Expressway congestion state detection method based on multi-source data fusion

Abstract

The invention relates to the field of expressway traffic management, in particular to an expressway congestion state detection method based on multi-source data fusion, which integrates data of a vehicle detector, an ETC (electronic toll collection) system, an Gade map and the like of a to-be-detected road section, accurately extracts traffic congestion characteristics by using a data analysis technology, and obtains a traffic congestion state through an efficient data fusion strategy. According to the invention, the road network service capability is enhanced, the real-time and consistent perception of the traffic state is realized, the traffic management efficiency is obviously improved, the traffic flow is optimized, the congestion is reduced, the travel experience of a driver is improved, and the highway traffic management efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of highway traffic management, and in particular to a highway congestion state detection method based on multi-source data fusion. Background Art

[0002] As an important part of urban transportation, highways have high vehicle speeds. Once traffic congestion occurs, it will not only affect road traffic efficiency, but may also cause traffic accidents and cause serious consequences. Therefore, in the field of highway traffic management, fast and accurate traffic congestion analysis technology is crucial. This technology can analyze the collected traffic data in real time, identify the traffic congestion status in time, and thus improve the effect of traffic control and induction. This has significant academic and practical value in alleviating traffic congestion and improving road use efficiency.

[0003] At present, the deployment and promotion of various detection equipment on highways provide conditions for obtaining multi-source data, such as vehicle inspection devices, ETC (Electronic Toll Collection) systems, and Amap data. Therefore, based on the existing detection conditions of highways, monitoring the traffic operation characteristics of highways from different angles and collecting data information from different sources provide a basis for studying multi-source data fusion models for detecting highway traffic congestion. However, the existing technology still has the following disadvantages when applying these multi-source data:

[0004] ① Limitations of a single data source: Due to data constraints, most of the data in previous studies came from a single data source, such as fixed detectors or video detectors. Although these data sources can provide some traffic information, it is difficult to fully reflect the real-time traffic operation status on the highway;

[0005] ② Limitations of detector structure: Some common detectors have different types of detected data and representations of road traffic conditions due to their own structures and advantages and disadvantages. For example, fixed detectors are only suitable for traffic status detection at nodes or sections. Whether they can accurately represent the traffic flow status of a road section is closely related to their layout density. Although video detectors can provide intuitive traffic images, they are easily affected by factors such as ambient light and weather, and the collected data may have deviations when expressing traffic flow status. Another example is ETC toll data. Although the ETC system can provide high-precision vehicle traffic data, the data it collects may be affected by factors such as traffic delays, and it mainly reflects the traffic conditions near toll stations, making it difficult to fully cover all sections of the highway.

[0006] ③ Complexity of the traffic system: The traffic system is continuous, volatile, and random. The traffic flow state perceived by a single detector is far from meeting the increasingly complex requirements of highway management and control. Through the redundancy and complementarity of multi-source detection data, the fusion result is more complete and accurate than the characterization and description of the highway traffic flow state by a single traffic detector, and can more realistically describe the actual road traffic conditions of the highway. Although multi-source data fusion can provide more comprehensive traffic information, how to effectively integrate different data sources, handle data inconsistencies, and optimize fusion algorithms are still challenges faced by existing technologies. Summary of the invention

[0007] The purpose of the present invention is to provide a highway congestion status detection method based on multi-source data fusion in response to the corresponding deficiencies in the prior art. By integrating data from vehicle inspection devices, ETC systems, and Amap, data analysis technology is used to accurately extract traffic congestion characteristics, and through efficient data fusion strategies, the road network service capability is enhanced, real-time and consistent perception of traffic status is achieved, traffic management efficiency is significantly improved, traffic flow is optimized, congestion is reduced, driver travel experience is improved, and highway traffic management efficiency is improved.

[0008] The objective of the present invention is achieved by adopting the following scheme:

[0009] A highway congestion state detection method based on multi-source data fusion includes the following steps:

[0010] 1) Using various traffic data obtained from different channels to detect the congestion status of the road section to be inspected, and obtaining the unblocked probability and / or congested probability of the road section to be inspected calculated based on the traffic data from different channels;

[0011] 2) Using fuzzy comprehensive evaluation method to perform decision-level fusion on the unblocked probability and / or congested probability of the road section to be inspected calculated based on traffic data from different channels, and calculate the comprehensive unblocked probability and comprehensive congested probability of the road section to be inspected;

[0012] 3) The congestion status of the road section to be inspected is determined based on the relationship between the comprehensive unblocked probability and the comprehensive congested probability of the road section to be inspected.

[0013] Preferably, in step 3), according to the relationship between the comprehensive unblocked probability and the comprehensive congested probability of the road section to be inspected, the specific manner of judging the congested state of the road section to be inspected is as follows:

[0014] If the comprehensive unblocked probability of the road section to be inspected is less than the comprehensive congested probability, the current road section to be inspected is in a congested state;

[0015] If the comprehensive unblocked probability of the road section to be inspected is greater than the comprehensive congested probability, the current road section to be inspected is in a non-congested state.

[0016] Preferably, the various types of traffic data obtained from different channels include vehicle inspection data, ETC data, and Amap data. The specific steps of calculating the comprehensive smooth traffic probability and the comprehensive congestion probability of the road section to be inspected include:

[0017] 1-1) Using the vehicle detector data, calculate the vehicle detector unblocked probability and the vehicle detector congestion probability of the current state of the road section to be inspected;

[0018] 1-2) Using ETC data, calculate the ETC smooth flow probability and ETC congestion probability of the current state of the road section to be inspected;

[0019] 1-3) Use Amap data to calculate the probability of small-scale congestion and large-scale congestion in the current state of the road section to be inspected;

[0020] 1-4) A fuzzy comprehensive evaluation method is used to perform decision-level fusion on the unobstructed probability and / or congested probability of the road section to be inspected obtained based on the vehicle inspection device data, ETC data, and Amap data, and calculate the comprehensive unobstructed probability and comprehensive congested probability of the road section to be inspected.

[0021] Preferably, in step 1-4), the specific method of calculating the comprehensive unblocked probability and the comprehensive congested probability of the road section to be inspected is as follows:

[0022] 1-4-1) Through experiments, the accuracy of the vehicle inspection data, ETC data, and Amap data in determining the current congestion status of the road section to be inspected is calculated respectively, and used as the initial factor weights of the corresponding data;

[0023] 1-4-2) Use the initial factor weights of vehicle inspection data, ETC data, and Amap data to form a factor weight vector;

[0024] 1-4-3) Normalizing the factor weight vector to obtain a normalized factor weight vector;

[0025] 1-4-4) A congestion detection result evaluation matrix is ​​established based on the vehicle inspection device real-time data, ETC real-time data, and Amap real-time data, as shown below:

[0026]

[0027] Where R is the congestion detection result evaluation matrix, R 11 is the probability of vehicle inspection device being unblocked, R 12 is the vehicle detector congestion probability, R 21 is the ETC smooth flow probability, R 22 is the ETC congestion probability, R 31 is the probability of small-scale congestion, R 32 is the probability of large-scale congestion;

[0028] 1-4-5) Perform fuzzy synthesis calculation according to the following formula to obtain the congestion fusion detection result:

[0029] B=A·R=(b1,b2)

[0030] Where B is the congestion fusion detection result, A is the factor weight vector after normalization, R is the congestion detection result evaluation matrix, and “·” represents the fuzzy synthesis algorithm.

[0031] Preferably, in step 1-1), the specific method of using the vehicle detector data to calculate the vehicle detector unblocked probability of the current state of the road section to be inspected and the vehicle detector congestion probability is as follows:

[0032] 1-1-1) According to the traffic parameters corresponding to several historical samples in the historical data of the vehicle detector, the combined feature vectors of the vehicle detector corresponding to the several historical samples are extracted, and the combined feature vectors of the vehicle detector corresponding to each historical sample in the historical data of the vehicle detector are clustered and analyzed using the fuzzy C-means clustering algorithm, and the number of clusters is determined to establish an FCM clustering model capable of judging the congestion status of a road section;

[0033] 1-1-2) According to the traffic parameters corresponding to several real-time samples in the vehicle detector real-time data, the vehicle detector combination feature vectors corresponding to several real-time samples are extracted and input into the FCM clustering model to obtain the vehicle detector unblocked probability and the vehicle detector congestion probability of the current state of the road section to be inspected.

[0034] Preferably, in step 1-1-2), the vehicle detector unblocked probability and the vehicle detector congested probability of the current state of the road section to be inspected are specifically obtained as follows:

[0035] 1-1-2-1) Obtain various traffic parameters corresponding to several real-time samples in the vehicle detector real-time data, extract the vehicle detector combination feature vectors corresponding to the several real-time samples, and input all the obtained vehicle detector combination feature vectors into the FCM clustering model to obtain the clustering center of various traffic feature parameters in all vehicle detector combination feature vectors;

[0036] 1-1-2-2) Calculate the Euclidean distance between each type of traffic characteristic parameter in each vehicle detector combination feature vector and the corresponding cluster center to form a Euclidean distance data set;

[0037] 1-1-2-3) Normalize the data in the Euclidean distance data set and map the data in the Euclidean distance data set to the interval (d1, d2), and d1+d2=1;

[0038] 1-1-2-4) The left endpoint d1 of the interval (d1, d2) is taken as the vehicle detector unblocked probability, and the right endpoint d2 of the interval (d1, d2) is taken as the vehicle detector congested probability.

[0039] Preferably, the vehicle detector combined feature vector includes the first, second and third traffic feature parameters, and the specific method for obtaining the vehicle detector combined feature vector is as follows:

[0040] a) Set a detection period T and extract the traffic parameters corresponding to several samples from the vehicle detector data:

[0041] The upstream location occupancy parameter s of the road section to be inspected during the detection cycle i-1 , downstream location occupancy parameter s i , upstream velocity parameter v i-1 , downstream location velocity parameter v i , and the upstream location flow parameter f i-1 , flow parameter f at downstream location i ;

[0042] b) Using the traffic parameters extracted in step a), calculate the absolute difference Δs between the upstream and downstream locations of the road section to be inspected within the detection period, and the ratio s between the absolute difference between the upstream and downstream locations and the upstream location occupancy rate of the road section to be inspected within the detection period: u , as the first traffic characteristic parameter of the road section to be inspected:

[0043]

[0044] Where Δs is the absolute difference in occupancy between upstream and downstream locations, s i-1 is the upstream location occupancy, s i is the downstream location occupancy, s u is the ratio of the absolute difference in the occupancy rates of the upstream and downstream locations to the occupancy rate of the upstream location;

[0045] c) Using the traffic parameters extracted in step a), analogy calculation is performed according to the following formula to obtain the absolute difference in speed between the upstream and downstream locations Δv, and the ratio of the absolute difference in speed between the upstream and downstream locations to the speed of the upstream location v u , as the second traffic characteristic parameter of the road section to be inspected:

[0046]

[0047] Where Δv is the absolute difference in velocity between the upstream and downstream locations, v u is the ratio of the absolute difference in velocity between the upstream and downstream locations to the velocity at the upstream location, v i-1 is the velocity parameter at the upstream location, v i is the velocity parameter at the downstream location;

[0048] d) Using the traffic parameters extracted in step a), analogy calculation is performed according to the following formula to obtain the absolute difference Δf between the upstream and downstream flow rates and the ratio f between the absolute difference between the upstream and downstream flow rates and the upstream flow rate u , as the third traffic characteristic parameter of the road section to be inspected:

[0049]

[0050] Where Δf is the absolute difference in flow rate between upstream and downstream locations, f u is the ratio of the absolute difference in flow between the upstream and downstream locations to the flow at the upstream location, f i-1 is the flow parameter of the upstream location, f i is the flow parameter of the downstream location;

[0051] e) The first, second and third traffic characteristic parameters of the road section to be inspected are combined to form a vehicle detector combined characteristic vector.

[0052] Preferably, in step 1-2), the specific method of using ETC data to calculate the ETC smooth flow probability of the current state of the road section to be inspected and the ETC congestion probability is as follows:

[0053] 1-2-1) According to the traffic parameters corresponding to several historical samples in the ETC historical data of the road section to be inspected, the ETC combination feature vectors corresponding to the several historical samples are extracted, and the ETC combination feature vectors corresponding to each historical sample in the ETC historical data are clustered by using the Gaussian mixture model to establish a GMM clustering model capable of judging the congestion status of the road section;

[0054] 1-2-2) Obtain various traffic parameters corresponding to several real-time samples in the ETC real-time data, extract the ETC combined feature vectors corresponding to the several real-time samples, and input all the obtained ETC combined feature vectors into the GMM clustering model, so that the GMM clustering model outputs the ETC smooth probability and ETC congestion probability of the current state of the road section to be inspected.

[0055] Preferably, the ETC combined feature vector includes the fourth, fifth and sixth traffic feature parameters, which are specifically obtained as follows:

[0056] ⑴ Set a detection period T and extract traffic parameters corresponding to several samples from ETC data:

[0057] The upstream flow rate f of the road section to be inspected during the detection period, and the time t when the vehicle passes the upstream ETC gantry i-1 , the time t when the vehicle passes the downstream ETC gantry i , the number of large vehicles passing through the upstream ETC gantry h, the number of small vehicles passing through the upstream ETC gantry c;

[0058] ⑵ Using the traffic parameters extracted in step ⑴, calculate the average travel speed of vehicles on the road section to be inspected during the inspection period according to the following formula: As the fourth traffic characteristic parameter of the road section to be inspected:

[0059]

[0060] In the formula, is the average travel speed of vehicles on the road section to be inspected during the inspection period, t i-1 is the time when the vehicle passes the upstream ETC gantry, t i is the time when the vehicle passes the downstream ETC gantry, and f is the upstream flow rate of the road section to be inspected within the detection period;

[0061] ⑶ Using the traffic parameters extracted in step ⑴, calculate the average upstream traffic density d of the road section to be inspected during the detection period according to the following formula as the fifth traffic characteristic parameter of the road section to be inspected:

[0062]

[0063] Where d is the average traffic density upstream of the road section to be inspected during the detection period, and f is the flow rate upstream of the road section to be inspected during the detection period. The average travel speed of vehicles on the road section to be inspected during the inspection period;

[0064] (4) Using the traffic parameters extracted in step (1), calculate the ratio p of the number of large vehicles to small vehicles in the road section to be inspected during the detection period according to the following formula as the sixth traffic characteristic parameter of the road section to be inspected:

[0065]

[0066] Where p is the ratio of large vehicles to small vehicles on the road section to be inspected during the detection period, h is the number of large vehicles passing through the ETC gantry upstream of the road section to be inspected during the detection period, and c is the number of small vehicles passing through the ETC gantry upstream of the road section to be inspected during the detection period.

[0067] (5) Use the fourth, fifth and sixth traffic characteristic parameters of the road section to be inspected to form an ETC combined feature vector.

[0068] Preferably, in step 1-3), the specific method of using Amap data to calculate the small-scale congestion probability of the current state of the road section to be inspected and the large-scale congestion probability is as follows:

[0069] 1-3-1) Determine the congestion length threshold based on Amap’s historical data;

[0070] 1-3-2) Determine the small-scale congestion probability and the large-scale congestion probability of the current state of the road section to be inspected in the following manner:

[0071] 1-3-2-1) When the actual congestion length is less than the congestion length threshold, a comparative experiment is conducted to determine the accuracy of Amap's judgment on the congestion status of the road section to be inspected, and the judgment accuracy is used as the probability of small-scale congestion;

[0072] 1-3-2-2) When the actual congestion length is ≥ the congestion length threshold, a comparative experiment is conducted to determine the accuracy of Amap's judgment on the congestion status of the road section to be inspected, and the judgment accuracy is used as the probability of large-scale congestion.

[0073] The beneficial effects of the present invention are as follows:

[0074] 1. The present invention effectively integrates traffic data from different sources such as vehicle detectors, ETC systems and Amap, and comprehensively utilizes multi-source data, which not only broadens the coverage of traffic status information, but also enhances the depth and breadth of the data;

[0075] 2. Based on the characteristics of each data source, the present invention constructs targeted detection models (such as FCM and GMM clustering algorithms). These models can deeply analyze the traffic status from different angles, ensuring the accuracy and reliability of the analysis results;

[0076] 3. The present invention adopts advanced data fusion technology to achieve effective integration of multi-source information at the decision-making level, giving play to the redundancy and complementarity of multi-source detection data in space or time, improving the consistency of traffic state perception, and significantly enhancing the stability and credibility of congestion detection results by reducing information conflicts and uncertainties;

[0077] 4. The present invention can not only accurately obtain real-time traffic congestion information, greatly improving the operational safety and efficiency of the expressway, but also help prevent traffic accidents by timely identifying and responding to traffic congestion, reduce the loss of social benefits caused by congestion, and improve overall traffic smoothness.

[0078] Glossary

[0079] Vehicle detector data: Vehicle detectors are usually installed at specific locations on highways, such as under lanes or on the side of the road. They can obtain high-precision parameters such as vehicle passing time, speed, and vehicle model. These data have high spatial location accuracy and record traffic information based on the actual installation location, but the coverage is mainly the lanes or sections around the installation point.

[0080] ETC data: ETC (Electronic Toll Collection) data contains vehicle travel records, including the time of passing through the ETC gantry, vehicle type (through vehicle type classification equipment), toll information, etc. ETC gantries have a certain distribution pattern along the highway, and their data can reflect the driving trajectory of vehicles on longer sections of the highway. The spatial scope is the location of the ETC gantry and the interval it covers.

[0081] Amap data: Amap data is mainly obtained through navigation software such as "Amap" on mobile devices (such as smartphones). It includes information such as the vehicle's location, speed, and driving direction. Its advantage is that it has a wide coverage area and a large number of users participate in data collection, but the data accuracy may be affected by multiple factors such as signals and user equipment. The spatial location is based on the positioning of the mobile device.

[0082] Congestion probability: Congestion probability refers to the probability of traffic congestion in a specific traffic environment (such as a certain road section, a certain area or a certain traffic network) and a specific time range (such as peak hours, holidays, etc.). It is a quantitative indicator, usually expressed as a value between 0 and 1. For example, a congestion probability of 0.3 means that under given conditions, there is a 30% probability of traffic congestion. The vehicle detector congestion probability in the present invention is the congestion probability obtained using the vehicle detector data, the ETC congestion probability is the congestion probability obtained using the ETC data, and the small-scale congestion probability and the large-scale congestion probability are the congestion probabilities obtained using Amap data. The unblocked probability is relative to the congestion probability, and in the present invention it refers to the non-congested probability. For example, a congestion probability of 0.3 means that under given conditions, there is a 30% probability of traffic congestion. Then, relatively speaking, the unblocked probability is 1-0.3=0.7, that is, there is a 70% probability that traffic congestion will not occur. The vehicle inspection device unblocked probability is the unblocked probability obtained using the vehicle inspection device data, and the ETC unblocked probability is the unblocked probability obtained using the ETC data. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 is a flow chart of the present invention;

[0084] Figure 2 Schematic diagram of a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0085] like Figure 1 As shown, a highway congestion state detection method based on multi-source data fusion includes the following steps:

[0086] ① Using various traffic data obtained from different channels to detect the congestion status of the road section to be inspected, and obtaining the unblocked probability and / or congested probability of the road section to be inspected calculated based on the traffic data from different channels;

[0087] ② The fuzzy comprehensive evaluation method is used to perform decision-level fusion on the unblocked probability and / or congested probability of the road section to be inspected calculated based on traffic data from different channels, and the comprehensive unblocked probability and comprehensive congested probability of the road section to be inspected are calculated;

[0088] ③ According to the relationship between the comprehensive unblocked probability and the comprehensive congested probability of the road section to be inspected, the congestion state of the road section to be inspected is judged in the following way:

[0089] If the comprehensive unblocked probability of the road section to be inspected is less than the comprehensive congested probability, the current road section to be inspected is in a congested state;

[0090] If the comprehensive unblocked probability of the road section to be inspected is greater than the comprehensive congested probability, the current road section to be inspected is in a non-congested state.

[0091] like Figure 2 As shown, the following examples are made according to the above method:

[0092] The various types of traffic data obtained from different channels include vehicle inspection data, ETC data, and Amap data.

[0093] 1) Use vehicle detector data, ETC data, and Amap data to detect the congestion status of the road section to be inspected, and obtain the smooth probability and / or congestion probability of the road section to be inspected corresponding to the vehicle detector data, ETC data, and Amap data respectively, and use the fuzzy comprehensive evaluation method for decision-level fusion to calculate the comprehensive smooth probability and comprehensive congestion probability of the road section to be inspected:

[0094] 1-1) Use the vehicle detector data to calculate the vehicle detector unblocked probability and the vehicle detector congestion probability of the current state of the road section to be inspected. The specific method is as follows:

[0095] 1-1-1) According to the traffic parameters corresponding to several historical samples in the historical data of the vehicle detector, the combined feature vectors of the vehicle detector corresponding to the several historical samples are extracted, and the combined feature vectors of the vehicle detector corresponding to each historical sample in the historical data of the vehicle detector are clustered and analyzed using the fuzzy C-means clustering algorithm, and the number of clusters is determined (the number of clusters k=2), and an FCM clustering model capable of judging the congestion status of a road section is established;

[0096] 1-1-2) According to the traffic parameters corresponding to several real-time samples in the real-time data of the vehicle detector, the combined feature vectors of the vehicle detector corresponding to several real-time samples are extracted and input into the FCM clustering model to obtain the vehicle detector unblocked probability and the vehicle detector congestion probability of the current state of the road section to be inspected. The specific method of obtaining is as follows:

[0097] 1-1-2-1) Obtain various traffic parameters corresponding to several real-time samples in the vehicle detector real-time data, extract the vehicle detector combination feature vectors corresponding to the several real-time samples, and input all the obtained vehicle detector combination feature vectors into the FCM clustering model to obtain the clustering center of various traffic feature parameters in all vehicle detector combination feature vectors;

[0098] 1-1-2-2) Calculate the Euclidean distance between each type of traffic characteristic parameter in each vehicle detector combination feature vector and the corresponding cluster center to form a Euclidean distance data set;

[0099] 1-1-2-3) Normalize the data in the Euclidean distance data set and map the data in the Euclidean distance data set to the interval (d1, d2), and d1+d2=1;

[0100] 1-1-2-4) The left endpoint d1 of the interval (d1, d2) is taken as the vehicle detector unblocked probability, and the right endpoint d2 of the interval (d1, d2) is taken as the vehicle detector congested probability.

[0101] Of course, if only the vehicle inspection data is used, it is also possible to preliminarily determine whether the road section to be inspected is in a congested state based on the size relationship between the left endpoint d1 and the right endpoint d2 of the interval (d1, d2):

[0102] If d1>d2, the road section to be inspected is not in a congested state;

[0103] If d1<d2, the road section to be inspected is in a congested state.

[0104] However, relying solely on a single vehicle inspection data type is not actually able to accurately determine the traffic congestion status of the road section to be inspected. Therefore, it is necessary to obtain data from different and representative aspects to comprehensively calculate the traffic congestion status of the road section to be inspected.

[0105] In this embodiment, the vehicle detector combined feature vector includes the first, second and third traffic feature parameters. The specific method for obtaining the vehicle detector combined feature vector is as follows:

[0106] a) Set a detection period T and extract the traffic parameters corresponding to several samples from the vehicle detector data:

[0107] The upstream location occupancy parameter s of the road section to be inspected during the detection cycle i-1 , downstream location occupancy parameter s i , upstream velocity parameter v i-1 , downstream location velocity parameter v i , and the upstream location flow parameter f i-1 , flow parameter f at downstream location i ;

[0108] b) Using the traffic parameters extracted in step a), calculate the absolute difference Δs between the upstream and downstream locations of the road section to be inspected within the detection period, and the ratio s between the absolute difference between the upstream and downstream locations and the upstream location occupancy rate of the road section to be inspected within the detection period: u , as the first traffic characteristic parameter of the road section to be inspected:

[0109]

[0110] Where Δs is the absolute difference in occupancy between upstream and downstream locations, s i-1 is the upstream location occupancy, s i is the downstream location occupancy, s u is the ratio of the absolute difference in the occupancy rates of the upstream and downstream locations to the occupancy rate of the upstream location;

[0111] c) Using the traffic parameters extracted in step a), analogy calculation is performed according to the following formula to obtain the absolute difference in speed between the upstream and downstream locations Δv, and the ratio of the absolute difference in speed between the upstream and downstream locations to the speed of the upstream location v u , as the second traffic characteristic parameter of the road section to be inspected:

[0112]

[0113] Where Δv is the absolute difference in velocity between the upstream and downstream locations, v u is the ratio of the absolute difference in velocity between the upstream and downstream locations to the velocity at the upstream location, v i-1 is the velocity parameter at the upstream location, v i is the velocity parameter at the downstream location;

[0114] d) Using the traffic parameters extracted in step a), analogy calculation is performed according to the following formula to obtain the absolute difference Δf between the upstream and downstream flow rates and the ratio f between the absolute difference between the upstream and downstream flow rates and the upstream flow rate u , as the third traffic characteristic parameter of the road section to be inspected:

[0115]

[0116] Where Δf is the absolute difference in flow rate between upstream and downstream locations, f u is the ratio of the absolute difference in flow between the upstream and downstream locations to the flow at the upstream location, f i-1 is the flow parameter of the upstream location, f i is the flow parameter of the downstream location;

[0117] e) Using the first, second and third traffic characteristic parameters of the road section to be inspected, a vehicle detector combination feature vector is formed. Each sample has a corresponding vehicle detector combination feature vector.

[0118] 1-2) Use ETC data to calculate the ETC smooth flow probability and ETC congestion probability of the current state of the road section to be inspected. The specific method is as follows:

[0119] 1-2-1) According to the traffic parameters corresponding to several historical samples in the ETC historical data of the road section to be inspected, the ETC combination feature vectors corresponding to the several historical samples are extracted, and the ETC combination feature vectors corresponding to each historical sample in the ETC historical data are clustered by using the Gaussian mixture model to establish a GMM clustering model capable of judging the congestion status of the road section;

[0120] 1-2-2) Obtain various traffic parameters corresponding to several real-time samples in the ETC real-time data, extract the ETC combined feature vectors corresponding to the several real-time samples, and input all the obtained ETC combined feature vectors into the GMM clustering model, so that the GMM clustering model outputs the ETC smooth probability and ETC congestion probability of the current state of the road section to be inspected.

[0121] Similarly, if only ETC data is used, the relationship between the ETC smooth probability and the ETC congestion probability of the current state of the road section to be inspected can also be used to preliminarily determine whether the road section to be inspected is in a congested state:

[0122] If the ETC unblocked probability of the current state of the road section to be inspected is greater than the ETC congested probability, then the road section to be inspected is not in a congested state;

[0123] If the ETC unblocked probability of the current state of the road section to be inspected is less than the ETC congested probability, the road section to be inspected is in a congested state.

[0124] However, relying solely on a single ETC data type is not actually able to accurately determine the traffic congestion status of the road section to be inspected. Therefore, it is also necessary to continue to obtain data from different and representative other aspects to comprehensively calculate the traffic congestion status of the road section to be inspected.

[0125] In this embodiment, the ETC combined feature vector includes the fourth, fifth and sixth traffic feature parameters, which are specifically obtained as follows:

[0126] ⑴ Set a detection period T and extract traffic parameters corresponding to several samples from ETC data:

[0127] The upstream flow rate f of the road section to be inspected during the detection period, and the time t when the vehicle passes the upstream ETC gantry i-1 , the time t when the vehicle passes the downstream ETC gantry i , the number of large vehicles passing through the upstream ETC gantry h, the number of small vehicles passing through the upstream ETC gantry c;

[0128] For example, in this embodiment, the upstream flow f is extracted by extracting the flow of each ETC gantry section with an interval period of T minutes;

[0129] (2) Using the traffic parameters extracted in step (1), calculate the average vehicle travel speed v of the road section to be inspected during the detection period according to the following formula as the fourth traffic characteristic parameter of the road section to be inspected:

[0130]

[0131] In the formula, is the average travel speed of vehicles on the road section to be inspected during the inspection period, t i-1 is the time when the vehicle passes the upstream ETC gantry, t i is the time when the vehicle passes the downstream ETC gantry, and f is the upstream flow rate of the road section to be inspected within the detection period;

[0132] ⑶ Using the traffic parameters extracted in step ⑴, calculate the average upstream traffic density d of the road section to be inspected during the detection period according to the following formula as the fifth traffic characteristic parameter of the road section to be inspected:

[0133]

[0134] Where d is the average traffic density upstream of the road section to be inspected during the detection period, and f is the flow rate upstream of the road section to be inspected during the detection period. The average travel speed of vehicles on the road section to be inspected during the inspection period;

[0135] (4) Using the traffic parameters extracted in step (1), calculate the ratio p of the number of large vehicles to small vehicles in the road section to be inspected during the detection period according to the following formula as the sixth traffic characteristic parameter of the road section to be inspected:

[0136]

[0137] Where p is the ratio of large vehicles to small vehicles on the road section to be inspected during the detection period, h is the number of large vehicles passing through the ETC gantry upstream of the road section to be inspected during the detection period, and c is the number of small vehicles passing through the ETC gantry upstream of the road section to be inspected during the detection period.

[0138] ⑸ Use the fourth, fifth, and sixth traffic characteristic parameters of the road section to be inspected to form an ETC combined feature vector. Each sample corresponds to such an ETC combined feature vector. After obtaining these ETC combined feature vectors, v, d, and p are introduced into the GMM (Gaussian Mixed Model) clustering model, the GMM model parameters are set, the number of clusters k=2 is determined, the maximum number of iterations and the minimum target parameters are set, and a GMM model that can be used to judge the current congestion state of the road section to be inspected is established, and then the traffic congestion state of the road section to be inspected is detected by the maximum a posteriori probability.

[0139] It is worth noting that before obtaining the various traffic parameters corresponding to several samples in the ETC data, in order to ensure the accuracy of subsequent judgments, it is also necessary to clean the data, such as matching license plates based on upstream and downstream ETC gantry information and eliminating abnormal matching data.

[0140] 1-3) Use Amap data to calculate the probability of small-scale congestion and large-scale congestion of the current state of the road section to be inspected. The specific method is as follows:

[0141] 1-3-1) Determine the congestion length threshold based on Amap’s historical data;

[0142] 1-3-2) Determine the small-scale congestion probability and the large-scale congestion probability of the current state of the road section to be inspected in the following manner:

[0143] 1-3-2-1) When the actual congestion length is less than the congestion length threshold, a comparative experiment is conducted to determine the accuracy of Amap's judgment on the congestion status of the road section to be inspected, and the judgment accuracy is used as the probability of small-scale congestion;

[0144] 1-3-2-2) When the actual congestion length is ≥ the congestion length threshold, a comparative experiment is conducted to determine the accuracy of Amap's judgment on the congestion status of the road section to be inspected, and the judgment accuracy is used as the probability of large-scale congestion.

[0145] The data provided by Amap can directly determine whether the road section to be inspected is currently congested, but this judgment is only a preliminary judgment of the road section to be inspected by navigation software such as Amap. Due to the influence of various factors, the results judged only by using this type of data may deviate from the actual situation. Therefore, when using this type of data, it is actually impossible to accurately judge the traffic congestion status of the road section to be inspected to the maximum extent. Through comparative experiments, we calculated the judgment accuracy of navigation software such as Amap in different situations of the road section to be inspected in advance, and used these judgment accuracy rates to participate in comprehensive judgment, so as to maximize the final judgment accuracy of the congestion status of the road section to be inspected.

[0146] For example, according to Amap's historical data, the congestion length threshold of the road section to be inspected is 1.5 km. After comparative experiments:

[0147] If the actual congestion length of the traffic is less than 1.5 km, the accuracy of Amap's judgment on the congestion status of the road section to be inspected is 75%, and the probability of small-scale congestion is 0.75;

[0148] If the actual congestion length of the traffic flow is ≥1.5km, the accuracy rate of Amap's judgment on the congestion status of the road section to be inspected is 95%, and the probability of large-scale congestion is 0.95.

[0149] It is worth noting that the reason why the congestion length threshold of the road section to be inspected is set at 1.5km according to Gaode's historical data is that based on the historical data analysis of the road section to be inspected, it can be found that 1.5km is a dividing point for distinguishing the congestion conditions of this road section to be inspected, which can more clearly distinguish between minor congestion and severe congestion.

[0150] 1-4) The fuzzy comprehensive evaluation method is used to perform decision-level fusion on the unobstructed probability and / or congested probability of the road section to be inspected obtained from the vehicle inspection device data, ETC data, and Amap data, and calculate the comprehensive unobstructed probability and comprehensive congested probability of the road section to be inspected. The specific method is as follows:

[0151] 1-4-1) Through experiments, the historical accuracy of the vehicle inspection data, ETC data, and Amap data on the congestion status of the road section to be inspected is calculated respectively, and used as the initial factor weights of the corresponding data;

[0152] In other words, we first need to conduct comparative experiments, compare the theoretical judgment results obtained using various types of data with the actual observation results one by one, and then calculate the accuracy corresponding to each theoretical judgment result as the historical judgment accuracy.

[0153] It is worth noting that the initial factor weights of these vehicle inspection device data, ETC data, and Amap data were obtained through experiments before the implementation of the present invention.

[0154] In this embodiment, the traffic characteristic parameters Δs and s obtained based on the vehicle detector data in the same detection period T of the highway section to be inspected are firstly extracted. u , Δv, v u , Δf, f u , traffic characteristic parameters obtained based on ETC data d, p and traffic characteristic parameters (i.e., congestion length threshold) obtained based on Amap data are used to perform congestion detection respectively, and the congestion detection results of each data source and the historical congestion discrimination rate are obtained as follows:

[0155] As shown in Table 1.

[0156]

[0157] Table 1

[0158] In Table 1, the historical congestion discrimination rate corresponding to each data source is calculated as follows:

[0159]

[0160]

[0161] Where a′1 is the historical congestion discrimination rate corresponding to the vehicle detector data, a′2 is the historical congestion discrimination rate corresponding to the ETC data, a′3 is the historical congestion discrimination rate corresponding to the Amap data, and Z A Z is the number of times when the theoretical judgment result and the actual observation result are consistent when the vehicle inspection device data are used to judge the congestion state of the road section to be inspected, B Z is the number of times when the theoretical judgment result is consistent with the actual observation result when the ETC data is used to judge the congestion status of the road section to be inspected. C G is the number of times that the theoretical judgment result is consistent with the actual observation result when using Amap data to judge the congestion status of the road section to be inspected. A G is the number of times when the theoretical judgment result is inconsistent with the actual observation result when the vehicle inspection device data is used to judge the congestion state of the road section to be inspected. B G is the number of times when the theoretical judgment result is inconsistent with the actual observation result when using ETC data to judge the congestion status of the road section to be inspected. C The number of times when the theoretical judgment result is inconsistent with the actual observation result when using Amap data to judge the congestion status of the road section to be inspected;

[0162] 1-4-2) Use the initial factor weights of vehicle inspection data, ETC data, and Amap data to form a factor weight vector;

[0163] In this embodiment, the initial factor weight of the vehicle detector data is a′1, the initial factor weight of the ETC data is a′2, and the initial factor weight of the Amap data is a′3;

[0164] Therefore, the initial factor weight vector is A′=(a′1,a′2,a′3);

[0165] 1-4-3) Normalize the factor weight vector to obtain the normalized factor weight vector A = (a1, a2, a3), and a1 + a2 + a3 = 1;

[0166] 1-4-4) A congestion detection result evaluation matrix is ​​established based on the vehicle inspection device real-time data, ETC real-time data, and Amap real-time data, as shown below:

[0167]

[0168] Where R is the congestion detection result evaluation matrix, R 11 is the probability of vehicle inspection device being unblocked, R 12 is the vehicle detector congestion probability, R 21 is the ETC smooth flow probability, R 22 is the ETC congestion probability, R 31 is the probability of small-scale congestion, R 32 is the probability of large-scale congestion;

[0169] Among them, R 11 +R 12 =1, R 21 +R 22 =1, R 31 +R 32 =1;

[0170] In fact, R 11 , R 12 , R 21 , R 22 , R 31 , R 32 It represents the degree of membership when the congestion detection results are comprehensively calculated using the fuzzy algorithm.

[0171] 1-4-5) Perform fuzzy synthesis calculation according to the following formula to obtain the congestion fusion detection result:

[0172] B=A·R=(b1,b2)

[0173] Where B is the congestion fusion detection result, A is the factor weight vector after normalization, R is the congestion detection result evaluation matrix, and “·” represents the fuzzy synthesis algorithm.

[0174] In this embodiment, A=(a1, a2, a3), and Substitute into the above formula, that is:

[0175]

[0176] B=(a1R 11 +a2R 21 +a3R 31 ,a1R 12 +a2R 22 +a3R 32 )=(b1,b2)

[0177] Finally, according to the relationship between the comprehensive unblocked probability b1 and the comprehensive congested probability b2 of the road section to be inspected, the congestion state of the road section to be inspected is determined in the following manner:

[0178] If the comprehensive unblocked probability b1 of the road section to be inspected is less than the comprehensive congested probability b2, the current road section to be inspected is in a congested state;

[0179] If the comprehensive unblocked probability b1 of the road section to be inspected is greater than the comprehensive congested probability b2, the current road section to be inspected is in a non-congested state.

[0180] Of course, it is also possible to use traffic data from different channels to calculate the probability of smooth traffic and / or congestion of the road section to be inspected in advance, and judge the congestion status of the road section to be inspected. If the congestion status determined is consistent, there is no need to perform decision-level fusion, that is, there is no need to calculate the comprehensive smooth traffic probability and the comprehensive congestion probability, and the judgment result can be directly output to save computing power.

[0181] In summary, the present invention uses a fuzzy comprehensive evaluation algorithm to determine the evaluation set domain U = {congestion detection result A based on vehicle inspection data, congestion detection result B based on ETC data, congestion detection result C based on AutoNavi data}; determine the highway traffic status evaluation level set domain V = {no congestion, congestion}; determine the factor weight vector A′ = {a′1, a′2, a′3}, a′1, a′2, a′3 are all historical congestion discrimination rates of each data source; normalize the factor weight vector to obtain the normalized factor weight vector A = (a1, a2, a3); determine the evaluation matrix R according to the membership of the congestion detection result. Finally, perform fuzzy operation, and the operation result is B = A·R = (b1, b2). When b1 < b2, the traffic state is congested.

[0182] It can be seen that compared with the traditional single data source detection method, the present invention fully considers the effectiveness of heterogeneous data sources, constructs targeted detection models respectively and integrates them innovatively, gives play to the redundancy and complementarity of each source detection data in space and time, improves the consistency of traffic status perception, and significantly enhances the stability and credibility of congestion detection results by reducing information conflicts and uncertainties.

[0183] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification made to the present invention by those skilled in the art without departing from the spirit of the present invention shall fall within the protection scope of the present invention.

Claims

1. A highway congestion detection method based on multi-source data fusion, characterized in that: The following steps are involved: 1) Using various traffic data obtained from different channels to detect the congestion status of the road section to be inspected, and obtaining the unblocked probability and / or congested probability of the road section to be inspected calculated based on the traffic data from different channels; 2) Using fuzzy comprehensive evaluation method to perform decision-level fusion on the unblocked probability and / or congested probability of the road section to be inspected calculated based on traffic data from different channels, and calculate the comprehensive unblocked probability and comprehensive congested probability of the road section to be inspected; 3) The congestion status of the road section to be inspected is determined based on the relationship between the comprehensive unblocked probability and the comprehensive congested probability of the road section to be inspected.

2. The highway congestion detection method according to claim 1, characterized in that: In step 3), according to the relationship between the comprehensive unblocked probability and the comprehensive congested probability of the road section to be inspected, the specific method of judging the congested state of the road section to be inspected is as follows: If the comprehensive unblocked probability of the road section to be inspected is less than the comprehensive congested probability, the current road section to be inspected is in a congested state; If the comprehensive unblocked probability of the road section to be inspected is greater than the comprehensive congested probability, the current road section to be inspected is in a non-congested state.

3. The highway congestion detection method according to claim 1, characterized in that: The various types of traffic data obtained from different channels include vehicle inspection data, ETC data, and Amap data. The specific steps for calculating the comprehensive smoothness probability and the comprehensive congestion probability of the road section to be inspected include: 1-1) Using the vehicle detector data, calculate the vehicle detector unblocked probability and the vehicle detector congestion probability of the current state of the road section to be inspected; 1-2) Using ETC data, calculate the ETC smooth flow probability and ETC congestion probability of the current state of the road section to be inspected; 1-3) Use Amap data to calculate the probability of small-scale congestion and large-scale congestion in the current state of the road section to be inspected; 1-4) A fuzzy comprehensive evaluation method is used to perform decision-level fusion on the unobstructed probability and / or congested probability of the road section to be inspected obtained based on the vehicle inspection device data, ETC data, and Amap data, and calculate the comprehensive unobstructed probability and comprehensive congested probability of the road section to be inspected.

4. The method for detecting highway congestion status according to claim 3, characterized in that: In step 1-4), the specific method of calculating the comprehensive unblocked probability and the comprehensive congested probability of the road section to be inspected is as follows: 1-4-1) Through experiments, the accuracy of the vehicle inspection data, ETC data, and Amap data in determining the current congestion status of the road section to be inspected is calculated respectively, and used as the initial factor weights of the corresponding data; 1-4-2) Use the initial factor weights of vehicle inspection data, ETC data, and Amap data to form a factor weight vector; 1-4-3) Normalizing the factor weight vector to obtain a normalized factor weight vector; 1-4-4) A congestion detection result evaluation matrix is ​​established based on the vehicle inspection device real-time data, ETC real-time data, and Amap real-time data, as shown below: Where R is the congestion detection result evaluation matrix, R 11 is the probability of vehicle inspection device being unblocked, R 12 is the vehicle detector congestion probability, R 21 is the ETC smooth flow probability, R 22 is the ETC congestion probability, R 31 is the probability of small-scale congestion, R 32 is the probability of large-scale congestion; 1-4-5) Perform fuzzy synthesis calculation according to the following formula to obtain the congestion fusion detection result: B=A·R=(b1,b2) Where B is the congestion fusion detection result, A is the factor weight vector after normalization, R is the congestion detection result evaluation matrix, and "·" represents the fuzzy synthesis algorithm.

5. The highway congestion detection method according to claim 3, characterized in that: In step 1-1), the specific method of using the vehicle detector data to calculate the vehicle detector unblocked probability and the vehicle detector congestion probability of the current state of the road section to be inspected is as follows: 1-1-1) According to the traffic parameters corresponding to several historical samples in the historical data of the vehicle detector, the combined feature vectors of the vehicle detector corresponding to the several historical samples are extracted, and the combined feature vectors of the vehicle detector corresponding to each historical sample in the historical data of the vehicle detector are clustered and analyzed using the fuzzy C-means clustering algorithm, and the number of clusters is determined to establish an FCM clustering model capable of judging the congestion status of a road section; 1-1-2) According to the traffic parameters corresponding to several real-time samples in the vehicle detector real-time data, the vehicle detector combination feature vectors corresponding to several real-time samples are extracted and input into the FCM clustering model to obtain the vehicle detector unblocked probability and the vehicle detector congestion probability of the current state of the road section to be inspected.

6. The highway congestion detection method according to claim 5, characterized in that: In step 1-1-2), the specific method of obtaining the vehicle detector unblocked probability and the vehicle detector congested probability of the current state of the road section to be inspected is as follows: 1-1-2-1) Obtain various traffic parameters corresponding to several real-time samples in the vehicle detector real-time data, extract the vehicle detector combination feature vectors corresponding to the several real-time samples, and input all the obtained vehicle detector combination feature vectors into the FCM clustering model to obtain the clustering center of various traffic feature parameters in all vehicle detector combination feature vectors; 1-1-2-2) Calculate the Euclidean distance between each type of traffic characteristic parameter in each vehicle detector combination feature vector and the corresponding cluster center to form a Euclidean distance data set; 1-1-2-3) Normalize the data in the Euclidean distance data set and map the data in the Euclidean distance data set to the interval (d1, d2), and d1+d2=1; 1-1-2-4) The left endpoint d1 of the interval (d1, d2) is taken as the vehicle detector unblocked probability, and the right endpoint d2 of the interval (d1, d2) is taken as the vehicle detector congested probability.

7. The highway congestion state detection method according to claim 5 or 6, characterized in that: The vehicle detector combined feature vector includes the first, second and third traffic feature parameters. The specific method for obtaining the vehicle detector combined feature vector is as follows: a) Set a detection period T and extract the traffic parameters corresponding to several samples from the vehicle detector data: The upstream location occupancy parameter s of the road section to be inspected during the detection cycle i-1 , downstream location occupancy parameter s i , upstream velocity parameter v i-1 , downstream location velocity parameter v i , and the upstream location flow parameter f i-1 , flow parameter f at downstream location i ; b) Using the traffic parameters extracted in step a), calculate the absolute difference Δs between the upstream and downstream locations of the road section to be inspected within the detection period, and the ratio s between the absolute difference between the upstream and downstream locations and the upstream location occupancy rate of the road section to be inspected within the detection period: u , as the first traffic characteristic parameter of the road section to be inspected: Where Δs is the absolute difference in occupancy between upstream and downstream locations, s i-1 is the upstream location occupancy, s i is the downstream location occupancy, s u is the ratio of the absolute difference in the occupancy rates of the upstream and downstream locations to the occupancy rate of the upstream location; c) Using the traffic parameters extracted in step a), analogy calculation is performed according to the following formula to obtain the absolute difference in speed between the upstream and downstream locations Δv, and the ratio of the absolute difference in speed between the upstream and downstream locations to the speed of the upstream location v u , as the second traffic characteristic parameter of the road section to be inspected: Where Δv is the absolute difference in velocity between the upstream and downstream locations, v u is the ratio of the absolute difference in velocity between the upstream and downstream locations to the velocity at the upstream location, v i-1 is the velocity parameter at the upstream location, v i is the velocity parameter at the downstream location; d) Using the traffic parameters extracted in step a), analogy calculation is performed according to the following formula to obtain the absolute difference Δf between the upstream and downstream flow rates and the ratio f between the absolute difference between the upstream and downstream flow rates and the upstream flow rate u , as the third traffic characteristic parameter of the road section to be inspected: Where Δf is the absolute difference in flow rate between upstream and downstream locations, f u is the ratio of the absolute difference in flow between the upstream and downstream locations to the flow at the upstream location, f i-1 is the flow parameter of the upstream location, f i is the flow parameter of the downstream location; e) The first, second and third traffic characteristic parameters of the road section to be inspected are combined to form a vehicle detector combined characteristic vector.

8. The highway congestion detection method according to claim 3, characterized in that: In step 1-2), the specific method of using ETC data to calculate the ETC smooth flow probability of the current state of the road section to be inspected and the ETC congestion probability is as follows: 1-2-1) According to the traffic parameters corresponding to several historical samples in the ETC historical data of the road section to be inspected, the ETC combination feature vectors corresponding to the several historical samples are extracted, and the ETC combination feature vectors corresponding to each historical sample in the ETC historical data are clustered by using the Gaussian mixture model to establish a GMM clustering model capable of judging the congestion status of the road section; 1-2-2) Obtain various traffic parameters corresponding to several real-time samples in the ETC real-time data, extract the ETC combined feature vectors corresponding to the several real-time samples, and input all the obtained ETC combined feature vectors into the GMM clustering model, so that the GMM clustering model outputs the ETC smooth probability and ETC congestion probability of the current state of the road section to be inspected.

9. The method for detecting highway congestion status according to claim 8, characterized in that: The ETC combined feature vector includes the fourth, fifth and sixth traffic feature parameters, and the specific acquisition method is as follows: ⑴ Set a detection period T and extract traffic parameters corresponding to several samples from ETC data: The upstream flow rate f of the road section to be inspected during the detection period, and the time t when the vehicle passes the upstream ETC gantry i-1 , the time t when the vehicle passes the downstream ETC gantry i , the number of large vehicles passing through the upstream ETC gantry h, the number of small vehicles passing through the upstream ETC gantry c; ⑵ Using the traffic parameters extracted in step ⑴, calculate the average travel speed of vehicles on the road section to be inspected during the inspection period according to the following formula: As the fourth traffic characteristic parameter of the road section to be inspected: In the formula, is the average travel speed of vehicles on the road section to be inspected during the inspection period, t i-1 is the time when the vehicle passes the upstream ETC gantry, t i is the time when the vehicle passes the downstream ETC gantry, and f is the upstream flow rate of the road section to be inspected within the detection period; ⑶ Using the traffic parameters extracted in step ⑴, calculate the average upstream traffic density d of the road section to be inspected during the detection period according to the following formula as the fifth traffic characteristic parameter of the road section to be inspected: Where d is the average traffic density upstream of the road section to be inspected during the detection period, and f is the flow rate upstream of the road section to be inspected during the detection period. The average travel speed of vehicles on the road section to be inspected during the inspection period; (4) Using the traffic parameters extracted in step (1), calculate the ratio p of the number of large vehicles to small vehicles in the road section to be inspected during the detection period according to the following formula as the sixth traffic characteristic parameter of the road section to be inspected: Where p is the ratio of large vehicles to small vehicles on the road section to be inspected during the detection period, h is the number of large vehicles passing through the ETC gantry upstream of the road section to be inspected during the detection period, and c is the number of small vehicles passing through the ETC gantry upstream of the road section to be inspected during the detection period. (5) Use the fourth, fifth and sixth traffic characteristic parameters of the road section to be inspected to form an ETC combined feature vector.

10. The highway congestion state detection method according to claim 3, characterized in that: In step 1-3), the specific method of using Amap data to calculate the probability of small-scale congestion and the probability of large-scale congestion of the current state of the road section to be inspected is as follows: 1-3-1) Determine the congestion length threshold based on Amap’s historical data; 1-3-2) Determine the small-scale congestion probability and the large-scale congestion probability of the current state of the road section to be inspected in the following manner: 1-3-2-1) When the actual congestion length is less than the congestion length threshold, a comparative experiment is conducted to determine the accuracy of Amap's judgment on the congestion status of the road section to be inspected, and the judgment accuracy is used as the probability of small-scale congestion; 1-3-2-2) When the actual congestion length is ≥ the congestion length threshold, a comparative experiment is conducted to determine the accuracy of Amap's judgment on the congestion status of the road section to be inspected, and the judgment accuracy is used as the probability of large-scale congestion.

Images