Intersection short-time traffic flow prediction method based on quantile regression and Bayesian network

Abstract

The invention discloses an intersection short-time traffic flow prediction method based on quantile regression and a Bayesian network, and relates to the technical field of intelligent traffic, and the method comprises the following steps: constructing the Bayesian network, analyzing traffic flow data through employing the quantile regression method, determining the probability space distribution of traffic flow under an abnormal traffic condition, and carrying out the prediction of the traffic flow. Determining parameters in the Bayesian network by using a quantile regression result; according to the method, the time-space and nonlinear characteristics of the traffic flow are accurately grasped, support is provided for strategy formulation and signal optimization of a management department, and the congestion is effectively dredged and the efficiency is effectively improved; network parameters determined by quantile regression can easily deal with various abnormal conditions, guarantee reliable prediction, build a foundation for emergency management and accident early warning, and reduce disordered and loss risks; the prediction system collects updated data in real time and optimizes the model, flexible adjustment is carried out according to scene changes, traffic management is enabled intelligently and finely, system performance and service quality are improved, multiple requirements are met, and sustainable development of traffic is driven.

Description

Technical Field

[0001] The invention relates to the field of intelligent transportation technology, and in particular to a method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network. Background Art

[0002] In the field of intelligent transportation technology, short-term traffic flow prediction at intersections plays a vital role in the efficient management and smooth operation of urban traffic. Accurate traffic flow prediction can provide key basis for traffic management departments to formulate reasonable control strategies, optimize traffic signal timing, and plan travel routes for travelers.

[0003] Traditional traffic flow prediction methods can be mainly divided into several categories, including parameter prediction methods, non-parameter prediction methods, and hybrid prediction methods. However, these methods have gradually exposed some limitations in practical applications.

[0004] On the one hand, traffic flow itself has complex spatiotemporal correlation and nonlinear change characteristics. Not only does it have periodic change patterns such as peaks and flat peaks in the time dimension, but in the spatial dimension, the traffic flows between intersections in the urban road network affect each other. Vehicles leaving the upstream intersection will affect the changes in traffic flow at the downstream intersection, and there is a time difference. The changes in traffic flow between different intersections show strong correlation. At the same time, this spatiotemporal change of traffic flow is nonlinear. Traditional prediction methods often find it difficult to fully and accurately capture these complex change patterns, resulting in a large deviation between the prediction results and the actual traffic flow situation, which cannot meet the needs of actual traffic management and travel guidance, affecting the scientific nature of traffic control strategy formulation and the improvement of road traffic efficiency.

[0005] On the other hand, in actual traffic scenarios, various abnormal conditions often occur, such as traffic accidents, road construction, holiday travel peaks, bad weather, and large-scale events held in the surrounding area. These abnormal conditions will cause the changes in traffic flow to deviate from its normal periodic law, and most traditional traffic flow prediction methods focus on normal data. They have poor adaptability to traffic flow changes under these abnormal conditions and cannot effectively handle abnormal data, which leads to a significant decrease in prediction accuracy and reliability. This will not only cause unreasonable traffic signal timing and aggravate traffic congestion, but may also cause applications such as traffic accident warnings to lose accurate data support, causing traffic chaos, increasing resource waste, and bringing many adverse effects to the normal operation of urban traffic.

[0006] The above existing traffic flow prediction methods have shortcomings in dealing with the spatiotemporal correlation, nonlinear change characteristics and abnormal conditions of traffic flow. There is an urgent need for a method that can more accurately and reliably predict the short-term traffic flow at intersections, so as to improve the level of traffic management, alleviate traffic congestion, ensure the orderly operation of urban traffic, meet the changes in traffic demand in different scenarios, and promote the development of urban traffic towards intelligence and refinement. In view of this, a method for predicting short-term traffic flow at intersections based on quantile regression and Bayesian network is provided to overcome the above problems. Summary of the invention

[0007] The purpose of the present invention is to provide a method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network to solve the problems raised in the above background technology.

[0008] In order to solve the above technical problems, the present invention provides a method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network, comprising the following steps:

[0009] Construct a Bayesian network and model the entire intersection as a Bayesian network structure to describe the spatiotemporal correlation of traffic flow at the intersection;

[0010] The traffic flow data is analyzed by using quantile regression method to determine the probability spatial distribution of traffic flow under abnormal traffic conditions, and the parameters in the Bayesian network are determined using the quantile regression results.

[0011] Traffic flow data is collected in real time and input into the constructed Bayesian network prediction system. The system is used to predict short-term traffic flow at intersections based on the spatiotemporal correlation of traffic flow and the parameters determined by quantile regression. The prediction results are output and the historical data set is continuously updated to optimize the timeliness and accuracy of the prediction results.

[0012] Furthermore, in the step of constructing the Bayesian network, the nodes and directed edges of the Bayesian network are determined based on the connection relationship between the intersections in the urban traffic road network and the flow direction and flow rate change rules of the traffic flow, where the nodes represent the relevant variables of the traffic flow and the directed edges represent the causal relationship between the variables.

[0013] Furthermore, the quantile regression method uses the method of the sum of absolute values ​​of weighted residuals to estimate parameters, which is used to simultaneously fit multiple regression curves of the selected quantile points, and the quantile regression method is tolerant to the random disturbance of a small number of specific data in the sample data.

[0014] Furthermore, in the step of determining the parameters in the Bayesian network using the quantile regression results, the prior probability required by the Bayesian network is calculated through quantile regression, and the prior probability is used to reflect the initial probability distribution of the traffic flow at different quantiles.

[0015] Furthermore, in the Bayesian network prediction system, the conditional probability is determined by calculating the spatiotemporal correlation between intersections. The calculation of the spatiotemporal correlation takes into account the impact of vehicles exiting the upstream intersection in the urban road network on the changes in traffic flow at the downstream intersection and the travel time of vehicles through the road section connecting the two intersections. The Bayesian network is used to accurately simulate the temporal and spatial changes in traffic flow.

[0016] Furthermore, in the step of collecting traffic flow data in real time, high-precision traffic flow monitoring equipment is used, and each key intersection and road section within the study area is covered.

[0017] Furthermore, in the step of optimizing the timeliness and accuracy of the prediction results, the parameters and historical data sets in the Bayesian network are dynamically adjusted according to the deviation between the real-time collected traffic flow data and the prediction results.

[0018] Furthermore, it also includes verification and evaluation steps of the prediction results. By comparing and analyzing the actual traffic flow data, a variety of evaluation indicators, including but not limited to root mean square error, mean absolute error, and mean absolute percentage error, are used to quantitatively evaluate the performance of the prediction method, and the prediction model is further optimized and improved based on the evaluation results.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] By constructing a Bayesian network combined with the quantile regression method, the spatiotemporal correlation and nonlinear change characteristics of traffic flow are fully considered. Compared with traditional methods, the prediction accuracy indicators are significantly improved in complex traffic scenarios in different regions. It can accurately capture the changing laws of traffic flow, provide a strong basis for traffic management departments to formulate control strategies and optimize signal timing, effectively alleviate traffic congestion and improve road traffic efficiency.

[0021] The use of quantile regression method to determine the parameters of the Bayesian network can properly handle various abnormal traffic conditions, enable the prediction method to better adapt to sudden changes in traffic flow, ensure the availability of prediction results, provide a reliable basis for applications such as traffic accident warning and traffic emergency management, reduce the risk of traffic chaos, and reduce economic losses and resource waste caused by congestion and delays.

[0022] The prediction system has the ability to collect traffic flow data in real time, update historical data sets, and optimize and improve the model based on verification and evaluation results. It can adjust the model structure or parameters in a targeted manner according to the changes in traffic flow in different scenarios, closely follow the real-time changes in traffic flow, and provide accurate prediction information to traffic management departments and travelers in a long-term and stable manner, promote the development of urban traffic management towards intelligence and refinement, improve the overall performance and service quality of the transportation system, meet diversified transportation needs, and promote the sustainable development of urban transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The schematic diagram of the present invention is a method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network. DETAILED DESCRIPTION

[0024] 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.

[0025] See also Figure 1 , the present invention provides a technical solution:

[0026] See also Figure 1 As shown, an embodiment 1 of a method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network:

[0027] Scene Setting:

[0028] A busy traffic area in Wuchang District, Wuhan City was selected, which contains multiple road intersections, such as the intersection of Zhongshan Road and Jiefang Road, the intersection of Zhongshan Road and Ziyang Road, etc. These intersections connect the main commercial areas, residential areas and office areas of the city, with large traffic flow and complex traffic flow changes, and have typical urban road traffic characteristics, which can fully verify the effectiveness of this traffic flow prediction method.

[0029] Construct a Bayesian network:

[0030] Detailed data on the connection relationship between each intersection in the urban traffic network in the area is collected, including information such as the connection direction of the road, the number of lanes, traffic control measures, and the flow direction and flow change law data of each intersection. For example, at the intersection of Zhongshan Road and Jiefang Road, in terms of the number of lanes, there are 4 entrance lanes, including 1 left-turn lane, 2 straight lanes, and 1 right-turn lane; there are 3 exit lanes. Through statistical analysis of traffic flow monitoring data at different times of the week (peak hours 7:00-9:00, 17:00-19:00, off-peak hours 9:00-17:00, 19:00-22:00, and nighttime hours 22:00-7:00), it was found that the average hourly entrance traffic volume during peak hours was about 1,785 vehicles, of which left-turning vehicles accounted for 20.3%, straight vehicles accounted for 59.8%, and right-turning vehicles accounted for 19.9%; the average hourly entrance traffic volume during off-peak hours was about 987 vehicles, and the corresponding vehicle types accounted for slightly different proportions; the average hourly entrance traffic volume during nighttime hours was about 312 vehicles. In terms of vehicle speed, the average speed during peak hours was about 24.8km / h, during off-peak hours was about 34.6km / h, and during nighttime hours was about 44.7km / h. The distribution of vehicle types also differs on weekdays and weekends. On weekdays, small cars account for 79.5%, buses account for 10.2%, and trucks account for 10.3%; on weekends, small cars account for 84.8%, buses account for 8.2%, and trucks account for 7%.

[0031] Based on these data, the nodes and directed edges of the Bayesian network are determined. For example, the traffic flow of the entrance lane, the traffic flow of the exit lane, the status of the signal lights, and the saturation of the surrounding parking lots at the intersection of Zhongshan Road and Jiefang Road are taken as nodes, because these variables directly affect the traffic flow conditions of the intersection, and there is a causal relationship between them. For example, the status of the signal lights will affect the traffic flow of the entrance lane, which in turn affects the traffic flow of the exit lane and the saturation of the surrounding parking lots. By reasonably setting directed edges, a Bayesian network structure that can accurately reflect the complex network characteristics of the traffic flow at the intersection is constructed.

[0032] This Bayesian network built based on actual traffic data can accurately describe the spatiotemporal correlation of traffic flow at intersections, providing a solid foundation for subsequent predictions. Compared with traditional prediction methods that do not consider the actual situation of the traffic network, it can more accurately capture the changing trends of traffic flow and improve the accuracy of predictions.

[0033] Quantile regression was used to analyze the data and determine the Bayesian network parameters:

[0034] The long-term traffic flow historical data of the area in the past three months was collected, including data under normal traffic conditions and data under abnormal traffic conditions, such as traffic flow data under special circumstances such as traffic accidents, road construction, and holiday travel peaks. During this period, there were 15 traffic accidents, 8 road constructions, and 6 holiday travel peaks.

[0035] The quantile regression method is used to analyze these data. For example, multiple quantile points such as 0.25, 0.5, and 0.75 are selected, and the parameters are estimated using the sum of absolute values ​​of weighted residuals, while multiple regression curves of these quantile points are fitted. Taking the traffic flow in the entrance lane of the intersection of Zhongshan Road and Jiefang Road as an example, during the peak hours of normal working days, for the regression curve of the 0.25 quantile, its fitting parameters show that the relationship between traffic flow and factors such as the saturation of surrounding roads and the cycle of traffic lights is: traffic flow = 0.32*saturation of surrounding roads + 0.18*cycle length of traffic lights + constant term (the constant term is about 487 calculated through historical data); there are also corresponding fitting equations for the 0.5 quantile and 0.75 quantile, and by calculating indicators such as the residual sum of squares under each quantile, it is found that its goodness of fit is significantly improved compared with the traditional linear regression method. For example, the residual sum of squares under the 0.25 quantile is reduced by about 19.5%, the 0.5 quantile is reduced by about 14.8%, and the 0.75 quantile is reduced by about 17.6%.

[0036] The parameters in the Bayesian network are determined based on the quantile regression results. Specifically, the prior probability required by the Bayesian network is calculated through quantile regression. For example, when calculating the prior probability of the traffic flow in the entrance lane of the intersection of Zhongshan Road and Jiefang Road, the historical traffic flow data under different time periods and weather conditions are considered to determine that during the peak hours of sunny weekdays, the probability of the traffic flow in the entrance lane being 1530-1980 vehicles / hour is 0.38, and the probability of being 1980-2450 vehicles / hour is 0.32. There are corresponding probability distributions in other ranges; and during the peak hours of rainy weekdays, due to factors such as slippery roads and reduced vehicle speeds, the probability of the traffic flow in the entrance lane being 1180-1580 vehicles / hour is 0.33, and the probability of being 1580-1960 vehicles / hour is 0.38, providing basic probability information for the prediction of the Bayesian network, thereby improving the adaptability of the prediction method to traffic flow changes caused by abnormal traffic conditions.

[0037] The application of quantile regression method enables this prediction method to better handle data under abnormal traffic conditions, improves the adaptability of the prediction model to complex traffic conditions, and compared with the prediction method that only relies on normal data, it can significantly reduce the impact of abnormal traffic conditions on prediction accuracy and make the prediction results more reliable.

[0038] Collect traffic flow data in real time and make predictions:

[0039] High-precision traffic flow monitoring equipment, such as microwave radar sensors and video cameras, are installed at key intersections and sections in the area. These devices can accurately collect data such as traffic flow, vehicle speed, and vehicle density in real time, and transmit the data to the data processing center through wireless communication technology. For example, at the intersection of Zhongshan Road and Jiefang Road, the measurement accuracy of microwave radar sensors can reach ±4.8 vehicles / minute (traffic flow) and ±1.8km / h (vehicle speed), and the vehicle recognition accuracy of video cameras is above 95.2%.

[0040] The real-time collected traffic flow data is input into the constructed Bayesian network prediction system. For example, at a certain moment (assuming it is 8:30 am), the traffic flow monitoring data of the entrance lane of the intersection of Zhongshan Road and Jiefang Road shows 1,640 vehicles / hour. The system predicts the traffic flow conditions of the intersection in the short term (such as within 10 minutes) in the future based on the previously constructed traffic flow spatiotemporal correlation model and the parameters determined by quantile regression. Through model calculation, it is estimated that the traffic flow of the exit lane will be about 1,480 vehicles / hour in 10 minutes, and the possible traffic congestion is mild congestion (the congestion index is calculated by the model to be about 0.62, and the congestion index is set to be greater than 0.5 as mild congestion), and the length of the vehicle queue is expected to be about 105 meters, and the prediction results are output.

[0041] At the same time, the prediction system will continuously update the historical data set, incorporate the newly collected data into the historical data, and optimize the parameters of the Bayesian network in real time.

[0042] For example, when there is an abnormal change in traffic flow at 9:30 (traffic volume suddenly increases by 28% due to the end of a large event nearby), the system will automatically adjust the probability distribution of related nodes to make the prediction results more timely and accurate, and able to adapt to real-time changes in traffic flow in a timely manner.

[0043] High-precision traffic flow monitoring equipment and real-time data collection and processing mechanisms ensure that the prediction system can obtain the latest and most accurate traffic information. Combined with the real-time optimization capability of the Bayesian network, the prediction results can closely follow the actual changes in traffic flow, effectively improving the real-time and accuracy of the prediction, and providing traffic management departments and travelers with more timely and reliable traffic flow prediction information, which helps to alleviate traffic congestion and optimize traffic travel planning.

[0044] Verify and evaluate the prediction results:

[0045] The prediction results were compared with the actual traffic flow data, and the performance of the prediction method was quantitatively evaluated using a variety of evaluation indicators, such as root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), etc. For example, the traffic flow prediction results at the intersection of Zhongshan Road and Jiefang Road were statistically analyzed over a week (from Monday to Sunday). The calculation showed that the RMSE was 49.8 vehicles / hour (a decrease of about 14.5 vehicles / hour compared to the traditional prediction method), the MAE was 34.6 vehicles / hour (a decrease of about 9.8 vehicles / hour), and the MAPE was 10.2% (a decrease of about 2.8 percentage points).

[0046] The prediction model is further optimized and improved based on the evaluation results. If the prediction error is found to be large within a certain time period (such as 14:00-16:00 on Wednesday afternoon), the possible reasons are found through in-depth analysis of the data and model parameters, such as unreasonable probability settings for a certain node (the surrounding parking lot saturation node), inaccurate spatiotemporal correlation model of traffic flow, etc., and then the model structure or parameters are adjusted in a targeted manner (for example, the weight of the impact of parking lot saturation on traffic flow is re-evaluated from the original 0.2 to 0.28) to continuously improve the accuracy and reliability of the short-term traffic flow prediction at the intersection.

[0047] By verifying and evaluating the prediction results, problems in the prediction method can be discovered in a timely manner, and targeted optimization and improvement can be carried out, so that the prediction model can continuously adapt to the changing characteristics of traffic flow, continuously improve the prediction performance, and provide guarantees for long-term stable and accurate traffic flow predictions, solving the problem that traditional traffic flow prediction methods are difficult to self-adjust and optimize according to actual conditions.

[0048] Through the specific steps and methods of the above embodiments, the intersection short-term traffic flow prediction method based on quantile regression and Bayesian network effectively solves the problems existing in the short-term traffic flow prediction of urban road intersections:

[0049] In view of the spatiotemporal correlation and nonlinear variation characteristics of traffic flow, by constructing a Bayesian network based on the actual traffic network conditions and using the quantile regression method, we can accurately capture the changing patterns of traffic flow and improve the accuracy of prediction. Compared with the traditional single prediction method, it better adapts to the complex and changeable characteristics of urban traffic flow, and provides strong support for traffic management departments to formulate scientific and reasonable traffic control strategies, which helps to alleviate traffic congestion and improve the traffic efficiency of urban roads.

[0050] In order to solve the problem of a few abnormal conditions interfering with the accuracy of traffic flow prediction, the application of quantile regression method enables the prediction method to effectively handle abnormal data, and the determined Bayesian network parameters can better adapt to the abnormal changes in traffic flow, thereby improving the reliability and stability of the prediction method under various traffic conditions, ensuring the availability of traffic flow prediction results, and providing a more accurate basis for applications such as traffic signal timing and traffic accident warning, reducing traffic chaos and waste of resources caused by inaccurate predictions.

[0051] By collecting traffic flow data in real time, continuously updating historical data sets, and verifying and evaluating prediction results, we have achieved adaptive optimization and continuous improvement of the prediction system, enabling it to respond to real-time changes in traffic flow in a timely manner and maintain a high level of prediction accuracy. This provides travelers with more accurate and real-time route planning suggestions, improves travel efficiency and satisfaction, and also provides key technical support for the development of urban intelligent transportation systems, promoting the intelligent and refined development of urban traffic management.

[0052] In summary, this intersection short-term traffic flow prediction method has significant advantages and innovations in both theory and practice. It can effectively solve practical problems in urban road traffic flow prediction and has broad application prospects and practical value.

[0053] See also Figure 1 As shown, a second embodiment of a method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network is as follows:

[0054] Scene Setting:

[0055] An emerging development area in Hanyang District, Wuhan City, Hubei Province was selected. The area is undergoing large-scale urban construction and development, covering multiple newly built residential communities, commercial complexes, and public facilities such as schools. The traffic demand is growing rapidly and changing in a variety of ways. There are major traffic arteries such as Jiangcheng Avenue and Sixin Avenue in the area, as well as many secondary roads and branches connected to them, forming a complex road network. Among them, the intersection of Jiangcheng Avenue and Sixin South Road, and the intersection of Sixin Avenue and Fangcao Road are key nodes with relatively concentrated traffic flow and complex traffic conditions, which are suitable for verifying the effectiveness of the improved traffic flow prediction method in urban construction and development areas.

[0056] Specific implementation steps:

[0057] Construct a Bayesian network:

[0058] During my internship at Wuhan Chengyuan Municipal Engineering Design Co., Ltd., I participated in actual projects and collected detailed data on the road network in the area. For example, at the intersection of Jiangcheng Avenue and Sixin South Road, the lanes are configured as two-way six lanes, with dedicated left and right turn lanes, and also equipped with bus lanes. In terms of traffic control measures, tidal lane management is implemented on some lanes during peak hours in the morning and evening to alleviate traffic congestion.

[0059] Through long-term traffic flow monitoring, traffic flow data at different time periods (7:00-9:00 in the morning peak, 9:00-16:00 in the off-peak, 16:00-19:00 in the evening peak, and 19:00-7:00 at night) are obtained, including information such as vehicle volume, vehicle speed, and vehicle type ratio. According to statistical analysis, the average hourly vehicle volume at the intersection during the morning peak period reaches about 4,480 vehicles, of which private cars account for about 69.5%, buses account for 15.3%, trucks account for 10.2%, and other vehicles account for 5%; the average vehicle speed is about 19.8 km / h. During the off-peak period, the vehicle volume drops to about 2,760 vehicles per hour, and the vehicle speed increases to about 29.8 km / h.

[0060] Based on these data, the nodes and directed edges of the Bayesian network are determined. The traffic flow of each entrance lane of the intersection, the cycle and timing of the signal light, the usage of the surrounding parking lots, the passenger flow of nearby bus stops, etc. are used as nodes to construct a network structure that reflects the complex relationship of traffic flow. For example, different timings of signal lights will directly affect the traffic flow of each entrance lane, and the saturation of the surrounding parking lots is interrelated with the traffic flow. By reasonably setting directed edges, the inherent logical relationship of traffic flow can be accurately described.

[0061] Quantile regression was used to analyze the data and determine the Bayesian network parameters:

[0062] Quantile regression analysis was performed using a large amount of historical traffic flow data accumulated by the company, covering data from normal working days, weekends, holidays, and special events (such as large-scale events held in the surrounding area, temporary road construction, etc.). Taking multiple quantile points such as 0.1, 0.3, 0.5, 0.7, and 0.9 as examples, the traffic flow in the entrance lane of the intersection of Jiangcheng Avenue and Sixin South Road was modeled.

[0063] During the analysis, the impact of different factors on traffic flow was taken into account, such as weather conditions (sunny, rainy, snowy, etc.), surrounding construction conditions (whether there is construction, construction scale, etc.) and seasonal factors (changes in travel patterns in spring, summer, autumn and winter). The parameters were estimated by the weighted residual absolute value method to obtain regression equations at different quantiles. For example, during the morning rush hour on a sunny weekday with no construction, the regression equation for the 0.5 quantile of traffic flow is: traffic flow = 0.38*saturation of surrounding roads + 0.28*effective green light duration of the signal light + 0.18*traffic flow in the previous period + constant term (calculated to be about 980), and the goodness of fit index shows that the model has been significantly improved compared with the traditional method, such as the residual sum of squares has been reduced by about 24.5%.

[0064] The parameters of the Bayesian network are determined based on the quantile regression results, and the prior probability distribution of each node is calculated. For example, under different weather and surrounding environmental conditions, the probability of the traffic flow in the entrance lane of the intersection being in different ranges is determined. During the morning rush hour on rainy days, the probability of the traffic flow being between 3450-3950 vehicles / hour is 0.28, and the probability of being between 3950-4450 vehicles / hour is 0.38, which provides accurate basic probability information for Bayesian network prediction and enhances adaptability to complex traffic conditions.

[0065] Collect traffic flow data in real time and make predictions:

[0066] Advanced traffic flow monitoring equipment, including high-definition video cameras and geomagnetic sensors, are installed at key intersections and sections in the area. The video cameras can accurately identify vehicle types and license plate information, and the geomagnetic sensors have a high degree of detection accuracy for traffic flow and speed, with the error controlled within a small range (traffic flow error ±2.8 vehicles / minute, speed error ±1.3km / h).

[0067] The real-time collected data is transmitted to the prediction system based on the Bayesian network. For example, at 8:30 a.m. on a certain weekday, the traffic volume of the entrance lane at the intersection of Jiangcheng Avenue and Sixin South Road was monitored to be 4,180 vehicles / hour. The system combined the constructed traffic flow model and the parameters determined by quantile regression to predict the traffic conditions of the intersection in the next 15 minutes. The traffic volume of the exit lane is expected to be about 3,750 vehicles / hour, and there may be local congestion areas (the congestion index calculated by the congestion index calculation model is 0.68, and the congestion index is set to be greater than 0.6 for congestion), the length of the vehicle queue is about 118 meters, and the prediction results are output in a timely manner.

[0068] At the same time, the prediction system updates the historical data set in real time and automatically adjusts the parameters of the Bayesian network according to the newly collected data. For example, when it is found that there is a promotion activity in the surrounding commercial complex during a certain period of time, which leads to an abnormal increase in traffic volume, the system quickly adjusts the probability distribution of related nodes so that the subsequent prediction results can reflect the changing trend of traffic flow in a timely manner, improving the accuracy and timeliness of the prediction.

[0069] Verify and evaluate the prediction results:

[0070] The prediction results were compared with the actual traffic flow data, and the performance of the prediction method was quantitatively evaluated using evaluation indicators such as mean square error (MSE), mean absolute deviation (MAD) and Nash efficiency coefficient (NSE). During the one-month test period, the traffic flow prediction results at the intersection of Jiangcheng Avenue and Sixin South Road were statistically analyzed, and the calculated MSE was 44.8 (reduced by about 17.5 compared with the traditional prediction method), MAD was 31.6 (reduced by about 11.5), and NSE was 0.81 (increased by about 0.14), indicating that this prediction method has high accuracy and reliability.

[0071] The prediction model is further optimized and improved based on the evaluation results. If the prediction error is found to be large in a certain time period (such as weekend afternoons), through in-depth analysis of the data and model parameters, it is found that this is due to the failure to fully consider the impact of weekend leisure travel patterns on traffic flow. Then, the model structure is adjusted in a targeted manner, and nodes and parameters that reflect weekend travel characteristics are added, such as the correlation between passenger flow and traffic flow in surrounding parks and shopping malls, so as to continuously improve the accuracy and stability of traffic flow prediction to adapt to traffic flow changes in different scenarios.

[0072] Through practical application in Hanyang District of Wuhan City, this traffic flow prediction method has shown significant advantages:

[0073] In view of the complex and changeable traffic flow characteristics in urban construction and development areas, by constructing a Bayesian network based on actual road networks and traffic data, and using the quantile regression method, the temporal and spatial variation of traffic flow is accurately captured, significantly improving the accuracy and reliability of predictions. Compared with traditional prediction methods, it more effectively copes with the complexity and uncertainty of traffic flow, provides strong support for traffic management departments to rationally plan traffic signal timing and optimize road resource allocation, helps to alleviate traffic congestion problems in urban construction, and improves road traffic efficiency.

[0074] In dealing with special events and abnormal traffic conditions, the quantile regression method enables the prediction method to fully consider various abnormal factors, accurately determine the Bayesian network parameters, and better adapt to sudden changes in traffic flow. Whether it is surrounding construction, large-scale activities or bad weather, it can provide reliable traffic flow prediction information, provide an accurate basis for traffic emergency management and travelers to plan their trips reasonably, and reduce economic losses and resource waste caused by traffic congestion and delays.

[0075] With the help of real-time traffic data collection, historical data update, and continuous verification and optimization mechanism of prediction results, the prediction system is dynamically adaptively adjusted, so that it can closely follow the real-time changes of traffic flow and maintain a high prediction accuracy. This not only provides timely and effective decision-making support for traffic management departments, but also provides accurate traffic information services for the majority of travelers, which helps to improve the level of intelligent management of urban traffic and promote the sustainable development of urban traffic. It has broad application prospects and practical value in the process of urban construction and development.

[0076] Summarize:

[0077] The above two embodiments respectively selected the mature and busy traffic area of ​​Wuchang District in Wuhan and the emerging development traffic area of ​​Hanyang District, and verified the short-term traffic flow prediction method of intersections based on quantile regression and Bayesian network in detail. In the stage of constructing the Bayesian network, nodes and directed edges are determined based on the actual road network data, such as detailed statistics of the number of lanes, traffic control measures, and traffic volume, speed, and vehicle type ratio in each period, so as to accurately reflect the complex relationship of traffic flow. When the quantile regression method analyzes data to determine the parameters of the Bayesian network, it fully considers various factors, including modeling of different quantile points, special events, and various environmental factors, effectively improving the model's goodness of fit and adaptability to abnormal traffic conditions, and determining the accurate prior probability distribution of each node. In the real-time collection of traffic flow data and prediction link, relying on high-precision monitoring equipment and real-time data transmission and processing mechanism, combined with the constructed model parameters to predict short-term traffic flow conditions, and can optimize the Bayesian network parameters in real time to adapt to traffic flow changes. Finally, the prediction results are verified and evaluated through a variety of evaluation indicators, and the model is optimized and improved according to the results to continuously improve the prediction performance.

[0078] Improved accuracy:

[0079] In Example 1, through the traffic flow prediction of the intersection of Zhongshan Road and Jiefang Road for one week, it was calculated that the RMSE was 49.8 vehicles / hour (a decrease of about 14.5 vehicles / hour compared with the traditional prediction method), the MAE was 34.6 vehicles / hour (a decrease of about 9.8 vehicles / hour), and the MAPE was 10.2% (a decrease of about 2.8 percentage points).

[0080] In Example 2, during the one-month test period at the intersection of Jiangcheng Avenue and Sixin South Road, the calculated MSE was 44.8 (reduced by about 17.5 compared to the traditional prediction method), the MAD was 31.6 (reduced by about 11.5), and the NSE was 0.81 (increased by about 0.14).

[0081] These data fully demonstrate that under complex traffic scenarios in different regions, this prediction method has significantly improved accuracy compared to traditional methods, can accurately capture the changing patterns of traffic flow, adapt to the spatiotemporal correlation and nonlinear change characteristics of traffic flow, and effectively support the management and control decisions of traffic management departments, improve road traffic efficiency, such as playing an important role in alleviating traffic congestion, providing more accurate basis for traffic signal timing, reducing unnecessary waiting time for vehicles, and improving overall traffic operation efficiency.

[0082] Improved adaptability to abnormal conditions:

[0083] Wuchang District, where Example 1 is located, takes into account abnormal situations such as traffic accidents, road construction, and holiday travel peaks. The Bayesian network parameters are determined by the quantile regression method to effectively handle abnormal data. For example, it can accurately predict traffic flow changes under special weather conditions such as rainy days, thereby ensuring the availability of traffic flow prediction results, providing a reliable basis for applications such as traffic accident warnings, reducing the risk of traffic congestion caused by abnormal situations, and reducing waste of resources. For example, it avoids chain congestion of surrounding roads caused by traffic accidents, allowing rescue vehicles to arrive at the scene more quickly and reducing the long-term impact of accidents on traffic.

[0084] In the second embodiment, Hanyang District is located in an urban construction and development area, so special events are more diverse, such as promotional activities in surrounding commercial complexes and construction in the surrounding areas. Through real-time monitoring and quantile regression analysis, the system can quickly adjust the Bayesian network parameters, accurately predict traffic flow changes, provide effective support for traffic emergency management, ensure that travelers plan their trips reasonably, and reduce economic losses caused by traffic congestion and delays, such as guiding vehicles to take reasonable detours during commercial activities to avoid severe congestion caused by a large number of vehicles rushing in.

[0085] The system's adaptive optimization and continuous improvement capabilities:

[0086] In both embodiments, the prediction system can collect traffic flow data in real time, continuously update historical data sets, and optimize and improve the prediction model based on the verification and evaluation results. For example, in Embodiment 1, the model structure or parameters are adjusted in a targeted manner based on the prediction error in a specific time period on Wednesday afternoon (such as adjusting the weight of the impact of the saturation of surrounding parking lots on traffic flow); in Embodiment 2, corresponding nodes and parameters are added based on the impact of leisure travel modes on traffic flow on weekend afternoons (such as the correlation between the passenger flow of surrounding parks and shopping malls and traffic flow).

[0087] This adaptive optimization and continuous improvement capability enables the prediction system to closely follow the real-time changes in traffic flow, maintain a high level of prediction accuracy, and provide long-term stable and accurate traffic flow prediction information to traffic management departments and travelers. It promotes the development of intelligent and refined urban traffic management, meets the changing traffic demand in different scenarios, and improves the overall performance and service quality of the urban transportation system. For example, it provides travelers with reliable route planning suggestions in the long term, improves travel efficiency and satisfaction, and promotes the sustainable development of urban transportation.

[0088] The two implementation examples cover areas at different stages of development and functional characteristics. Wuchang District represents a mature urban core area, where traffic flow has typical peak-to-peak variation patterns and complex mixed traffic characteristics of commercial, residential and office areas; while Hanyang District, as an emerging development area, has a rapid increase in traffic demand and is greatly affected by urban construction activities, and traffic flow changes are more diverse and uncertain, including temporary traffic control caused by construction projects, changes in travel patterns caused by the opening of commercial facilities, etc. By verifying these two areas with significant differences, the effectiveness and adaptability of the traffic flow prediction method in various complex situations can be fully demonstrated, which more powerfully proves that the method can not only cope with conventional traffic scenarios, but also play a stable role in special environments and abnormal conditions. It has broad application potential and practical value, can meet the traffic flow prediction needs of different urban areas, and provide comprehensive and reliable technical support for urban traffic management and travelers, further highlighting its advantages in theory and practice.

Claims

1. A method for predicting short-term traffic flow at intersections based on quantile regression and Bayesian network, characterized in that: The following steps are involved: Construct a Bayesian network and model the entire intersection as a Bayesian network structure to describe the spatiotemporal correlation of traffic flow at the intersection; The traffic flow data is analyzed by using quantile regression method to determine the probability spatial distribution of traffic flow under abnormal traffic conditions, and the parameters in the Bayesian network are determined using the quantile regression results. Traffic flow data is collected in real time and input into the constructed Bayesian network prediction system. The system is used to predict short-term traffic flow at intersections based on the spatiotemporal correlation of traffic flow and the parameters determined by quantile regression. The prediction results are output and the historical data set is continuously updated to optimize the timeliness and accuracy of the prediction results.

2. The method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network as claimed in claim 1, characterized in that: In the step of constructing the Bayesian network, the nodes and directed edges of the Bayesian network are determined based on the connection relationship between the intersections in the urban traffic network and the flow direction and flow change rules of the traffic flow, where the nodes represent the relevant variables of the traffic flow and the directed edges represent the causal relationship between the variables.

3. The method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network as claimed in claim 1, characterized in that: The quantile regression method uses the method of the sum of the absolute values ​​of weighted residuals to estimate parameters, which is used to simultaneously fit multiple regression curves of the selected quantile points. The quantile regression method is also tolerant to the random disturbance of a few specific data in the sample data.

4. The method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network as claimed in claim 1, characterized in that: In the step of determining the parameters in the Bayesian network using the quantile regression results, the prior probability required by the Bayesian network is calculated through quantile regression, and the prior probability is used to reflect the initial probability distribution of traffic flow under different quantiles.

5. The method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network as claimed in claim 1, characterized in that: In the Bayesian network prediction system, the conditional probability is determined by calculating the spatiotemporal correlation between intersections. The calculation of spatiotemporal correlation takes into account the impact of vehicles exiting the upstream intersection in the urban road network on the traffic flow changes at the downstream intersection and the travel time of vehicles passing through the road section connecting the two intersections. The Bayesian network is used to accurately simulate the temporal and spatial changes in traffic flow.

6. The method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network as claimed in claim 1, characterized in that: In the step of collecting traffic flow data in real time, high-precision traffic flow monitoring equipment is used, covering all key intersections and sections within the study area.

7. The method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network as claimed in claim 1, characterized in that: In the step of optimizing the timeliness and accuracy of the prediction results, the parameters and historical data sets in the Bayesian network are dynamically adjusted according to the deviation between the real-time collected traffic flow data and the prediction results.

8. The method for predicting short-term traffic flow at an intersection based on quantile regression and Bayesian network as claimed in claim 1, characterized in that: It also includes verification and evaluation steps for the prediction results. By comparing and analyzing the actual traffic flow data, a variety of evaluation indicators, including but not limited to root mean square error, mean absolute error, and mean absolute percentage error, are used to quantitatively evaluate the performance of the prediction method, and the prediction model is further optimized and improved based on the evaluation results.

Images