets. The association rules are converted into feature vectors as additional inputs of the recurrent neural network to enhance the model's perception of historical association patterns. The long short-term memory network (LSTM) is used to predict the impact of different control measures on traffic flow and user behavior. The network structure includes an LSTM layer (128 units), a fully connected layer (64 neurons) and an output layer. The historical data sequence is divided into 30-minute time steps, and each sample contains data from the past 4 hours. The Adam optimizer is used for training, with a learning rate of 0.001, a batch size of 64, and 100 training rounds. The optimization iteration unit of the control measures is constructed based on the genetic algorithm. The binary encoding method is used, and each gene represents the parameters of a control measure. The fitness function combines the traffic flow parameters and user response indicators predicted by LSTM, as well as the confidence of the association rules, in the form of f = w1 * fluency + w2 * user satisfaction + w3 * confidence. f is the value of the fitness function, which combines multiple factors to evaluate the quality of a solution, which can be understood as an overall evaluation of the traffic flow management strategy; w1, w2, and w3 are the weight coefficients of the corresponding parameters, which are used to indicate the importance of different factors in the overall evaluation. Fluency refers to the smoothness and efficiency of traffic flow, such as the speed and smoothness of vehicles passing through the road network. User satisfaction reflects the subjective feelings of users (such as drivers or passengers) about traffic conditions, which may include factors such as travel time, waiting time, comfort or convenience. Confidence refers to the credibility of a rule, that is, the probability that condition B will also occur when condition A occurs. It refers to the strength of the association between traffic patterns discovered based on historical data and actual user behavior. The selection operation adopts the roulette method, with the crossover probability set to 0.8 and the mutation probability set to 0.1. Each iteration generates 100 new solutions, and the top 10 are selected as elites. The optimization results are serialized in JSON format, written into the optimization solution table of the knowledge base in real time, and the relevant statistical indicators are updated. In a certain intelligent traffic control system, the distributed database stores 1 million control records for the past 5 years. The TalendETL tool automatically extracts data from 10 different source systems at 2 a.m. every day, processing about 50GB of incremental information.During the data cleaning process, the regular expression pattern = "^[A-Z]{2}\d{3}￥" was used to verify the license plate format, removing about 2% of abnormal records. Z-Score standardization unified the speed data from different sources to a distribution with a mean of 0 and a standard deviation of 1. The Apriori algorithm was run on the processed data set, with a minimum support of 0.05, which means that at least 50,000 records contain the item set, and a minimum confidence of 0.7 to ensure that the rule has high reliability. The algorithm found 200 strong association rules, such as "morning rush hour + main road + speed limit measures → average speed increase by 15%", with a support of 0.06 and a confidence of 0.75. The LSTM network training used data from the past 4 hours with a time step of 30 minutes, forming a total of 2,880,000 training samples. On the validation set, the mean absolute percentage error (MAPE) for traffic flow prediction was 5.2%, and the accuracy of user satisfaction prediction reached 83%. In the genetic algorithm, each individual is represented by a 50-bit binary code, containing 10 parameters of 5 control measures. The weights in the fitness function are set to w1=0.5, w2=0.3, and w3=0.2. After 500 generations of evolution, the genetic algorithm converged to the local optimal solution. The recommended control scheme improved traffic flow by 12% and user satisfaction by 8% compared with the baseline scheme. The optimization results are written to the knowledge base in real time in the JSON format of {"measure_id": 1, "params": [60, 120, 0.8], "fitness": 0.876}, and the cache update is triggered, so that the new scheme takes effect on the entire system within 3 seconds.

[0043] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for controlling the stability of intelligent transportation, characterized in that: The method includes: obtaining the user's travel time preference, location preference and sensitivity to traffic control measures, building a user attribute model to describe the user's acceptance of different traffic control measures; obtaining real-time traffic flow data and road condition data of the road network, combining historical traffic data and historical abnormal road conditions, training a traffic prediction model, and judging the dynamic change trend of traffic flow data and road condition data; generating candidate combination schemes of control measures based on the user attribute model and the traffic prediction model, screening the candidate combination schemes, and obtaining the optimal control combination strategy; converting the optimal control combination strategy into traffic control instructions to dynamically control traffic flow. Control and guide; During the dynamic control and guidance of traffic flow, real-time monitoring of traffic flow, vehicle speed, and queue length of the road network during the execution of traffic control instructions is carried out to evaluate the actual control effect, and dynamically adjust traffic control instructions according to the actual evaluation results; at the same time, based on the user's feedback on each control measure, the matching degree between user behavior and traffic control is calculated. If the matching degree is lower than the preset threshold, the user behavior is identified as risky behavior; construct a virtual simulation scenario corresponding to traffic control, simulate traffic flow and user behavior under different control measures based on the identified risk behavior, and evaluate the effect and risk of each control measure in the virtual simulation scenario; analyze historical control measures The effects and risks of traffic control measures are analyzed, the correlation between control measures and user behavior and traffic flow is extracted, and an optimization iteration unit of control measures is established; the user's travel time preference, location preference and sensitivity to traffic control measures are obtained, and a user attribute model is constructed to describe the user's acceptance of different traffic control measures, including: obtaining the user's historical travel data, classifying the user's travel time according to the historical travel data, and obtaining the user's time preference type; spatially clustering the destinations in the historical travel data to obtain the user's location preference type; calculating the user's traffic control measures according to the user's compliance rate and complaint rate with different traffic control measures. Sensitivity index; construct a user attribute model, take the time preference type, the location preference type and the traffic control sensitivity index as input variables, and obtain the user's acceptance scores and feedback classifications for different traffic control measures through training on historical data; for new traffic control measures, extract the corresponding acceptance scores and feedback classifications, combine the current transportation mode selection, travel frequency and environmental factors, determine the weights of each factor, and calculate the comprehensive acceptance index; based on the comprehensive acceptance index, combined with the road condition information sensitivity, economic cost and travel purpose classification, predict the user's specific response to the new traffic control measures.

2. The method according to claim 1, characterized in that The method obtains real-time traffic flow data and road condition data of the road network, combines historical traffic flow data and historical abnormal road conditions, trains a traffic prediction model, and determines the dynamic change trend of traffic flow data and road condition data, including: obtaining vehicle quantity information collected by traffic monitoring cameras, driving speed data recorded by vehicle-mounted GPS equipment, and lane occupancy data detected by roadside sensors; generating a traffic status data set in a unified format based on the vehicle quantity information, the driving speed data, and the lane occupancy data; constructing a traffic prediction model based on the traffic status data set, combined with road network topology information and traffic signal control parameters; inputting real-time traffic flow data and road condition data into the traffic prediction model, and obtaining the traffic flow change trend and road condition change trend within a preset time in the future through a sliding time window method.

3. The method according to claim 1, characterized in that The method generates candidate combinations of control measures according to the user attribute model and the traffic prediction model, screens the candidate combinations, and obtains the optimal control combination strategy, including: generating candidate combinations of control measures according to user acceptance data, traffic flow prediction results, and real-time traffic status data, wherein the real-time traffic status data adjusts the candidate generation process by affecting the decision tree splitting condition; encoding the candidate combinations by using a genetic algorithm, wherein each gene represents a control measure parameter, and establishing a fitness function, wherein the fitness function adopts a weighted summation form, and the weight is pre-set according to user acceptance, travel time and location preferences, and traffic stability level; screening the candidate solutions to obtain a set of non-dominated solutions, and calculating the congestion of each solution, wherein the congestion calculation is based on the normalized value of each objective function; selecting the solution with the largest congestion from the set of non-dominated solutions as the optimal control combination strategy, matching the optimal control combination strategy with the real-time traffic status data, and judging whether a preset trigger condition is met, and if the trigger condition is met, sending the control strategy information to the relevant traffic management equipment.

4. The method according to claim 1, characterized in that The method converts the optimal control combination strategy into traffic control instructions to dynamically regulate and guide traffic flow, including: generating a standardized traffic control instruction set according to the optimal control combination strategy, the instruction set including signal light timing adjustment, lane allocation and speed limit setting parameters; if the strategy includes speed limit measures, generating speed limit setting instructions; verifying the traffic control instructions, the verification including generating polynomial selection, data grouping, division operation and remainder comparison; transmitting the verified instructions to the intelligent traffic control system through a secure encrypted channel; after receiving the traffic control instructions, obtaining current road condition parameters, the road condition parameters including vehicle flow and average vehicle speed; real-time regulation of traffic lights, the regulation including fuzzification, reasoning and defuzzification; after receiving the traffic control information, the on-board terminal obtains the current position and destination information of the vehicle and calculates the optimal path.

5. The method according to claim 1, characterized in that In the process of dynamically regulating and guiding the traffic flow, the traffic flow, vehicle speed and queue length of the road network during the execution of the traffic control instructions are monitored in real time, the actual control effect is evaluated, and the traffic control instructions are dynamically adjusted according to the actual evaluation results, including: obtaining the traffic flow, vehicle speed and queue length data collected by the road induction coil and the video monitoring equipment, and preprocessing the original data; judging whether the value exceeds the preset range, if it exceeds, removing the abnormal value; smoothing the data after removing the abnormal value to obtain the preprocessed data; fusing the preprocessed data; obtaining the optimal state estimation value through the iterative update process, and the optimal state estimation value includes the average vehicle speed, traffic density and queue length; calculating the control effect evaluation index according to the optimal state estimation value; obtaining a comprehensive score by using the fuzzy comprehensive evaluation method; establishing a factor set U, a comment set V and a weight set W, and the weight set W is determined by the hierarchical analysis method; generating an adjustment strategy according to the comprehensive score and the preset threshold, obtaining the rules from the rule library, and obtaining the adjustment strategy by using the forward chain reasoning method; generating a control instruction according to the adjustment strategy, and sending the control instruction to the traffic control device for execution.

6. The method according to claim 1, characterized in that The method also calculates the matching degree between the user behavior and the traffic control based on the user's feedback on each control measure, and identifies the user behavior as a risky behavior if the matching degree is lower than a preset threshold, including: obtaining traffic behavior data carrying a user identification number; sending a feature extraction instruction according to the traffic behavior data, the feature extraction instruction is used to extract a keyword vector and a sentiment score from the user feedback; receiving a matching degree calculation request with a user identification number and a feature vector, the feature vector being constructed by combining the keyword vector and the sentiment score with the user behavior data; judging whether the feature vector matches the traffic control measure, including: using a support vector machine algorithm to calculate the matching degree score; if the matching degree score is lower than a preset threshold, it is determined to be a risky behavior.

7. The method according to claim 1, characterized in that The method of constructing a virtual simulation scene corresponding to traffic control simulates traffic flow and user behavior under different control measures based on the identified risk behavior, and evaluates the effect and risk of each control measure in the virtual simulation scene, including: constructing a virtual traffic network model based on geographic information system data including road geometry, intersection layout and traffic facility location; simulating traffic flow operation, determining vehicle acceleration, deceleration and lane change operations according to preset behavior rules; if the vehicle position is updated, calculating new speed and acceleration data; simulating user response behavior using a multi-agent method, in which each agent represents a driver; adjusting the parameter settings of the virtual simulation scene according to the identified risk behavior; calculating the average travel time, congestion index and accident risk assessment index, and comparing the effects of different control measures.

8. The method according to claim 1, characterized in that The method of analyzing the effects and risks of historical control measures, extracting the association rules between control measures and user behavior and traffic flow, and establishing an optimization iteration unit for control measures includes: obtaining historical control measure data, the historical control measure data including control type, implementation time, location, duration, traffic flow change and user feedback; using the Apriori algorithm to analyze the association between control measures and user behavior and traffic flow in the historical control measure data according to preset minimum support and minimum confidence, and obtaining strong association rules; converting the strong association rules into feature vectors as inputs of a long short-term memory network; using the long short-term memory network to predict the impact of different control measures on traffic flow and user behavior, and obtaining prediction results; based on the prediction results, constructing an optimization iteration unit for control measures based on a genetic algorithm, and the optimization iteration unit adopts a binary encoding method; if the optimization iteration unit generates a new control plan, the new control plan is written into the optimization plan table of the traffic control knowledge base.

Images