Expressway abnormal state detection method and system based on support vector machine

Abstract

The invention relates to the technical field of expressway monitoring, and provides an expressway abnormal state detection method and system based on a support vector machine, which processes vehicle trajectory data and ETC gantry data through Kalman filtering and adaptive weighted average to improve the data accuracy, introduces a Gaussian kernel function to construct an SVM model, and improves the detection accuracy of the expressway abnormal state. An original feature space is mapped to a high-dimensional space to deal with a nonlinear problem, samples with labels are used to train a multi-classification SVM model, and accurate detection of various abnormal states such as traffic congestion, accidents and vehicle faults is realized through a voting mechanism. According to the method, the accuracy of abnormal state detection is improved, powerful decision support is provided for expressway managers, the traffic flow is optimized, the road traffic capacity is improved, and the development of an intelligent traffic system is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of highway monitoring, and in particular to a method and system for detecting abnormal states of highways based on a support vector machine. Background Art

[0002] The abnormal state of the highway is mainly caused by excessive traffic, traffic accidents, road and bridge collapse, vehicle breakdown and other events. Since the acquisition area of ​​hardware equipment such as video is limited and cannot fully cover the entire section of the highway, it is difficult for road network monitoring personnel to obtain the type and location information of the abnormal state in real time when an abnormal state occurs. However, with the coverage of other acquisition equipment on the highway, it is possible to monitor the operation status of the highway network in real time. For example, the gantry data can interact with ETC to obtain data such as the lane and speed of the vehicle. At the same time, the total amount of data entering and leaving the highway entrance and exit toll station can be obtained through the toll data. In addition, for two passenger and one dangerous goods and truck vehicles, the vehicle's running trajectory data can be obtained in real time. Therefore, this study uses the above-mentioned highway multi-source monitoring data and the support vector machine model to realize the recognition of the abnormal state of the highway, solve the problem that the video detector cannot fully cover, and use this method to realize the reasoning and recognition of the occurrence time, type, and spread length of the abnormal state of the road network.

[0003] The existing technology has the following technical pain points. Aiming at the abnormal state of highways caused by unexpected events, a highway abnormal state detection scheme based on the support vector machine model is proposed. First, the vehicle breakdown accident model of the basic section of the one-way three-lane of Kunyu Expressway is established using VISSIM simulation software; abnormal events under different conditions are simulated, and the impact of abnormal states on the traffic capacity of the highway section is analyzed; then, a highway abnormal state detection model is established based on the support vector machine model, and the difference between single-sided and double-sided input is analyzed. The simulation results show that the established support vector machine abnormal state detection model has good comprehensive performance, and single-sided input can also detect the occurrence of abnormal states well.

[0004] The vehicle abnormal behavior detection method and system based on trajectory data analyze the distribution characteristics of vehicle trajectory data under different spatiotemporal characteristics. In order to avoid the disadvantage that the abnormal behavior data screening method with a predetermined threshold ignores the internal information of the data, the invention regards the data that does not conform to the characteristics of most data as abnormal behavior data, and proposes an abnormal behavior determination and screening method for vehicle micro-trajectory data. In combination with supervised machine learning methods, a data diversion module and a comprehensive processing module are introduced on the basis of the traditional support vector machine model, and an abnormal behavior detection method for micro-trajectory data is proposed. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method and system for detecting abnormal states of highways based on support vector machines, which solves the problem that the existing methods and systems for detecting abnormal states of highways can be divided into abnormal detection methods based on macro traffic flow and abnormal detection methods based on micro vehicle trajectories. The macro traffic abnormality detection and identification is mainly based on the basic diagram of traffic flow, and the abnormal state is identified by the time mutation points of the basic diagram. The method of micro vehicle trajectory is mainly for the GPS trajectory points of the vehicle, and the vehicle speed, vehicle acceleration, and headway spacing are selected as features for model training and calibration. The macro traffic map can identify the abnormal time at a certain detector position, but there is a certain lag, and the abnormal state recognition result has a large deviation when the vehicle flow is small; the micro-based recognition method needs to rely on the vehicle GPS, and the data accessed by the vehicles traveling on the highway only includes the trajectory data of two passenger vehicles, one dangerous goods vehicle, and some trucks. The proportion of vehicles in the traffic flow is low, and the recognition result is inaccurate.

[0006] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0007] The present invention provides a method for detecting abnormal state of a highway based on a support vector machine, comprising:

[0008] Step S101, obtaining vehicle GPS trajectory data and ETC gantry data on the expressway, preprocessing the vehicle GPS trajectory data and ETC gantry data on the expressway to obtain preprocessed vehicle GPS trajectory data and ETC gantry data on the expressway, preprocessing the vehicle GPS trajectory data and ETC gantry data on the expressway includes identifying and removing breakpoints in the vehicle trajectory data, projecting the trajectory point data onto the expressway network data using a map matching algorithm, sorting the ETC gantry data by transaction time to generate vehicle transaction gantry trajectory data, and removing data in an erroneous sequence;

[0009] Step S102: Associating the pre-processed vehicle GPS trajectory data and ETC gantry data on the highway according to the license plate number, time, and location, and identifying a set of vehicles using GPS and ETC in the same time period;

[0010] Step S103: performing Kalman filtering on the trajectory data and the ETC gantry data, and performing adaptive weighted averaging on the data after the Kalman filtering;

[0011] Step S104: setting a confidence range and a classification threshold, and performing speed fusion according to the confidence. The speed fusion according to the confidence includes classifying the speed fusion of the road section into three categories according to different confidences, performing speed fusion by weighted average of GPS and ETC data, obtaining a fusion result, and using data with high confidence to estimate the speed in the fusion result;

[0012] Step S105: Introduce the Gaussian kernel function to construct a multi-classification support vector machine model, and determine the final classification result through a voting mechanism, determine the input sample, extract features from the input sample, calculate through a binary classification support vector machine, and use a voting mechanism to determine the category to obtain the detection result of the abnormal state of the highway.

[0013] Furthermore, in the method for detecting abnormal state of a highway based on a support vector machine according to the present invention, step S103 comprises:

[0014] Kalman filter initialization is used to initialize the Kalman filter for each vehicle trajectory and ETC gantry data sequence, setting the initial state and covariance matrix;

[0015] Kalman prediction step, using the vehicle dynamics model to predict the state at the next time point, the state at the time point includes position and speed;

[0016] When new GPS track points or ETC gantry data are received, the new GPS track points or ETC gantry data are used as observations to update the state estimate of the Kalman filter;

[0017] Use the observation value and Kalman gain to update the state estimate and covariance matrix, perform adaptive weighted averaging on the data processed by Kalman filtering according to the covariance matrix, output the filtered and weighted data, and output the vehicle trajectory data and ETC gantry data after Kalman filtering and adaptive weighted averaging.

[0018] Furthermore, in the method for detecting abnormal state of a highway based on a support vector machine according to the present invention, step S104 comprises:

[0019] Divide the confidence into high confidence, medium confidence and low confidence;

[0020] For each GPS track point and ETC gantry data point, the confidence of each GPS track point and ETC gantry data point is calculated to obtain a set of GPS and ETC data point confidences;

[0021] The data points are divided into three categories according to the confidence level, including high confidence data, medium confidence data and low confidence data, and the fusion result is obtained;

[0022] Check the confidence of each data point, select the data point or data point set with the highest confidence as the final velocity estimate, and output the velocity fusion result.

[0023] Furthermore, in the method for detecting abnormal state of a highway based on a support vector machine according to the present invention, step S105 comprises:

[0024] Determine the input sample. The input sample is the feature representation of the vehicle trajectory data and ETC gantry data processed in the previous steps. Label the input sample, that is, mark the normal state and abnormal state;

[0025] The Gaussian kernel function is introduced to construct the support vector machine model. The Gaussian kernel function maps the original feature space to a high-dimensional space.

[0026] Use labeled input samples to train a binary SVM model. For normal or abnormal states, train the SVM model to distinguish between categories.

[0027] The abnormal status of highways includes traffic congestion type, accident type and vehicle failure type. A binary classification SVM is trained for each pair of categories, and then the final classification result is determined through a voting mechanism to obtain the detection result of the abnormal status of highways:

[0028] Based on the final classification result, each input sample is labeled as normal or abnormal.

[0029] In a second aspect, the present invention provides a highway abnormal state detection system based on a support vector machine, which is applied to a highway abnormal state detection method based on a support vector machine, comprising:

[0030] A data acquisition module is used to acquire vehicle GPS trajectory data and ETC gantry data on the expressway, pre-process the vehicle GPS trajectory data and ETC gantry data on the expressway to obtain pre-processed vehicle GPS trajectory data and ETC gantry data on the expressway, and pre-process the vehicle GPS trajectory data and ETC gantry data on the expressway, including identifying and removing breakpoints in the vehicle trajectory data, projecting the trajectory point data onto the expressway network data using a map matching algorithm, sorting the ETC gantry data by transaction time to generate vehicle transaction gantry trajectory data, and removing data in an erroneous sequence;

[0031] Data association module: used to associate the pre-processed vehicle GPS trajectory data and ETC gantry data on the highway according to the license plate number, time and location, and identify the set of vehicles using GPS and ETC in the same time period;

[0032] Data processing module: used to perform Kalman filtering on trajectory data and ETC gantry data, and perform adaptive weighted averaging on the data after Kalman filtering;

[0033] Data analysis module: used to set the confidence range and classification threshold, and perform speed fusion according to the confidence. Speed ​​fusion according to the confidence includes classifying the speed fusion of the road section into three categories according to different confidence levels, and performing speed fusion by weighted average of GPS and ETC data to obtain the fusion result. In the fusion result, the data with high confidence is used for speed estimation.

[0034] Result output module: used to introduce Gaussian kernel function to build a multi-classification support vector machine model, and determine the final classification result through a voting mechanism, determine the input sample, extract features from the input sample, calculate through a binary classification support vector machine, and use a voting mechanism to determine the category to obtain the detection result of the abnormal state of the highway.

[0035] Furthermore, in the highway abnormal state detection system based on support vector machine described in the present invention, the data processing module is also used for:

[0036] Kalman filter initialization is used to initialize the Kalman filter for each vehicle trajectory and ETC gantry data sequence, setting the initial state and covariance matrix;

[0037] Kalman prediction step, using the vehicle dynamics model to predict the state at the next time point, the state at the time point includes position and speed;

[0038] When new GPS track points or ETC gantry data are received, the new GPS track points or ETC gantry data are used as observations to update the state estimate of the Kalman filter;

[0039] Use the observation value and Kalman gain to update the state estimate and covariance matrix, perform adaptive weighted averaging on the data processed by Kalman filtering according to the covariance matrix, output the filtered and weighted data, and output the vehicle trajectory data and ETC gantry data after Kalman filtering and adaptive weighted averaging.

[0040] Furthermore, in the highway abnormal state detection system based on support vector machine described in the present invention, the data analysis module is also used for:

[0041] Divide the confidence into high confidence, medium confidence and low confidence;

[0042] For each GPS track point and ETC gantry data point, the confidence of each GPS track point and ETC gantry data point is calculated to obtain a set of GPS and ETC data point confidences;

[0043] The data points are divided into three categories according to the confidence level, including high confidence data, medium confidence data and low confidence data, and the fusion result is obtained;

[0044] Check the confidence of each data point, select the data point or data point set with the highest confidence as the final velocity estimate, and output the velocity fusion result.

[0045] Furthermore, in the highway abnormal state detection system based on support vector machine described in the present invention, the result output module is also used for:

[0046] Determine the input sample. The input sample is the feature representation of the vehicle trajectory data and ETC gantry data processed in the previous steps. Label the input sample, that is, mark the normal state and abnormal state;

[0047] The Gaussian kernel function is introduced to construct the support vector machine model. The Gaussian kernel function maps the original feature space to a high-dimensional space.

[0048] Use labeled input samples to train a binary SVM model. For normal or abnormal states, train the SVM model to distinguish between categories.

[0049] The abnormal status of highways includes traffic congestion type, accident type and vehicle failure type. A binary classification SVM is trained for each pair of categories, and then the final classification result is determined through a voting mechanism to obtain the detection result of the abnormal status of highways:

[0050] Based on the final classification result, each input sample is labeled as normal or abnormal.

[0051] Beneficial effects of the present invention:

[0052] By combining Kalman filtering and adaptive weighted average processing, the present invention can more accurately estimate the speed and position of the vehicle and reduce the influence of data noise and error. Furthermore, in the training process of the support vector machine (SVM) model, using these high-quality data can significantly improve the accuracy of abnormal state detection.

[0053] The present invention uses Gaussian kernel function to map the original feature space to high-dimensional space, so that the SVM model can handle nonlinear problems and better adapt to changes in different highway environments and traffic conditions. By constructing a multi-classification SVM model and using a voting mechanism, the present invention can simultaneously detect multiple types of highway abnormal conditions including traffic congestion, accidents and vehicle failures.

[0054] The method of the present invention has advantages in processing speed and efficiency, and can realize real-time detection of abnormal conditions of highways. This is crucial for timely response and handling of abnormal situations, and helps to reduce traffic delays and safety hazards. By accurately and in real time detecting abnormal conditions of highways, the present invention provides powerful decision support for highway managers. This helps managers better understand road conditions, optimize traffic flow, improve road capacity, and improve the overall highway management level and service quality.

[0055] The successful application of the present invention will promote the development of the entire intelligent transportation system. By combining with other intelligent transportation technologies (such as vehicle networking, traffic signal control, etc.), it can further realize the intelligent and automated management of highways and improve traffic efficiency.

[0056] In summary, the highway abnormal state detection method based on support vector machine described in the present invention shows significant beneficial effects in improving detection accuracy, enhancing model generalization ability, realizing multi-type abnormal state detection, improving real-time performance, enhancing highway management level and promoting the development of intelligent transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0058] Figure 1 A schematic flow chart of a method for detecting abnormal state of a highway based on a support vector machine is provided in the present invention.

[0059] Figure 2 A schematic diagram of the classification form of abnormal data and normal data of the gantry trajectory data provided by the present invention.

[0060] Figure 3 This is a schematic diagram of voting classification of samples to be identified provided by the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. 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. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.

[0062] The present invention provides a method for detecting abnormal state of a highway based on a support vector machine, comprising:

[0063] Step S101, obtaining vehicle GPS trajectory data and ETC gantry data on the expressway, preprocessing the vehicle GPS trajectory data and ETC gantry data on the expressway to obtain preprocessed vehicle GPS trajectory data and ETC gantry data on the expressway, preprocessing the vehicle GPS trajectory data and ETC gantry data on the expressway includes identifying and removing breakpoints in the vehicle trajectory data, projecting the trajectory point data onto the expressway network data using a map matching algorithm, sorting the ETC gantry data by transaction time to generate vehicle transaction gantry trajectory data, and removing data in an erroneous sequence;

[0064] Real-time data is obtained from vehicle GPS devices and ETC gantry devices on the highway. Vehicle GPS trajectory data contains information such as the vehicle's location, speed, and time; ETC gantry data contains the speed, time, and location information of the vehicle passing through the gantry.

[0065] Check the vehicle's trajectory data, identify and remove breakpoints in the trajectory, and ensure the continuity and integrity of the data.

[0066] A map matching algorithm is applied to project the trajectory point data onto the highway network data to correct the position deviation in the data.

[0067] Sort the ETC gantry data according to transaction time to generate vehicle transaction gantry trajectory data.

[0068] Error sequence elimination: Check the ETC gantry data and eliminate missing, duplicate or abnormal sequence data caused by collection errors.

[0069] Get the original vehicle GPS track data and ETC gantry data from the data source. These data may come from multiple different devices or systems, so they need to be integrated and cleaned. Check and remove breakpoints in the GPS track data to ensure the continuity of the track. This step can be done automatically by the algorithm or combined with manual inspection. Use the map matching algorithm to accurately map the GPS track point data to the highway network map.

[0070] Sort the ETC gantry data by transaction time and remove any data in the wrong sequence. This step can be done automatically by data cleaning and filtering tools. After the above preprocessing steps, the preprocessed vehicle GPS trajectory data and ETC gantry data on the highway are obtained.

[0071] Through the above method and process, the quality and accuracy of the acquired vehicle GPS trajectory data and ETC gantry data can be ensured.

[0072] Step S102: Associating the pre-processed vehicle GPS trajectory data and ETC gantry data on the highway according to the license plate number, time, and location, and identifying a set of vehicles using GPS and ETC in the same time period;

[0073] After completing the data preprocessing of step S101, the cleaned vehicle GPS trajectory data and ETC gantry data on the highway are obtained. Next, the goal of step S102 is to associate these data according to license plate number, time, and location, and identify the set of vehicles that use both GPS and ETC in the same time period.

[0074] The vehicle information in the GPS track data and the ETC gantry data is matched by the common identifier of the license plate number. For each GPS track record, try to find a record with the same license plate number in the ETC gantry data.

[0075] Based on the license plate number matching, the data is further associated through time information. The timestamp of the GPS track record and the transaction time of the ETC gantry data are compared to ensure that they occurred in a similar or identical time period.

[0076] Verify data relevance through location information. Check whether the location points recorded by the GPS track are close to or consistent with the gantry locations in the ETC gantry data. This helps confirm that the vehicle actually passed the ETC gantry at a specific time and place.

[0077] After completing the data association, a comprehensive dataset containing vehicle GPS tracks and ETC gantry data will be obtained. Next, traverse this dataset to identify vehicles that have both GPS track records and ETC gantry records in the same time period. Mark the vehicle as "vehicles using both GPS and ETC" and organize the relevant data (such as license plate number, timestamp, location information, etc.) into a vehicle set.

[0078] Through the processing of step S102, the set of vehicles that use both GPS and ETC in the same time period is accurately associated and identified.

[0079] Step S103: performing Kalman filtering on the trajectory data and the ETC gantry data, and performing adaptive weighted averaging on the data after the Kalman filtering;

[0080] The trajectory data and ETC gantry data are processed by Kalman filtering. Kalman filtering is a recursive algorithm that can estimate the state of the system through a series of measurement data in the presence of noise and other inaccuracies.

[0081] For trajectory data, Kalman filtering can help smooth the vehicle's driving trajectory and remove trajectory jitter or deviation caused by signal interference or equipment errors. Through Kalman filtering, more accurate and continuous vehicle position information can be obtained.

[0082] For ETC gantry data, Kalman filtering can also help remove noise and outliers in the data and improve the accuracy and reliability of the data.

[0083] After the Kalman filter processing is completed, more accurate and reliable trajectory data and ETC gantry data are obtained. Adaptive weighted averaging is a method of assigning weights based on the importance and reliability of data. The weights are determined based on factors such as the source, quality, and timestamp of the data. For example, more accurate and recent data are given a higher weight; less accurate or old data are given a lower weight.

[0084] Through adaptive weighted average processing, the trajectory data and ETC gantry data are effectively integrated to obtain a more comprehensive and accurate description of the vehicle's driving status.

[0085] In summary, step S103 further improves the accuracy of the data by performing Kalman filtering and adaptive weighted averaging on the trajectory data and ETC gantry data.

[0086] Step S104: setting a confidence range and a classification threshold, and performing speed fusion according to the confidence. The speed fusion according to the confidence includes classifying the speed fusion of the road section into three categories according to different confidences, performing speed fusion by weighted average of GPS and ETC data, obtaining a fusion result, and using data with high confidence to estimate the speed in the fusion result;

[0087] Set the confidence range and classification threshold. Set a confidence range and set the classification threshold. The classification threshold is used to divide the confidence into different levels so that different speed fusion strategies can be adopted according to different confidence levels.

[0088] According to the set confidence classification threshold, the speed fusion of the road section is divided into three categories for processing. These three categories can correspond to high confidence, medium confidence and low confidence situations respectively.

[0089] Under each confidence level, the weighted average of GPS and ETC data is used for speed fusion. The weight of the weighted average can be determined based on factors such as the confidence level, accuracy, and historical performance of the data. For high-confidence data, a larger weight can be given; for low-confidence data, a smaller weight can be given or even ignored.

[0090] The fused speed result is obtained through weighted average calculation. This result should be more accurate and reliable than the speed of a single data source because it combines information from multiple data sources.

[0091] In the fusion result, if the confidence of a certain data source is particularly high (for example, exceeding a certain set threshold), the speed estimation of this data source can be directly used as the final speed result.

[0092] In summary, step S104 sets the confidence range and classification threshold and performs speed fusion according to the confidence level, aiming to obtain a more accurate and reliable speed estimation result, making full use of the respective advantages of GPS data and ETC data, and improving the accuracy of speed estimation through weighted averaging and confidence evaluation. The final speed fusion result will provide important input information for subsequent highway abnormal state detection.

[0093] Step S105: Introduce the Gaussian kernel function to construct a multi-classification support vector machine model, and determine the final classification result through a voting mechanism, determine the input sample, extract features from the input sample, calculate through a binary classification support vector machine, and use a voting mechanism to determine the category to obtain the detection result of the abnormal state of the highway.

[0094] Support vector machine (SVM) is a commonly used machine learning algorithm suitable for classification problems. In highway abnormal state detection, SVM can be used to distinguish different abnormal states.

[0095] Since abnormal states may contain multiple types (such as traffic congestion, vehicle accidents, speeding, etc.), a multi-classification SVM model is needed. The Gaussian kernel function (also known as radial basis function, RBF) is introduced here because it performs well in processing nonlinear data and can map the original data to a high-dimensional space, making it easier to find hyperplanes that distinguish different categories in the high-dimensional space.

[0096] When building a multi-classification SVM model, you can use a "one-to-one" or "one-to-many" strategy to train multiple binary SVMs and combine them in some way (such as a voting mechanism) to achieve multi-classification.

[0097] Input samples refer to vehicle data processed in the previous steps, including speed fusion results, vehicle location, timestamp and other information. Feature extraction is to extract features useful for classification tasks from input samples. In highway abnormal state detection, features may include the average speed of vehicles, speed change rate, distance between vehicles, time for vehicles to pass a specific section, etc.

[0098] Calculated by binary support vector machine:

[0099] For each binary classification SVM, the extracted features are used as input, and calculations are performed to obtain the classification results of the SVM for the input samples.

[0100] The process involves the calculation of the SVM hyperplane, the construction of the decision function, and the calculation of the distance from the sample point to the hyperplane.

[0101] Use voting mechanism to determine the category: After obtaining the classification results of multiple binary SVMs, use voting mechanism to determine the final category of the input sample.

[0102] The voting mechanism can be simply understood as "the minority obeys the majority", that is, the final category is determined based on the classification results of the majority SVM. More complex voting strategies, such as weighted voting, can also be used to further improve the accuracy of classification.

[0103] Obtaining the detection result of the highway abnormal state: After the above steps, the detection result of the highway abnormal state is finally obtained. This result can be one or more abnormal state labels, indicating the abnormal state category to which the input sample belongs.

[0104] According to the detection results of abnormal conditions on the highway, abnormal conditions on the highway can be responded and handled in a timely manner, such as issuing traffic warning information, dispatching rescue resources, etc., to ensure the safety and smooth flow of the highway.

[0105] In summary, step S105 constructs a multi-classification support vector machine model by introducing a Gaussian kernel function, extracts features and performs binary classification calculations on the input samples, and finally uses a voting mechanism to determine the category, thereby obtaining the detection result of the abnormal state of the highway. This process fully utilizes the advantages of SVM in dealing with classification problems and the role of the voting mechanism in improving classification accuracy, providing an effective solution for abnormal state detection of highways.

[0106] Specifically, the method for detecting abnormal state of a highway based on a support vector machine according to the present invention, step S103 comprises:

[0107] Kalman filter initialization is used to initialize the Kalman filter for each vehicle trajectory and ETC gantry data sequence, setting the initial state and covariance matrix;

[0108] Kalman prediction step, using the vehicle dynamics model to predict the state at the next time point, the state at the time point includes position and speed;

[0109] When new GPS track points or ETC gantry data are received, the new GPS track points or ETC gantry data are used as observations to update the state estimate of the Kalman filter;

[0110] Use the observation value and Kalman gain to update the state estimate and covariance matrix, perform adaptive weighted averaging on the data processed by Kalman filtering according to the covariance matrix, output the filtered and weighted data, and output the vehicle trajectory data and ETC gantry data after Kalman filtering and adaptive weighted averaging.

[0111] The Kalman filter and adaptive weighted average processing in the highway abnormal state detection method based on support vector machine described in the present invention are as follows:

[0112] Kalman filter initialization, initialize a Kalman filter for each vehicle trajectory and ETC gantry data sequence. This includes setting the initial state of the filter (such as the initial estimate of position and velocity) and the covariance matrix (representing the uncertainty of the state estimate). The initial state can be set based on prior knowledge or the first observation. The covariance matrix is ​​usually set as a diagonal matrix, and the elements on the diagonal represent the initial uncertainty of each state component.

[0113] The vehicle dynamics model (such as the uniform linear motion model, uniformly accelerated linear motion model, etc.) is used to predict the state at the next time point, which includes the vehicle's position and speed. The prediction step calculates the predicted state and predicted covariance matrix at the next time point based on the current state estimate and covariance matrix, as well as the parameters of the vehicle dynamics model.

[0114] Kalman update step: When new GPS trajectory points or ETC gantry data are received, these data are used as observations to update the state estimate of the Kalman filter.

[0115] First, the residuals (i.e., differences) between the observed and predicted states are calculated, along with the covariance matrix of the residuals.

[0116] Then, the Kalman gain is calculated based on the residual and the covariance matrix, which determines the weight of the observations when updating the state estimate.

[0117] The Kalman gain and residual are used to update the state estimate and covariance matrix to obtain a more accurate and reliable state estimate.

[0118] Adaptive weighted averaging: The data processed by the Kalman filter (i.e., the updated state estimate and covariance matrix) is adaptively weighted averaged according to the covariance matrix.

[0119] The diagonal elements of the covariance matrix represent the uncertainty of each state component and can therefore be used as weights for weighted averaging. The smaller the uncertainty (i.e., the smaller the covariance), the greater the weight of the state component in the weighted average.

[0120] Through adaptive weighted averaging, the data can be further smoothed and fused to obtain more accurate and continuous vehicle trajectory data and ETC gantry data.

[0121] Output filtered and weighted data: Output vehicle trajectory data and ETC gantry data after Kalman filtering and adaptive weighted averaging. These data will be used in subsequent speed fusion and abnormal state detection steps.

[0122] In summary, step S103 smoothes, fuses and optimizes the vehicle trajectory data and ETC gantry data through Kalman filtering and adaptive weighted averaging, providing a more accurate data basis for subsequent highway abnormal state detection.

[0123] Specifically, the method for detecting abnormal state of a highway based on a support vector machine according to the present invention, step S104 comprises:

[0124] Divide the confidence into high confidence, medium confidence and low confidence;

[0125] For each GPS track point and ETC gantry data point, the confidence of each GPS track point and ETC gantry data point is calculated to obtain a set of GPS and ETC data point confidences;

[0126] The data points are divided into three categories according to the confidence level, including high confidence data, medium confidence data and low confidence data, and the fusion result is obtained;

[0127] Check the confidence of each data point, select the data point or data point set with the highest confidence as the final velocity estimate, and output the velocity fusion result.

[0128] The speed fusion process in the method for detecting abnormal state of highway based on support vector machine described in the present invention is as follows:

[0129] Velocity fusion is based on confidence;

[0130] Confidence division: First, the confidence is divided into three levels: high confidence, medium confidence, and low confidence. This is usually determined based on experience or experimental data, and each level corresponds to a confidence range.

[0131] Calculate confidence: For each GPS track point and ETC gantry data point, calculate its confidence. The confidence calculation can be based on a variety of factors, such as data accuracy, historical performance, consistency with other data, etc. Get a set of GPS and ETC data point confidences, which contains the confidence information of all data points.

[0132] Data classification: According to the calculated confidence, the data points are divided into three categories: high confidence data, medium confidence data and low confidence data. The three types of data will be used in the subsequent velocity fusion process, among which high confidence data will be given a higher weight.

[0133] Speed ​​fusion: For classified data, the weighted average method is used for speed fusion. The weight of the weighted average can be determined according to the confidence of the data, and high confidence data is given a greater weight.

[0134] Specifically, the weighted average speed of each type of data can be calculated, and then the weighted average speeds of the three types of data can be weighted averaged again to obtain the final speed fusion result. In this process, the speed of high-confidence data will dominate.

[0135] Select the final speed estimate: Check the confidence of each data point and determine the data point or set of data points with the highest confidence. In some cases, there may be only one data point with a particularly high confidence, in which case the speed of that data point can be directly selected as the final speed estimate. In other cases, there may be multiple data points with high confidence, in which case the average or weighted average of the speeds of these data points can be selected as the final speed estimate.

[0136] Output speed fusion result: Output the final result after speed fusion. The final result after speed fusion will be used in the subsequent highway abnormal state detection step.

[0137] In summary, step S104 implements confidence-based velocity fusion through the steps of calculating confidence, classifying data, velocity fusion, and selecting final velocity estimation.

[0138] Specifically, the method for detecting abnormal state of a highway based on a support vector machine according to the present invention, step S105 comprises:

[0139] Determine the input sample. The input sample is the feature representation of the vehicle trajectory data and ETC gantry data processed in the previous steps. Label the input sample, that is, mark the normal state and abnormal state;

[0140] The Gaussian kernel function is introduced to construct the support vector machine model. The Gaussian kernel function maps the original feature space to a high-dimensional space.

[0141] Use labeled input samples to train a binary SVM model. For normal or abnormal states, train the SVM model to distinguish between categories.

[0142] The abnormal status of highways includes traffic congestion type, accident type and vehicle failure type. A binary classification SVM is trained for each pair of categories, and then the final classification result is determined through a voting mechanism to obtain the detection result of the abnormal status of highways:

[0143] Based on the final classification result, each input sample is labeled as normal or abnormal.

[0144] The multi-classification support vector machine model construction and abnormal state detection process in the highway abnormal state detection method based on support vector machine described in the present invention are as follows:

[0145] Determine input samples and label annotations: Input samples are feature representations of vehicle trajectory data and ETC gantry data processed by previous steps (such as Kalman filtering, velocity fusion, etc.). These features may include vehicle speed, acceleration, position, travel time, ETC gantry passing time, etc.

[0146] The input samples are labeled, that is, the samples are marked as normal or abnormal according to the actual situation of the highway. Abnormal conditions may include traffic congestion, accidents, vehicle failures, etc.

[0147] Introducing Gaussian kernel function to build SVM model: Gaussian kernel function (also known as radial basis function, RBF) is selected to build support vector machine (SVM) model. Gaussian kernel function can map the original feature space to high-dimensional space, making it easier to find hyperplanes that distinguish different categories in high-dimensional space. The parameters of Gaussian kernel function (such as width parameter σ) need to be adjusted according to actual conditions to obtain the best classification performance.

[0148] Training a binary SVM model: Use labeled input samples to train a binary SVM model. For the normal state and each abnormal state (such as traffic congestion, accident, vehicle failure, etc.), you need to train a separate binary SVM model. During the training process, the SVM model will learn how to distinguish between normal and abnormal states, or distinguish between different abnormal states based on the characteristics of the input samples.

[0149] Constructing a multi-classification SVM model and voting mechanism: Since highway abnormal conditions include many types, it is necessary to construct a multi-classification SVM model. This can be achieved by training a binary classification SVM for each pair of categories (such as normal state and traffic congestion, normal state and accident, etc.). When a new input sample needs to be classified, all trained binary classification SVMs are used for prediction, and the final classification result is determined by a voting mechanism. The voting mechanism can be a simple "majority vote", that is, the category with the most votes is selected as the final classification result.

[0150] Obtain the detection results of highway abnormal conditions: According to the final classification results, each input sample is marked as a normal state or an abnormal state (such as traffic congestion, accident, vehicle failure, etc.). The detection results can be used for real-time monitoring and management of highways, helping managers to understand and respond to abnormal conditions on highways in a timely manner.

[0151] According to the detection results of abnormal status, corresponding measures can be taken, such as issuing traffic warning information, dispatching rescue resources, adjusting traffic signals, etc., to ensure the safety and smooth flow of highways.

[0152] In summary, step S105 achieves accurate detection of abnormal conditions of highways by constructing a multi-classification support vector machine model and using a voting mechanism.

[0153] In a second aspect, the present invention provides a highway abnormal state detection system based on a support vector machine, which is applied to a highway abnormal state detection method based on a support vector machine, comprising:

[0154] A data acquisition module is used to acquire vehicle GPS trajectory data and ETC gantry data on the expressway, pre-process the vehicle GPS trajectory data and ETC gantry data on the expressway to obtain pre-processed vehicle GPS trajectory data and ETC gantry data on the expressway, and pre-process the vehicle GPS trajectory data and ETC gantry data on the expressway, including identifying and removing breakpoints in the vehicle trajectory data, projecting the trajectory point data onto the expressway network data using a map matching algorithm, sorting the ETC gantry data by transaction time to generate vehicle transaction gantry trajectory data, and removing data in an erroneous sequence;

[0155] Data association module: used to associate the pre-processed vehicle GPS trajectory data and ETC gantry data on the highway according to the license plate number, time and location, and identify the set of vehicles using GPS and ETC in the same time period;

[0156] Data processing module: used to perform Kalman filtering on trajectory data and ETC gantry data, and perform adaptive weighted averaging on the data after Kalman filtering;

[0157] Data analysis module: used to set the confidence range and classification threshold, and perform speed fusion according to the confidence. Speed ​​fusion according to the confidence includes classifying the speed fusion of the road section into three categories according to different confidence levels, and performing speed fusion by weighted average of GPS and ETC data to obtain the fusion result. In the fusion result, the data with high confidence is used for speed estimation.

[0158] Result output module: used to introduce Gaussian kernel function to build a multi-classification support vector machine model, and determine the final classification result through a voting mechanism, determine the input sample, extract features from the input sample, calculate through a binary classification support vector machine, and use a voting mechanism to determine the category to obtain the detection result of the abnormal state of the highway.

[0159] Specifically, in the highway abnormal state detection system based on support vector machine described in the present invention, the data processing module is also used for:

[0160] Kalman filter initialization is used to initialize the Kalman filter for each vehicle trajectory and ETC gantry data sequence, setting the initial state and covariance matrix;

[0161] Kalman prediction step, using the vehicle dynamics model to predict the state at the next time point, the state at the time point includes position and speed;

[0162] When new GPS track points or ETC gantry data are received, the new GPS track points or ETC gantry data are used as observations to update the state estimate of the Kalman filter;

[0163] Use the observation value and Kalman gain to update the state estimate and covariance matrix, perform adaptive weighted averaging on the data processed by Kalman filtering according to the covariance matrix, output the filtered and weighted data, and output the vehicle trajectory data and ETC gantry data after Kalman filtering and adaptive weighted averaging.

[0164] Specifically, in the highway abnormal state detection system based on support vector machine described in the present invention, the data analysis module is also used for:

[0165] Divide the confidence into high confidence, medium confidence and low confidence;

[0166] For each GPS track point and ETC gantry data point, the confidence of each GPS track point and ETC gantry data point is calculated to obtain a set of GPS and ETC data point confidences;

[0167] The data points are divided into three categories according to the confidence level, including high confidence data, medium confidence data and low confidence data, and the fusion result is obtained;

[0168] Check the confidence of each data point, select the data point or data point set with the highest confidence as the final velocity estimate, and output the velocity fusion result.

[0169] Specifically, in the highway abnormal state detection system based on support vector machine described in the present invention, the result output module is also used for:

[0170] Determine the input sample. The input sample is the feature representation of the vehicle trajectory data and ETC gantry data processed in the previous steps. Label the input sample, that is, mark the normal state and abnormal state;

[0171] The Gaussian kernel function is introduced to construct the support vector machine model. The Gaussian kernel function maps the original feature space to a high-dimensional space.

[0172] Use labeled input samples to train a binary SVM model. For normal or abnormal states, train the SVM model to distinguish between categories.

[0173] The abnormal status of highways includes traffic congestion type, accident type and vehicle failure type. A binary classification SVM is trained for each pair of categories, and then the final classification result is determined through a voting mechanism to obtain the detection result of the abnormal status of highways:

[0174] Based on the final classification result, each input sample is labeled as normal or abnormal.

[0175] In order to better understand the purpose of the present invention, the present invention is described in further detail below.

[0176] See also Figure 1 As shown in the figure, the data used in this study include vehicle GPS trajectory data and ETC gantry data. Since both data contain speed data features, for sections that contain both GPS trajectory points and ETC gantries, the confidence level is considered to perform a cross-section speed fusion model to identify the actual vehicle speed of the section. At the same time, for sections with only GPS and sections with only gantries, the speed of a single data source is estimated. According to the confidence estimation model, there are three types of results: high confidence in vehicle trajectory and gantry data, high confidence in vehicle trajectory data or gantry data, and low confidence in vehicle trajectory data and gantry data. The first two types of results are used as a set of sections to be identified. By integrating features such as section speed, flow, and weather, a highway abnormal state detection model based on a support vector machine is used to identify abnormal states through model training and prediction. The method also includes a complete system architecture, including a data acquisition module, a data processing module, an anomaly detection module, and an early warning output module. The data acquisition module is responsible for capturing various types of data on the highway in real time; the data processing module is responsible for data preprocessing and feature extraction; the anomaly detection module uses the trained SVM model to quickly analyze real-time data to determine whether there is an abnormal state; finally, the early warning output module presents the detection results to the manager in an intuitive manner, and can automatically trigger the corresponding emergency response mechanism according to the preset early warning rules.

[0177] Data source: Vehicle trajectory data: Vehicle trajectory data comes from GPS, mobile devices, vehicle terminals, etc., and contains vehicle location, speed, time and other information.

[0178] ETC gantry data: The ETC gantry collects speed, flow and time information of vehicles passing through.

[0179] Data preprocessing, the vehicle trajectory data includes the identification and elimination of breakpoints, and the map matching algorithm, and the trajectory point data is projected onto the highway network data under study.

[0180] ETC gantry data includes the formation ID, transaction gantry and transaction time. The transaction data generated by the interaction between the traffic medium and the gantry during the vehicle's driving process is sorted by transaction time and the generated transaction data sequence is called vehicle transaction gantry trajectory data. The gantry trajectory data can be further divided into abnormal data and normal data. For specific classification forms, please refer to Figure 2 .

[0181] ETC data is affected by gantry acquisition errors and may have three error sequences: missing, object, and duplication. Therefore, the data of the three error sequences need to be eliminated during data preprocessing.

[0182] Both ETC data and vehicle trajectory data record the speed of vehicles passing through the gantry or in real time. However, for vehicles that hold GPS data and pass through the gantry at the same time, using both data will cause duplicate calculations. Therefore, it is necessary to first associate the trajectory data and ETC data according to the license plate number, time, and location to identify the set of vehicles that have both GPS and ETC in the same time period. For the above vehicle set, the confidence of the gantry speed collection can be identified by mutual verification of multi-source data.

[0183] Based on the velocity fusion model of Kalman filter, in actual monitoring scenarios, sensor data is often interfered by noise, resulting in decreased accuracy and stability of the data. To this end, this paper uses the Kalman filter algorithm to fuse the data of the same type of sensors. The algorithm can reduce the impact of noise and uncertainty through the prediction and filtering of prior estimates and posterior observations, and improve the accuracy and stability of sensor data, thereby effectively solving the problem of unstable sensor data in actual monitoring scenarios.

[0184] The main steps of the Kalman filter algorithm include two stages: prediction and update. In the prediction stage, the state estimation and covariance matrix of the current moment are predicted based on the state estimation of the previous moment and the dynamic model of the system through the state transfer equation. That is:

[0185]

[0186] in, is the predicted estimated value of the state at time t; is the optimal value at time t-1; A tis the state transfer matrix; B t is the control matrix; U t is the covariance matrix of process noise; Q t is the system process noise covariance matrix.

[0187] In the update phase, the prediction estimate and the covariance matrix are modified according to the observation equation and the measurement data collected by the sensor, that is:

[0188]

[0189] P t|t =(IK G H t ) t|t-1

[0190] K G is the Kalman gain; is the observation matrix; Z t is the measurement data collected by the sensor; R t is the covariance matrix of the measurement noise; is the revised estimated value of the state at time t; P t|t is the corrected covariance matrix.

[0191] The adaptive weighted average fusion method, when processing data collected by the same type of sensors, due to the difference between the trajectory point data and the vehicle speed collected by ETC data, the collected data needs to be fused to improve accuracy. The Kalman filter algorithm can reduce noise and uncertainty, but cannot solve the differences between sensors. Therefore, the adaptive weighted average algorithm is used to process the data after Kalman filtering to better reflect the true value. By weighted averaging the data of different sensors, the accuracy and stability of the fusion results can be further improved.

[0192] There may be some deviations between the trajectory data speed and the vehicle speed collected by ETC. Assume that the variances of the two data sources are The total variance is:

[0193]

[0194] Since each sensor is located at a different position in the detection environment, it can be approximately considered that the data of each sensor is independent of each other, so:

[0195] E[(XX i )(XX j )]=0 (i≠j)

[0196]

[0197] The total variance is a multivariate quadratic function of the weighted coefficient, so it has a minimum value. According to the extreme value theory of multivariate functions, the weighted coefficient corresponding to the total minimum variance is

[0198]

[0199] The total variance is now:

[0200]

[0201] The resulting fusion value is:

[0202]

[0203] In the formula The mean of the historical data measured k times by each sensor.

[0204] Confidence judgment model, set the confidence range to C GPS and C ETC , where C GPS Indicates the confidence of GPS data, C ETC Indicates the confidence of ETC data. Confidence can be calculated based on factors such as data quality and signal coverage. The classification of confidence can be defined using the following thresholds:

[0205] C High : High confidence threshold

[0206] C low : Low confidence threshold

[0207] According to the different confidence levels, the speed fusion of road sections is divided into the following three categories:

[0208] The vehicle trajectory and gantry data have high confidence;

[0209] Condition: C GPS >C High And C ETC >C Higj

[0210] Processing method: Use the weighted average of GPS and ETC data for speed fusion.

[0211] The vehicle trajectory data or gantry data has high confidence;

[0212] Condition: C GPS >C Higj or C ETC >C High

[0213] Approach: Only use data with high confidence for velocity estimation.

[0214] The confidence of both vehicle trajectory data and gantry data is low;

[0215] Condition: C GPS <C low And C ETC <C low

[0216] Processing method: Mark such road sections as road sections to be identified, and use other features (such as historical data, traffic volume, and weather) to assist in estimating speed. Machine learning models or interpolation methods can be used to estimate speed.

[0217] Support vector machine, model construction based on multi-classification support vector machine;

[0218] Introduction of kernel function;

[0219] In order to handle nonlinear problems, support vector machines use kernel functions to map input data to high-dimensional space. In multi-classification SVM, Gaussian kernel functions (i.e. radial basis functions, RBF) are usually used:

[0220]

[0221] Here, σ is the bandwidth parameter of the kernel function, x i and x j is the feature vector of the sample.

[0222] Voting mechanism for multi-classification SVM;

[0223] Support vector machine is a classifier based on the principle of structural risk minimization. The basic SVM is a binary classifier that divides data into positive and negative classes by constructing an optimal hyperplane as the classification interface. This hyperplane is formed by feature vectors in a high-dimensional space. However, high-speed abnormal events include:

[0224] A multi-classification SVM is composed of multiple binary classification SVMs. If there are M categories of data to be classified, then a binary classification SVM needs to be trained between each two categories, and a total of M(M-1) / 2 binary classification SVMs are required. The collection of these binary classification SVMs plus the voting mechanism forms a multi-classification SVM. When performing pattern recognition, each binary classification SVM in the multi-classification SVM is used to classify the data to be recognized, and the output results of each classifier are counted. The category with the most votes is used as the final category of the data to be recognized. Please refer to Figure 3 .

[0225] For each binary support vector machine, its classification hyperplane function is determined by the training samples containing these two categories:

[0226]

[0227] in is the classification result of the identified sample (1 for the positive class and -1 for the negative class); α i is the Lagrange multiplier; y i is the category value of the i-th sample (1 for positive class and -1 for negative class); t i is the feature vector of the i-th sample; t is the feature vector of the identified sample; b is the intercept of the hyperplane curve; k is the inner product function, and the Gaussian inner product function is used in this paper.

[0228] 3) Final classification of abnormal status

[0229] The final classification result is determined by a voting mechanism, that is, the result of each binary SVM is recorded as one vote, and the category with the most votes is selected as the final classification result. Assuming that there is a CCC class to be identified, the final classification result can be expressed as:

[0230]

[0231] Among them, f ij (x) is the binary classifier output of categories i and j. By counting the votes for each category, the category y with the most votes is final Selected as the final abnormal state classification result.

[0232] Mathematical expression of highway abnormal state recognition: for each input sample x, based on features (such as vehicle speed, flow, etc.), all binary classification SVMf ij (x) is calculated and a voting mechanism is used to determine its final category:

[0233]

[0234] This process can identify abnormal conditions on the highway through the output of multiple SVM classifiers. This method can accurately classify and detect different types of abnormal conditions, ensuring that the real-time monitoring and management system of the highway can effectively identify and respond to various emergencies.

Claims

1. A method for detecting abnormal state of highway based on support vector machine, characterized in that: include: Step S101, obtaining vehicle GPS trajectory data and ETC gantry data on the expressway, preprocessing the vehicle GPS trajectory data and ETC gantry data on the expressway to obtain preprocessed vehicle GPS trajectory data and ETC gantry data on the expressway, preprocessing the vehicle GPS trajectory data and ETC gantry data on the expressway includes identifying and removing breakpoints in the vehicle trajectory data, projecting the trajectory point data onto the expressway network data using a map matching algorithm, sorting the ETC gantry data by transaction time to generate vehicle transaction gantry trajectory data, and removing data in an erroneous sequence; Step S102: Associating the pre-processed vehicle GPS trajectory data and ETC gantry data on the highway according to the license plate number, time, and location, and identifying a set of vehicles using GPS and ETC in the same time period; Step S103: performing Kalman filtering on the trajectory data and the ETC gantry data, and performing adaptive weighted averaging on the data after the Kalman filtering; Step S104: setting a confidence range and a classification threshold, and performing speed fusion according to the confidence. The speed fusion according to the confidence includes classifying the speed fusion of the road section into three categories according to different confidences, performing speed fusion by weighted average of GPS and ETC data, obtaining a fusion result, and using data with high confidence to estimate the speed in the fusion result; Step S105: Introduce the Gaussian kernel function to construct a multi-classification support vector machine model, and determine the final classification result through a voting mechanism, determine the input sample, extract features from the input sample, calculate through a binary classification support vector machine, and use a voting mechanism to determine the category to obtain the detection result of the abnormal state of the highway.

2. A method for detecting abnormal state of a highway based on a support vector machine as claimed in claim 1, characterized in that: The step S103 includes: Kalman filter initialization is used to initialize the Kalman filter for each vehicle trajectory and ETC gantry data sequence, setting the initial state and covariance matrix; Kalman prediction step, using the vehicle dynamics model to predict the state at the next time point, the state at the time point includes position and speed; When new GPS track points or ETC gantry data are received, the new GPS track points or ETC gantry data are used as observations to update the state estimate of the Kalman filter; Use the observation value and Kalman gain to update the state estimate and covariance matrix, perform adaptive weighted averaging on the data processed by Kalman filtering according to the covariance matrix, output the filtered and weighted data, and output the vehicle trajectory data and ETC gantry data after Kalman filtering and adaptive weighted averaging.

3. The method for detecting abnormal state of a highway based on a support vector machine as claimed in claim 1, characterized in that: The step S104 includes: Divide the confidence into high confidence, medium confidence and low confidence; For each GPS track point and ETC gantry data point, the confidence of each GPS track point and ETC gantry data point is calculated to obtain a set of GPS and ETC data point confidences; The data points are divided into three categories according to the confidence level, including high confidence data, medium confidence data and low confidence data, and the fusion result is obtained; Check the confidence of each data point, select the data point or data point set with the highest confidence as the final velocity estimate, and output the velocity fusion result.

4. A method for detecting abnormal state of a highway based on a support vector machine as claimed in claim 1, characterized in that: The step S105 includes: Determine the input sample. The input sample is the feature representation of the vehicle trajectory data and ETC gantry data processed in the previous steps. Label the input sample, that is, mark the normal state and abnormal state; The Gaussian kernel function is introduced to construct the support vector machine model. The Gaussian kernel function maps the original feature space to a high-dimensional space. The input samples with labels are used to train the binary classification SVM model. For normal or abnormal states, the SVM model is trained to distinguish categories. The abnormal status of highways includes traffic congestion type, accident type and vehicle failure type. A binary classification SVM is trained for each pair of categories, and then the final classification result is determined through a voting mechanism to obtain the detection result of the abnormal status of highways: Based on the final classification result, each input sample is labeled as normal or abnormal.

5. A highway abnormal state detection system based on support vector machine, applied to the highway abnormal state detection method based on support vector machine of claims 1 to 4, characterized in that: include: A data acquisition module is used to acquire vehicle GPS trajectory data and ETC gantry data on the expressway, pre-process the vehicle GPS trajectory data and ETC gantry data on the expressway to obtain pre-processed vehicle GPS trajectory data and ETC gantry data on the expressway, and pre-process the vehicle GPS trajectory data and ETC gantry data on the expressway, including identifying and removing breakpoints in the vehicle trajectory data, projecting the trajectory point data onto the expressway network data using a map matching algorithm, sorting the ETC gantry data by transaction time to generate vehicle transaction gantry trajectory data, and removing data in an erroneous sequence; Data association module: used to associate the pre-processed vehicle GPS trajectory data and ETC gantry data on the highway according to the license plate number, time and location, and identify the set of vehicles using GPS and ETC in the same time period; Data processing module: used to perform Kalman filtering on trajectory data and ETC gantry data, and perform adaptive weighted averaging on the data after Kalman filtering; Data analysis module: used to set the confidence range and classification threshold, and perform speed fusion according to the confidence. Speed ​​fusion according to the confidence includes classifying the speed fusion of the road section into three categories according to different confidence levels, and performing speed fusion by weighted average of GPS and ETC data to obtain the fusion result. In the fusion result, the data with high confidence is used for speed estimation. Result output module: used to introduce Gaussian kernel function to build a multi-classification support vector machine model, and determine the final classification result through a voting mechanism, determine the input sample, extract features from the input sample, calculate through a binary classification support vector machine, and use a voting mechanism to determine the category to obtain the detection result of the abnormal state of the highway.

6. A highway abnormal state detection system based on support vector machine as claimed in claim 5, characterized in that: The data processing module is also used for: Kalman filter initialization is used to initialize the Kalman filter for each vehicle trajectory and ETC gantry data sequence, setting the initial state and covariance matrix; Kalman prediction step, using the vehicle dynamics model to predict the state at the next time point, the state at the time point includes position and speed; When new GPS track points or ETC gantry data are received, the new GPS track points or ETC gantry data are used as observations to update the state estimate of the Kalman filter; Use the observation value and Kalman gain to update the state estimate and covariance matrix, perform adaptive weighted averaging on the data processed by Kalman filtering according to the covariance matrix, output the filtered and weighted data, and output the vehicle trajectory data and ETC gantry data after Kalman filtering and adaptive weighted averaging.

7. A highway abnormal state detection system based on support vector machine as claimed in claim 5, characterized in that: The data analysis module is also used for: Divide the confidence into high confidence, medium confidence and low confidence; For each GPS track point and ETC gantry data point, the confidence of each GPS track point and ETC gantry data point is calculated to obtain a set of GPS and ETC data point confidences; The data points are divided into three categories according to the confidence level, including high confidence data, medium confidence data and low confidence data, and the fusion result is obtained; Check the confidence of each data point, select the data point or data point set with the highest confidence as the final velocity estimate, and output the velocity fusion result.

8. A highway abnormal state detection system based on support vector machine as claimed in claim 5, characterized in that: The result output module is also used for: Determine the input sample. The input sample is the feature representation of the vehicle trajectory data and ETC gantry data processed in the previous steps. Label the input sample, that is, mark the normal state and abnormal state; The Gaussian kernel function is introduced to construct the support vector machine model. The Gaussian kernel function maps the original feature space to a high-dimensional space. The input samples with labels are used to train the binary classification SVM model. For normal or abnormal states, the SVM model is trained to distinguish categories. The abnormal status of highways includes traffic congestion type, accident type and vehicle failure type. A binary classification SVM is trained for each pair of categories, and then the final classification result is determined through a voting mechanism to obtain the detection result of the abnormal status of highways: Based on the final classification result, each input sample is labeled as normal or abnormal.

Images