Driver global speed evolution mode classification and prediction method and system under progressive variable speed limit, electronic equipment and storage medium

Abstract

The invention discloses a driver global speed evolution mode classification and prediction method and system under progressive variable speed limit, electronic equipment and a storage medium. The mode classification and prediction method comprises the following steps: arranging lane-level variable speed limit information issuing equipment; arranging a vehicle track data acquisition device; according to a preset control strategy, a variable information system is used for issuing a corresponding lane-level speed-limiting control instruction, and the speed-limiting value of the lane-level variable speed-limiting plate is adjusted; clustering is carried out based on a driver global vehicle speed track mode; constructing a driver global speed evolution model under progressive lane-level variable speed limit control; and constructing a driver global speed evolution prediction model under the lane-level variable speed limit control. According to the method, global speed prediction of the driver can be carried out by utilizing the speed evolution condition of the driver under the gradual variable speed limit condition, a customized personalized driving behavior early warning scheme is provided for the driver, the cost is low, and the method has the advantages of being capable of being copied and popularized and high in robustness.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent traffic control, and relates to a pattern classification and prediction method, and in particular to a method, system, electronic device and storage medium for classifying and predicting a driver's global speed evolution pattern under a progressive variable speed limit. Background Art

[0002] With the development of sensor technology, artificial intelligence, 5G and Internet of Things technology, precise and personalized vehicle command push technology is gradually accelerating towards implementation, and intelligent networked vehicle technology is constantly updated and iterated. Intelligent networked vehicles will reconstruct the relationship between people, vehicles and roads, bringing vehicle perception and control to a new level.

[0003] With the popularization of motor vehicles and the growth of transportation demand, the static speed control method and the information release method based on variable information boards used in the current highway safety active control technology are gradually unable to meet the needs of the intelligent network environment. First, the static speed limit method cannot provide safe and reasonable speed limit values ​​for different working conditions and specific conditions in a dynamic road traffic environment; second, the means of information release based on variable information boards cannot provide personalized speed limit guidance for specific vehicles, resulting in large differences in management effects between different individuals, which has a negative impact on highway traffic safety; third, due to the different layout spacing and published speed values ​​of the variable speed limit boards set continuously on the highway, different drivers have different driving behavior strategies under the influence of the progressive lane-level variable speed limit conditions, and targeted speed limit guidance is required for the driving behavior chain of different drivers.

[0004] In view of this, there is an urgent need to design a new road safety control method to overcome at least some of the above-mentioned defects of the existing road safety control method. Summary of the invention

[0005] The present invention provides a method, system, electronic device and storage medium for classifying and predicting the evolution pattern of the driver's global speed under a progressively variable speed limit. The method can use the driver's speed evolution under the progressively variable speed limit to predict the driver's global speed and provide a customized and personalized driving behavior warning solution. The solution is independent of vehicle-side equipment, has low cost, and is replicable and promotable and has strong robustness.

[0006] To solve the above technical problem, according to one aspect of the present invention, the following technical solution is adopted:

[0007] A method for classifying and predicting a driver's global speed evolution pattern under a progressive variable speed limit, the method comprising:

[0008] Step S1, deploying a series of lane-level variable speed limit information publishing devices at set intervals on a high-speed expressway section with a free flow state;

[0009] Step S2: deploying vehicle trajectory data collection devices at set intervals on the road area to collect and analyze vehicle time-series trajectory data;

[0010] Step S3: according to the preset control strategy, the variable information system is used to issue and control the corresponding lane-level speed limit control command, the speed limit value of the lane-level variable speed limit plate is adjusted, and the vehicle trajectory data and lane-level speed limit information release data are exported in real time;

[0011] Step S4: based on the clustering of the driver's global vehicle speed trajectory pattern, the driver's vehicle trajectory is classified according to different time series distribution characteristics, so as to classify the driver under the progressive variable speed limit;

[0012] Step S5: constructing a global speed evolution model of the driver under the progressive lane-level variable speed limit control, segmenting the global speed of the driver's vehicle, and obtaining the state transition probability between different trajectory segments, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit;

[0013] Step S6: construct a prediction model for the global speed evolution of the driver under lane-level variable speed limit control, and predict the speed evolution trend according to the global speed distribution characteristics of the driver under progressive lane-level variable speed limit control, so as to predict the driver's driving behavior and actively predict the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

[0014] As an implementation mode of the present invention, in step S1, a lane-level variable speed limit information release device is arranged on the expressway section;

[0015] In step S2, radars, cameras and other vehicle trajectory data collection devices are deployed at set intervals on the road area to obtain continuous and uninterrupted sub-meter-level time-series vehicle trajectories, with a speed collection accuracy of no less than 1km / h and a collection frequency of 10Hz, providing a data basis for the collection and analysis of the driver's real-time time-series driving behavior characteristics;

[0016] In step S3, the vehicle time-series trajectory data and lane-level speed limit information release data are all collected in real time and use UTC time, and the location of data release uses a unified stake number coordinate;

[0017] In step S6, on the basis of completing the classification of the driver's global speed evolution pattern, the driver's global speed evolution pattern architecture is constructed based on the graph neural network, and the graph neural network is combined with CNN-LSTM to further construct the driver's global speed evolution prediction model, that is, the MTK-GNN-CNN-LSTM global speed evolution prediction model, to predict its speed evolution trend, thereby predicting its driving behavior, and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limits.

[0018] As an implementation of the present invention, in step S4, the K-Shape clustering algorithm is used to eliminate the amplitude influence of the time series through normalization, local alignment is achieved by moving the time series, and the fast Fourier algorithm is introduced in the calculation of similarity to reduce the calculation cost of the high-dimensional time series. The K-Shape algorithm combines the concepts of DTW distance and shape features and can effectively capture the intrinsic structure of the time series; step S4 includes the following steps:

[0019] Step S41, unifying the length of vehicle time series trajectory data and the distribution of acquisition point coordinates; since the radar acquisition frequency is fixed, vehicles of different speeds take different time lengths to pass through the same road section, resulting in different lengths of speed time series data. In order to achieve matching based on vehicle position in the clustering process of different time series, the speed time series data of each vehicle uses the longitudinal position as the "time series" and performs interpolation processing at a spacing of 1m, and processes the speed time series data of each vehicle into equal-length time series data;

[0020] Step S42, initialize cluster centers: randomly select K time series from the data set as initial cluster centers. These series will become representative shapes of clusters; set the number of clusters to 3;

[0021] Step S43, shape distance measurement: using DTW or other shape distance measurement to calculate the distance between each time series and each cluster center;

[0022] Step S44, assigning data points: assigning each time series to the cluster with the closest cluster center; thus, each series will be classified into the cluster with the most similar shape to it;

[0023] Step S45, update cluster centers: update cluster centers using the average shape of the time series assigned to each cluster. In this way, cluster centers will be continuously adjusted to better represent the shape characteristics of the cluster;

[0024] Step S46, repeat iteration: Repeat the above steps until the cluster center no longer changes and reaches a predetermined number of iterations; eventually, the obtained cluster will contain all time series in the data set, and each cluster represents a group of sequences with similar shapes.

[0025] As an implementation mode of the present invention, in step S5, a global speed evolution model of the driver under the progressive lane-level variable speed limit control is constructed, the global speed of the driver's vehicle is segmented, and the state transition probability between different trajectory segments is obtained, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit; step S5 includes the following steps:

[0026] Step S51: discretize and perform quantile discretization processing on the vehicle speed time series to obtain vehicle speed discretization partitions.

[0027] Step S52: Establish a Markov transition matrix for the partitioned time series data to obtain the matrix M1

[0028]

[0029] Step S53, normalize the values ​​and calculate the conversion probability matrix M′1:

[0030]

[0031] Step S54, calculating the Markov transition field;

[0032] Step S55, the time-series segmented subsequences processed by the Markov transition field are encoded and mapped by a graph neural network, and a graph neural network is further constructed to describe different speed fluctuation states and the relationship between them;

[0033] Step S56, use the module partitioning algorithm in the ts ia graph neural network analysis tool to perform community partitioning on the obtained graph neural network; the driver's global speed state is divided into different state intervals, and the edges between the neural vertices represent the conversion probability between different speed state intervals; the process of the driver making decisions in the face of speed limits is divided into several state intervals as needed, so as to more finely describe the driver's global speed evolution process.

[0034] According to another aspect of the present invention, the following technical solution is adopted: a classification and prediction system for the global speed evolution pattern of a driver under a progressive variable speed limit, the classification and prediction system comprising:

[0035] A number of information release devices are arranged on a high-speed road section with a free-flow state, and are arranged in sequence according to a set spacing rule;

[0036] A plurality of vehicle trajectory data collection devices are arranged in sequence on the road area according to a set spacing rule to realize the collection and analysis of vehicle time series trajectory data;

[0037] The speed limit value adjustment module is used to use the variable information system to issue and control the corresponding lane-level speed limit control instructions according to the preset control strategy, adjust the speed limit value of the lane-level variable speed limit plate, and export the vehicle trajectory data and lane-level speed limit information release data in real time;

[0038] A classification module is used to cluster the driver's global speed trajectory pattern and classify the driver's vehicle trajectory according to different time series distribution characteristics, so as to classify the driver under the progressive variable speed limit;

[0039] The speed evolution model construction module is used to construct the driver's global speed evolution model under the progressive lane-level variable speed limit control, divide the driver's vehicle global speed into segments, and obtain the state transition probability between different trajectory segments, so as to effectively judge their abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit;

[0040] The speed evolution prediction model construction module is used to construct a driver's global speed evolution prediction model under lane-level variable speed limit control. According to the driver's global speed distribution characteristics under progressive lane-level variable speed limit control, its speed evolution trend is predicted, thereby predicting its driving behavior and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

[0041] As an implementation mode of the present invention, the information publishing device is a lane-level variable speed limit information publishing device;

[0042] The vehicle trajectory data acquisition device includes a radar and / or a camera to obtain continuous and uninterrupted sub-meter-level time-series vehicle trajectories, with a speed acquisition accuracy of no less than 1km / h and an acquisition frequency of 10Hz, providing a data basis for the acquisition and analysis of the driver's real-time time-series driving behavior characteristics;

[0043] The vehicle time-series trajectory data and lane-level speed limit information release data are all collected in real time and use UTC time, and the locations where the data is released use unified stake number coordinates;

[0044] The speed evolution prediction model construction module, based on the classification of the driver's global speed evolution pattern, further constructs the driver's global speed evolution prediction model based on the driver's global speed evolution pattern architecture constructed by the graph neural network, and uses the graph neural network combined with CNN-LSTM, that is, the MTK-GNN-CNN-LSTM global speed evolution prediction model, to predict the driver's speed evolution trend, thereby predicting the driver's driving behavior, and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limits.

[0045] As an implementation mode of the present invention, the classification module adopts the K-Shape clustering algorithm, eliminates the amplitude influence of the time series by normalization, realizes local alignment by moving the time series, and introduces the fast Fourier algorithm in calculating the similarity to reduce the calculation cost of the high-dimensional time series. The K-Shape algorithm combines the concepts of DTW distance and shape features, and can effectively capture the intrinsic structure of the time series; the classification process of the classification module includes the following steps:

[0046] Step S41, unifying the length of vehicle time series trajectory data and the distribution of acquisition point coordinates; since the radar acquisition frequency is fixed, vehicles of different speeds take different time lengths to pass through the same road section, resulting in different lengths of speed time series data. In order to achieve matching based on vehicle position in the clustering process of different time series, the speed time series data of each vehicle uses the longitudinal position as the "time series" and performs interpolation processing at a spacing of 1m, and processes the speed time series data of each vehicle into equal-length time series data;

[0047] Step S42, initialize cluster centers: randomly select K time series from the data set as initial cluster centers. These series will become representative shapes of clusters; set the number of clusters to 3;

[0048] Step S43, shape distance measurement: using DTW or other shape distance measurement to calculate the distance between each time series and each cluster center;

[0049] Step S44, assigning data points: assigning each time series to the cluster with the closest cluster center; thus, each series will be classified into the cluster with the most similar shape to it;

[0050] Step S45, update cluster centers: update cluster centers using the average shape of the time series assigned to each cluster. In this way, cluster centers will be continuously adjusted to better represent the shape characteristics of the cluster;

[0051] Step S46, repeat iteration: Repeat the above steps until the cluster center no longer changes and reaches a predetermined number of iterations; eventually, the obtained cluster will contain all time series in the data set, and each cluster represents a group of sequences with similar shapes.

[0052] As an implementation mode of the present invention, the model building process of the velocity evolution model building module includes the following steps:

[0053] Step S51: discretize and perform quantile discretization processing on the vehicle speed time series to obtain vehicle speed discretization partitions.

[0054] Step S52: Establish a Markov transition matrix for the partitioned time series data to obtain the matrix M1

[0055]

[0056] Step S53, normalize the values ​​and calculate the conversion probability matrix M′1:

[0057]

[0058] Step S54, calculating the Markov transition field;

[0059] Step S55, the time-series segmented subsequences processed by the Markov transition field are encoded and mapped by a graph neural network, and a graph neural network is further constructed to describe different speed fluctuation states and the relationship between them;

[0060] Step S56, use the module partitioning algorithm in the ts ia graph neural network analysis tool to perform community partitioning on the obtained graph neural network; the driver's global speed state is divided into different state intervals, and the edges between the neural vertices represent the conversion probability between different speed state intervals; the process of the driver making decisions in the face of speed limits is divided into several state intervals as needed, so as to more finely describe the driver's global speed evolution process.

[0061] According to another aspect of the present invention, the following technical solution is adopted: an electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0062] According to another aspect of the present invention, the following technical solution is adopted: a storage medium stores computer program instructions, and the computer program instructions implement the steps of the above method when executed by a processor.

[0063] The present invention finds a high-speed expressway with a free-flow period and continuous lane-level variable speed limit plates; arranges vehicle trajectory data acquisition devices at a certain interval on the roadside; according to a preset control strategy, a variable information system is used to issue and control corresponding lane-level speed limit control instructions to adjust the speed limit value of the lane-level variable speed limit plates; the vehicle trajectory characteristic parameters are collected and analyzed by the vehicle trajectory data acquisition device, the vehicle trajectory characteristic data of the driver under different speed limit conditions are extracted, and a classification and prediction model of the driver's global speed evolution pattern under a progressive variable speed limit based on the vehicle driving trajectory data is constructed. The classification and prediction method of the driver's global speed evolution pattern under a progressive variable speed limit provided by the present invention applies time-series vehicle trajectory feature data to realize the prediction of the driver's speed evolution under the progressive variable speed limit condition, solves the problem that the continuous lane-level speed limit scenarios are ever-changing, the combination methods cannot be exhausted, resulting in unclear mutual influence mechanism between lane speed limit measures and driving behavior, resulting in the inability to predict the driver's speed distribution characteristic law under continuous speed limit conditions, and the traditional traffic variable speed limit control cannot perform targeted speed limit according to the driver's driving behavior characteristics. At the same time, it solves the problem that the current variable speed limit control on highways lacks an optimization mechanism for individual drivers, resulting in low driver compliance.

[0064] The beneficial effects of the present invention are as follows: the classification and prediction method, system, electronic device and storage medium of the driver's global speed evolution pattern under a progressively variable speed limit proposed in the present invention can use the driver's speed evolution under the progressively variable speed limit to predict the driver's global speed, and provide a customized and personalized driving behavior warning solution for the driver, which does not rely on vehicle-side equipment, has low cost, and is replicable and promotable and has strong robustness.

[0065] The method of the present invention utilizes a continuous vehicle trajectory acquisition device to collect and analyze the time series data of driving behavior under progressive lane-level variable speed limits, and utilizes K-Shape shape-based clustering to realize the prediction of the driver's speed evolution under progressive variable speed limit conditions, thereby solving the problems that the continuous lane-level speed limit scenarios are ever-changing, the combination methods cannot be exhausted, resulting in unclear mutual influence mechanism between lane speed limit measures and driving behavior, resulting in the inability to predict the speed distribution characteristics of the driver under continuous speed limit conditions, and the traditional traffic variable speed limit control cannot perform targeted speed limit according to the driver's driving behavior characteristics.

[0066] The present invention utilizes a graph neural network model to discretize time series fluctuations into different subsequence fluctuations, thereby achieving accurate quantification of the global speed evolution law and state transition probability of an individual driver under continuous speed limit conditions, accurately extracting the driver's abnormal speed state, and finely describing the evolution of the driver's vehicle speed state under continuous speed limit. This provides a theoretical basis and analysis method for refined variable speed limit control, and can provide personalized speed limit guidance strategies for driver behavior characteristics, with the characteristics of low cost, replicability and strong robustness.

[0067] On the basis of completing the classification of the driver's global speed evolution pattern, the Markov transition field and graph neural network combined with CNN-LSTM are used to further construct the MTK-GNN-CNN-LSTM driver's global speed evolution prediction model architecture. Compared with the conventional CNN-LSTM model, the global speed evolution pattern label obtained based on the Markov transition field-graph neural network community partitioning algorithm is added to the CNN model input. The subsequence dependency and conversion probability of the time series can be input in the input layer to improve the accuracy and stability of the global speed prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 Flow chart of an evolution pattern classification and prediction method in one embodiment of the present invention.

[0069] Figure 2 Schematic diagram of the composition of an evolutionary pattern classification and prediction system in one embodiment of the present invention.

[0070] Figure 3 Schematic diagram of the principle of a driver's global speed evolution pattern classification and prediction method in one embodiment of the present invention.

[0071] Figure 4 The figure is a schematic diagram of a method for deploying a variable speed limit information publishing device in an embodiment of the present invention.

[0072] Figure 5 It is a schematic diagram of a method for deploying traffic flow parameter collection equipment in one embodiment of the present invention.

[0073] Figure 6 Schematic diagram of vehicle speed distribution of three vehicle speed trajectory modes in one embodiment of the present invention.

[0074] Figure 7 Schematic diagram showing comparison of vehicle speed distribution in three global vehicle speed trajectory modes under different speed limit scenarios in one embodiment of the present invention.

[0075] Figure 8 Schematic diagram of vehicle speed time series selected by the speed evolution pattern classification and prediction method in one embodiment of the present invention.

[0076] Fig. 9Schematic diagram of the Markov transformation field generated by the vehicle speed time series in one embodiment of the present invention.

[0077] Fig.10 It is a schematic diagram of neural network construction and parameter statistics in one embodiment of the present invention.

[0078] Fig.11 Schematic diagram of the relationship between the neural network community division and the mapping with time series data in one embodiment of the present invention.

[0079] Fig.12 This is a schematic diagram of a global speed evolution prediction model based on MTK-GNN-CNN-LSTM in one embodiment of the present invention.

[0080] Fig.13 FIG. 4 is a schematic diagram of the composition of an electronic device in one embodiment of the present invention. DETAILED DESCRIPTION

[0081] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0082] In order to further understand the present invention, preferred embodiments of the present invention are described below in conjunction with examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, rather than limiting the claims of the present invention.

[0083] The description in this section is only for several typical embodiments, and the present invention is not limited to the scope of the embodiments. The same or similar prior art means and some technical features in the embodiments are mutually replaced within the scope of the present invention.

[0084] The description of the steps in each embodiment in the specification is only for the convenience of explanation, and the implementation method of the present application is not limited by the order of implementation of the steps.

[0085] The term “connection” in the specification includes both direct connection and indirect connection.

[0086] The present invention discloses a classification and prediction method of the global speed evolution pattern of a driver under a progressive variable speed limit. Figure 1 is a flow chart of an evolution pattern classification and prediction method in one embodiment of the present invention, Figure 3 FIG. 1 is a schematic diagram showing the principle of a method for classifying and predicting a driver's global speed evolution pattern in one embodiment of the present invention; please refer to FIG. Figure 1 , Figure 3 , the pattern classification and prediction method includes:

[0087] [Step S1] A series of lane-level variable speed limit information release devices are arranged at set intervals on a free-flow highway section;

[0088] [Step S2] Deploy vehicle trajectory data collection devices at set intervals on the road area to collect and analyze vehicle time series trajectory data;

[0089] [Step S3] According to the preset control strategy, the variable information system is used to issue and control the corresponding lane-level speed limit control command, adjust the speed limit value of the lane-level variable speed limit plate, and export the vehicle trajectory data and lane-level speed limit information release data in real time;

[0090] [Step S4] Based on the clustering of the driver's global vehicle speed trajectory pattern, the driver's vehicle trajectory is classified according to different time series distribution characteristics, so as to classify the driver under the progressive variable speed limit;

[0091] [Step S5] constructing a global speed evolution model of the driver under the progressive lane-level variable speed limit control, segmenting the global speed of the driver's vehicle, and obtaining the state transition probability between different trajectory segments, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit;

[0092] [Step S6] Construct a prediction model for the driver's global speed evolution under lane-level variable speed limit control, and predict the driver's speed evolution trend based on the driver's global speed distribution characteristics under progressive lane-level variable speed limit control, so as to predict the driver's driving behavior and actively predict the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

[0093] In one embodiment of the present invention, in step S1, a lane-level variable speed limit information issuing device is deployed on the expressway section.

[0094] In step S2, vehicle trajectory data collection devices such as radars and cameras are deployed at set intervals on the road to obtain continuous and uninterrupted sub-meter-level time-series vehicle trajectories. The speed collection accuracy is not less than 1km / h, and the collection frequency reaches 10Hz, providing a data basis for the collection and analysis of the driver's real-time time-series driving behavior characteristics.

[0095] In step S3, the vehicle time-series trajectory data and lane-level speed limit information release data are all collected in real time and use UTC time, and the locations where the data is released use unified stake number coordinates.

[0096] In step S6, on the basis of completing the classification of the driver's global speed evolution pattern, the driver's global speed evolution pattern architecture is constructed based on the graph neural network, and the graph neural network is combined with CNN-LSTM to further construct the driver's global speed evolution prediction model, that is, the MTK-GNN-CNN-LSTM global speed evolution prediction model, to predict its speed evolution trend, thereby predicting its driving behavior, and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limits.

[0097] In one embodiment of the present invention, in step S4, the K-Shape clustering algorithm is used to eliminate the amplitude influence of the time series by normalization, local alignment is achieved by moving the time series, and the fast Fourier algorithm is introduced in the calculation of similarity to reduce the calculation cost of the high-dimensional time series. The K-Shape algorithm combines the concepts of DTW distance and shape features and can effectively capture the intrinsic structure of the time series; step S4 includes the following steps:

[0098] Step S41, unifying the length of vehicle time series trajectory data and the distribution of acquisition point coordinates; since the radar acquisition frequency is fixed, vehicles of different speeds take different time lengths to pass through the same road section, resulting in different lengths of speed time series data. In order to achieve matching based on vehicle position in the clustering process of different time series, the speed time series data of each vehicle uses the longitudinal position as the "time series" and performs interpolation processing at a spacing of 1m, and processes the speed time series data of each vehicle into equal-length time series data;

[0099] Step S42, initialize cluster centers: randomly select K time series from the data set as initial cluster centers. These series will become representative shapes of clusters; set the number of clusters to 3;

[0100] Step S43, shape distance measurement: using DTW or other shape distance measurement to calculate the distance between each time series and each cluster center;

[0101] Step S44, assigning data points: assigning each time series to the cluster with the closest cluster center; thus, each series will be classified into the cluster with the most similar shape to it;

[0102] Step S45, update cluster centers: update cluster centers using the average shape of the time series assigned to each cluster. In this way, cluster centers will be continuously adjusted to better represent the shape characteristics of the cluster;

[0103] Step S46, repeat iteration: Repeat the above steps until the cluster center no longer changes and reaches a predetermined number of iterations; eventually, the obtained cluster will contain all time series in the data set, and each cluster represents a group of sequences with similar shapes.

[0104] In one embodiment of the present invention, in step S5, a global speed evolution model of the driver under the progressive lane-level variable speed limit control is constructed, the global speed of the driver's vehicle is segmented, and the state transition probability between different trajectory segments is obtained, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit; step S5 includes the following steps:

[0105] Step S51: discretize and perform quantile discretization processing on the vehicle speed time series to obtain vehicle speed discretization partitions.

[0106] Step S52: Establish a Markov transition matrix for the partitioned time series data to obtain the matrix M1

[0107]

[0108] Step S53, normalize the values ​​and calculate the conversion probability matrix M′1:

[0109]

[0110] Step S54, calculating the Markov transition field;

[0111] Step S55, the time-series segmented subsequences processed by the Markov transition field are encoded and mapped by a graph neural network, and a graph neural network is further constructed to describe different speed fluctuation states and the relationship between them;

[0112] Step S56, the module partitioning algorithm in the TSIA graph neural network analysis tool developed by the Massachusetts Institute of Technology can be used to perform community partitioning on the obtained graph neural network; the driver's global speed state is divided into different state intervals, and the edges between the neural vertices represent the conversion probability between different speed state intervals; the process of the driver making decisions in the face of speed limits is divided into several state intervals as needed, so as to more finely describe the driver's global speed evolution process.

[0113] The present invention further discloses a classification and prediction system for the global speed evolution pattern of a driver under a progressive variable speed limit. Figure 2 FIG. 1 is a schematic diagram of the composition of an evolutionary pattern classification and prediction system in one embodiment of the present invention; please refer to FIG. Figure 2 The pattern classification and prediction system includes: a number of information publishing devices 1, a number of vehicle trajectory data acquisition devices 2, a speed limit value adjustment module 3, a classification module 4, a speed evolution model construction module 5 and a speed evolution prediction model construction module 6.

[0114] The information release device 1 is arranged on a high-speed road section with a free flow state, and is arranged in sequence according to a set spacing rule; in one embodiment of the present invention, the information release device is a lane-level variable speed limit information release device;

[0115] The vehicle trajectory data acquisition device 2 is arranged in the road area in sequence according to the set spacing rules to realize the collection and analysis of vehicle time-series trajectory data. The vehicle trajectory data acquisition device includes a radar and / or a camera to obtain continuous and uninterrupted sub-meter-level time-series vehicle trajectories, with a speed acquisition accuracy of not less than 1km / h and an acquisition frequency of 10Hz, providing a data basis for the collection and analysis of the driver's real-time time-series driving behavior characteristics.

[0116] The speed limit value adjustment module 3 is used to use the variable information system to issue and control the corresponding lane-level speed limit control instructions according to the preset control strategy, adjust the speed limit value of the lane-level variable speed limit plate, and export the vehicle trajectory data and lane-level speed limit information release data in real time. In one embodiment, the vehicle time-series trajectory data and lane-level speed limit information release data are all real-time collected data and use UTC time, and the location of data release uses a unified stake number coordinate;

[0117] The classification module 4 is used to cluster the driver's global vehicle speed trajectory pattern and classify the driver's vehicle trajectory according to different time series distribution characteristics, thereby classifying the driver under the progressive variable speed limit.

[0118] The speed evolution model construction module 5 is used to construct a driver's global speed evolution model under progressive lane-level variable speed limit control, segment the driver's vehicle global speed, obtain the state transition probability between different trajectory segments, so as to effectively judge the driver's abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

[0119] The speed evolution prediction model construction module 6 is used to construct a driver's global speed evolution prediction model under lane-level variable speed limit control. According to the driver's global speed distribution characteristics under progressive lane-level variable speed limit control, the speed evolution trend is predicted, thereby predicting the driver's driving behavior and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

[0120] The speed evolution prediction model construction module 6, based on the classification of the driver's global speed evolution pattern, constructs a driver's global speed evolution pattern architecture based on the graph neural network, and uses the graph neural network combined with CNN-LSTM to further construct a driver's global speed evolution prediction model, that is, a global speed evolution prediction model of MTK-GNN-CNN-LSTM, to predict the driver's speed evolution trend, thereby predicting his driving behavior, and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limits.

[0121] In one embodiment of the present invention, the classification module 4 adopts the K-Shape clustering algorithm, eliminates the amplitude influence of the time series by normalization, realizes local alignment by moving the time series, and introduces the fast Fourier algorithm in calculating the similarity to reduce the calculation cost of the high-dimensional time series. The K-Shape algorithm combines the concepts of DTW distance and shape features, and can effectively capture the intrinsic structure of the time series; the classification process of the classification module 4 includes the following steps:

[0122] Step S41, unifying the length of vehicle time series trajectory data and the distribution of acquisition point coordinates; since the radar acquisition frequency is fixed, vehicles of different speeds take different time lengths to pass through the same road section, resulting in different lengths of speed time series data. In order to achieve matching based on vehicle position in the clustering process of different time series, the speed time series data of each vehicle uses the longitudinal position as the "time series" and performs interpolation processing at a spacing of 1m, and processes the speed time series data of each vehicle into equal-length time series data;

[0123] Step S42, initialize cluster centers: randomly select K time series from the data set as initial cluster centers. These series will become representative shapes of clusters; set the number of clusters to 3;

[0124] Step S43, shape distance measurement: using DTW or other shape distance measurement to calculate the distance between each time series and each cluster center;

[0125] Step S44, assigning data points: assigning each time series to the cluster with the closest cluster center; thus, each series will be classified into the cluster with the most similar shape to it;

[0126] Step S45, update cluster centers: update cluster centers using the average shape of the time series assigned to each cluster. In this way, cluster centers will be continuously adjusted to better represent the shape characteristics of the cluster;

[0127] Step S46, repeat iteration: Repeat the above steps until the cluster center no longer changes and reaches a predetermined number of iterations; eventually, the obtained cluster will contain all time series in the data set, and each cluster represents a group of sequences with similar shapes.

[0128] In one embodiment of the present invention, the model building process of the velocity evolution model building module 5 includes the following steps:

[0129] Step S51: discretize and perform quantile discretization processing on the vehicle speed time series to obtain vehicle speed discretization partitions.

[0130] Step S52: Establish a Markov transition matrix for the partitioned time series data to obtain the matrix M1

[0131]

[0132] Step S53, normalize the values ​​and calculate the conversion probability matrix M′1:

[0133]

[0134] Step S54, calculating the Markov transition field;

[0135] Step S55, the time-series segmented subsequences processed by the Markov transition field are encoded and mapped by a graph neural network, and a graph neural network is further constructed to describe different speed fluctuation states and the relationship between them;

[0136] Step S56, the module partitioning algorithm in the TSIA graph neural network analysis tool developed by the Massachusetts Institute of Technology (of course, other tools can also be used) can be used to perform community partitioning processing on the obtained graph neural network; the driver's global speed state is divided into different state intervals, and the edges between the neural vertices represent the conversion probability between different speed state intervals; the process of the driver making decisions in the face of speed limits is divided into several state intervals as needed, so as to more finely describe the driver's global speed evolution process.

[0137] In one embodiment of the present invention, the method for classifying and predicting the global speed evolution pattern of a driver under a progressive variable speed limit of the present invention comprises the following steps:

[0138] Step 1) A series of lane-level variable speed limit information release devices are arranged at a certain interval on a high-speed expressway section with free flow status.

[0139] The alignment of the road needs to comply with the requirements of the Highway Alignment Design Specifications (JTG D20-2017); the setting of signs and markings in the scene needs to comply with the requirements of the Road Traffic Signs and Markings (GB 5768-2009) and the Highway Speed ​​Limit Sign Design Specifications (JTG / T3381-02-2020). Figure 4 shown.

[0140] Step 2) Deploy vehicle trajectory data collection devices at a certain interval on the road area to achieve continuous and uninterrupted sub-meter level time-series vehicle trajectory collection and analysis.

[0141] The speed acquisition accuracy is not less than 1km / h, the acquisition frequency reaches 10Hz, and the acquisition equipment is arranged as follows: Figure 5 shown.

[0142] Step 3) According to the preset control strategy, the variable information system is used to issue and control the corresponding lane-level speed limit control instructions, adjust the speed limit value of the lane-level variable speed limit plate, and export the traffic flow parameters and lane-level speed limit information release data in real time.

[0143] The lane-level variable speed limit board is an information instruction issuing device for the variable speed limit control system. It is arranged on the gantry at a spacing of 750m. Each lane-level variable speed limit board issues variable speed limit information for one lane. The lane-level speed limit instructions on different lanes and upstream and downstream can have different speed limit values. By changing and adjusting the speed limit value of the variable speed limit sign in real time, the operating state of the road traffic flow can be changed in real time.

[0144] Step 4) Based on the K-Shape-based clustering of the driver's global speed trajectory pattern, the driver's vehicle trajectory is divided into three categories according to different time series distribution characteristics, thereby dividing the driver under the progressive variable speed limit into three categories:

[0145] Specifically, the length of vehicle time series trajectory data and the distribution of acquisition point coordinates are unified. Since the radar acquisition frequency is fixed, vehicles of different speeds take different time to pass the same road section, resulting in different lengths of speed time series data. In order to achieve matching based on vehicle position in the clustering process of different time series, the speed time series data of each vehicle uses the longitudinal position as the "time series" and interpolates at a spacing of 1m, processing the speed time series data of each vehicle into equal-length time series data.

[0146] Initialize cluster centers: Randomly select K time series from the data set as initial cluster centers. These series will become the representative shapes of the clusters; set the number of clusters to 3.

[0147] Shape distance metric: Calculate the distance between each time series and each cluster center using DTW or other shape distance metric.

[0148] Assign data points: Assign each time series to the cluster with the closest cluster center. In this way, each series will be classified into the cluster with the most similar shape to it.

[0149] Update cluster centers: Update the cluster centers using the average shape of the time series assigned to each cluster. In this way, the cluster centers will be continuously adjusted to better represent the shape characteristics of the cluster.

[0150] Repeat iteration: Repeat the above steps until the cluster center no longer changes and reaches the predetermined number of iterations. Eventually, the resulting cluster will contain all time series in the data set, and each cluster represents a group of sequences with similar shapes.

[0151] Finally, the speed distribution diagrams of the three speed trajectory modes are obtained, as shown in Figure 6 As shown in the figure, the comparison of vehicle speed distribution of three global vehicle speed trajectory modes under different speed limit scenarios is shown in the figure. Figure 7 shown.

[0152] Step 5) construct a global speed evolution model of the driver under the progressive lane-level variable speed limit control, segment the global speed of the driver and vehicle, and obtain the state transition probability between different trajectory segments, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit;

[0153] The vehicle speed time series is discretized and processed into discrete quantiles to obtain the vehicle speed discretization partitions, such as Figure 8 As shown;

[0154] The Markov transition matrix is ​​established for the partitioned time series data to obtain the matrix M1, as shown in Fig. 9 As shown;

[0155]

[0156] Specifically, the values ​​are normalized and the transition probability matrix M′1 ​​is calculated;

[0157]

[0158] Computes a Markov transition field.

[0159] The time-series segmented subsequences processed by the Markov transition field are encoded and mapped by the graph neural network, and the graph neural network is further constructed to describe the different speed fluctuation states and their relationships, such as Fig.10 shown.

[0160] The obtained graph neural network is processed by community partitioning using the module partitioning algorithm in the ts ia graph neural network analysis tool developed by MIT. The driver's global speed state is divided into different state intervals, and the edges between neural vertices represent the conversion probability between different speed state intervals. The process of the driver making decisions in the face of speed limits can be divided into several state intervals as needed, so as to more finely describe the driver's global speed evolution process, such as Fig.11 shown.

[0161] Step 6) Construct a prediction model for the driver's global speed evolution under lane-level variable speed limit control. According to the driver's global speed distribution characteristics under progressive lane-level variable speed limit control, predict the speed evolution trend, thereby predicting the driver's driving behavior and actively predicting the occurrence of abnormal driving behavior under progressive lane-level variable speed limit.

[0162] Specifically, on the basis of completing the classification of the driver's global speed evolution pattern, the driver's global speed evolution pattern architecture is constructed based on the graph neural network, and the graph neural network is combined with CNN-LSTM to further construct the driver's global speed evolution prediction model, that is, the MTK-GNN-CNN-LSTM global speed evolution prediction model, which predicts the speed evolution trend and thus predicts the driving behavior. It actively predicts the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limits. Its prediction accuracy is compared with the other two models. Fig.12 Shown

[0163] The lane-level variable speed limit group compliance prediction method for cross-sectional traffic flow parameters provided by the present invention applies cross-sectional traffic flow characteristic data to realize the prediction of downstream driver group compliance, solves the problem that traditional traffic variable speed limit control cannot optimize the speed limit according to the driver's compliance, and at the same time solves the problem that most of the current roads do not have the function of vehicle trajectory collection but only have the function of cross-sectional traffic flow data collection. Limited cross-sectional traffic flow characteristic parameters can be used to predict compliance, and the cost is relatively low.

[0164] Table 1 Model error and performance table

[0165]

[0166] The present invention also discloses an electronic device, Fig.13 FIG. 1 is a schematic diagram of the composition of an electronic device in an embodiment of the present invention; please refer to FIG. Fig.13 At the hardware level, the electronic device includes a memory, a processor and at least one network interface; the processor may be a microprocessor, and the memory may include a memory, such as a random access memory (RAM), or a non-volatile memory, etc. Of course, the electronic device may also be provided with other hardware as required.

[0167] The processor, network interface and memory may be interconnected via an internal bus, which may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus may include an address bus, a data bus, a control bus, etc. The memory is used to store programs (which may include operating system programs and application programs); the program may include program code, which may include computer operation instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0168] In one embodiment, the processor can read the corresponding program from the non-volatile memory into the memory and then run it; the processor can execute the program stored in the memory and is specifically used to perform the following operations (such as Figure 1 shown):

[0169] [Step S1] A series of lane-level variable speed limit information release devices are arranged at set intervals on a free-flow highway section;

[0170] [Step S2] Deploy vehicle trajectory data collection devices at set intervals on the road area to collect and analyze vehicle time series trajectory data;

[0171] [Step S3] According to the preset control strategy, the variable information system is used to issue and control the corresponding lane-level speed limit control command, adjust the speed limit value of the lane-level variable speed limit plate, and export the vehicle trajectory data and lane-level speed limit information release data in real time;

[0172] [Step S4] Based on the clustering of the driver's global vehicle speed trajectory pattern, the driver's vehicle trajectory is classified according to different time series distribution characteristics, so as to classify the driver under the progressive variable speed limit;

[0173] [Step S5] constructing a global speed evolution model of the driver under the progressive lane-level variable speed limit control, segmenting the global speed of the driver's vehicle, and obtaining the state transition probability between different trajectory segments, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit;

[0174] [Step S6] Construct a prediction model for the driver's global speed evolution under lane-level variable speed limit control, and predict the driver's speed evolution trend based on the driver's global speed distribution characteristics under progressive lane-level variable speed limit control, so as to predict the driver's driving behavior and actively predict the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

[0175] The present invention further discloses a storage medium having computer program instructions stored thereon, which, when executed by a processor, implements the following steps of the method of the present invention (eg Figure 1 shown):

[0176] [Step S1] A series of lane-level variable speed limit information release devices are arranged at set intervals on a free-flow highway section;

[0177] [Step S2] Deploy vehicle trajectory data collection devices at set intervals on the road area to collect and analyze vehicle time series trajectory data;

[0178] [Step S3] According to the preset control strategy, the variable information system is used to issue and control the corresponding lane-level speed limit control command, adjust the speed limit value of the lane-level variable speed limit plate, and export the vehicle trajectory data and lane-level speed limit information release data in real time;

[0179] [Step S4] Based on the clustering of the driver's global vehicle speed trajectory pattern, the driver's vehicle trajectory is classified according to different time series distribution characteristics, so as to classify the driver under the progressive variable speed limit;

[0180] [Step S5] constructing a global speed evolution model of the driver under the progressive lane-level variable speed limit control, segmenting the global speed of the driver's vehicle, and obtaining the state transition probability between different trajectory segments, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit;

[0181] [Step S6] Construct a prediction model for the driver's global speed evolution under lane-level variable speed limit control, and predict the driver's speed evolution trend based on the driver's global speed distribution characteristics under progressive lane-level variable speed limit control, so as to predict the driver's driving behavior and actively predict the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

[0182] To sum up, the classification and prediction method of the driver's global speed evolution pattern under a progressive variable speed limit proposed in the present invention can use the driver's speed evolution under the progressive variable speed limit condition to predict the driver's global speed, and provide a customized personalized driving behavior warning solution for the driver. It does not rely on vehicle-side equipment, has low cost, and is replicable and promotable and has strong robustness.

[0183] The method of the present invention utilizes a continuous vehicle trajectory acquisition device to collect and analyze the time series data of driving behavior under progressive lane-level variable speed limits, and utilizes K-Shape shape-based clustering to realize the prediction of the driver's speed evolution under progressive variable speed limit conditions, thereby solving the problems that the continuous lane-level speed limit scenarios are ever-changing, the combination methods cannot be exhausted, resulting in unclear mutual influence mechanism between lane speed limit measures and driving behavior, resulting in the inability to predict the speed distribution characteristics of the driver under continuous speed limit conditions, and the traditional traffic variable speed limit control cannot perform targeted speed limit according to the driver's driving behavior characteristics.

[0184] The present invention utilizes a graph neural network model to discretize time series fluctuations into different subsequence fluctuations, thereby achieving accurate quantification of the global speed evolution law and state transition probability of an individual driver under continuous speed limit conditions, accurately extracting the driver's abnormal speed state, and finely describing the evolution of the driver's vehicle speed state under continuous speed limit. This provides a theoretical basis and analysis method for refined variable speed limit control, and can provide personalized speed limit guidance strategies for driver behavior characteristics, with the characteristics of low cost, replicability and strong robustness.

[0185] On the basis of completing the classification of the driver's global speed evolution pattern, the Markov transition field and graph neural network combined with CNN-LSTM are used to further construct the MTK-GNN-CNN-LSTM driver's global speed evolution prediction model architecture. Compared with the conventional CNN-LSTM model, the global speed evolution pattern label obtained based on the Markov transition field-graph neural network community partitioning algorithm is added to the CNN model input. The subsequence dependency and conversion probability of the time series can be input in the input layer to improve the accuracy and stability of the global speed prediction.

[0186] It should be noted that the present application can be implemented in software and / or a combination of software and hardware; for example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of the present application can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including related data structures) can be stored in a computer-readable recording medium; for example, a RAM memory, a magnetic or optical drive or a floppy disk and the like. In addition, some steps or functions of the present application can be implemented in hardware; for example, as a circuit that cooperates with a processor to perform various steps or functions.

[0187] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0188] The description and application of the present invention here are illustrative, and it is not intended to limit the scope of the present invention to the above-mentioned embodiments. The effects or advantages involved in the embodiments may not be embodied in the embodiments due to interference from various factors, and the description of the effects or advantages is not used to limit the embodiments. The deformation and change of the embodiments disclosed here are possible, and the replacement of the embodiments and the various equivalent parts are well known to those of ordinary skill in the art. It should be clear to those skilled in the art that the present invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials and parts without departing from the spirit or essential features of the present invention. Other deformations and changes can be made to the embodiments disclosed here without departing from the scope and spirit of the present invention.

Claims

1. A method for classifying and predicting the global speed evolution pattern of a driver under a progressive variable speed limit, characterized in that: The pattern classification and prediction method comprises: Step S1, deploying a series of lane-level variable speed limit information publishing devices at set intervals on a high-speed expressway section with a free flow state; Step S2: deploying vehicle trajectory data collection devices at set intervals on the road area to collect and analyze vehicle time-series trajectory data; Step S3: according to the preset control strategy, the variable information system is used to issue and control the corresponding lane-level speed limit control command, the speed limit value of the lane-level variable speed limit plate is adjusted, and the vehicle trajectory data and lane-level speed limit information release data are exported in real time; Step S4: based on the clustering of the driver's global vehicle speed trajectory pattern, the driver's vehicle trajectory is classified according to different time series distribution characteristics, so as to classify the driver under the progressive variable speed limit; Step S5: constructing a global speed evolution model of the driver under the progressive lane-level variable speed limit control, segmenting the global speed of the driver's vehicle, and obtaining the state transition probability between different trajectory segments, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit; Step S6: construct a prediction model for the global speed evolution of the driver under lane-level variable speed limit control, and predict the speed evolution trend according to the global speed distribution characteristics of the driver under progressive lane-level variable speed limit control, so as to predict the driver's driving behavior and actively predict the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

2. The method for classifying and predicting the global speed evolution pattern of a driver under a progressive variable speed limit according to claim 1 is characterized by: In step S1, a lane-level variable speed limit information release device is arranged on the expressway section; In step S2, radars, cameras and other vehicle trajectory data collection devices are deployed at set intervals on the road area to obtain continuous and uninterrupted sub-meter-level time-series vehicle trajectories, with a speed collection accuracy of no less than 1km / h and a collection frequency of 10Hz, providing a data basis for the collection and analysis of the driver's real-time time-series driving behavior characteristics; In step S3, the vehicle time-series trajectory data and lane-level speed limit information release data are all collected in real time and use UTC time, and the location of data release uses a unified stake number coordinate; In step S6, on the basis of completing the classification of the driver's global speed evolution pattern, the driver's global speed evolution pattern architecture is constructed based on the graph neural network, and the graph neural network is combined with CNN-LSTM to further construct the driver's global speed evolution prediction model, that is, the MTK-GNN-CNN-LSTM global speed evolution prediction model, to predict its speed evolution trend, thereby predicting its driving behavior, and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limits.

3. The method for classifying and predicting the global speed evolution pattern of a driver under a progressive variable speed limit according to claim 1 is characterized by: In step S4, the K-Shape clustering algorithm is used to eliminate the amplitude influence of the time series through normalization, local alignment is achieved by moving the time series, and the fast Fourier algorithm is introduced in the calculation of similarity to reduce the calculation cost of high-dimensional time series. The K-Shape algorithm combines the concepts of DTW distance and shape features and can effectively capture the intrinsic structure of the time series; step S4 includes the following steps: Step S41, unifying the length of vehicle time series trajectory data and the distribution of acquisition point coordinates; since the radar acquisition frequency is fixed, vehicles of different speeds take different time lengths to pass through the same road section, resulting in different lengths of speed time series data. In order to achieve matching based on vehicle position in the clustering process of different time series, the speed time series data of each vehicle uses the longitudinal position as the "time series" and performs interpolation processing at a spacing of 1m, and processes the speed time series data of each vehicle into equal-length time series data; Step S42, initializing cluster centers: randomly selecting K time series from the data set as initial cluster centers. These sequences will become representative shapes of the clusters; Set the number of clusters to 3; Step S43, shape distance measurement: using DTW or other shape distance measurement to calculate the distance between each time series and each cluster center; Step S44, assigning data points: assigning each time series to the cluster with the closest cluster center; thus, each series will be classified into the cluster with the most similar shape to it; Step S45, updating cluster centers: updating cluster centers using the average shape of the time series assigned to each cluster. In this way, the cluster center will be continuously adjusted to better represent the shape characteristics of the cluster; Step S46, repeat iteration: Repeat the above steps until the cluster center no longer changes and reaches a predetermined number of iterations; eventually, the obtained cluster will contain all time series in the data set, and each cluster represents a group of sequences with similar shapes.

4. The method for classifying and predicting the global speed evolution pattern of a driver under a progressive variable speed limit according to claim 1 is characterized by: In step S5, a global speed evolution model of the driver under the progressive lane-level variable speed limit control is constructed, the global speed of the driver and vehicle is segmented, and the state transition probability between different trajectory segments is obtained, so as to effectively judge the abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit; step S5 includes the following steps: Step S51: discretize and perform quantile discretization processing on the vehicle speed time series to obtain vehicle speed discretization partitions. Step S52: Establish a Markov transition matrix for the partitioned time series data to obtain the matrix M1 Step S53, normalize the values ​​and calculate the conversion probability matrix M′1: Step S54, calculating the Markov transition field; Step S55, the time-series segmented subsequences processed by the Markov transition field are encoded and mapped by a graph neural network, and a graph neural network is further constructed to describe different speed fluctuation states and the relationship between them; Step S56, use the module partitioning algorithm in the tsia graph neural network analysis tool to perform community partitioning on the obtained graph neural network; the driver's global speed state is divided into different state intervals, and the edges between the neural vertices represent the conversion probability between different speed state intervals; the process of the driver making decisions in the face of speed limits is divided into several state intervals as needed, so as to more finely describe the driver's global speed evolution process.

5. A classification and prediction system for the global speed evolution pattern of drivers under progressive variable speed limit, characterized in that: The pattern classification and prediction system comprises: A number of information release devices are arranged on a high-speed road section with a free-flow state, and are arranged in sequence according to a set spacing rule; A plurality of vehicle trajectory data collection devices are arranged in sequence on the road area according to a set spacing rule to realize the collection and analysis of vehicle time series trajectory data; The speed limit value adjustment module is used to use the variable information system to issue and control the corresponding lane-level speed limit control instructions according to the preset control strategy, adjust the speed limit value of the lane-level variable speed limit plate, and export the vehicle trajectory data and lane-level speed limit information release data in real time; A classification module is used to cluster the driver's global speed trajectory pattern and classify the driver's vehicle trajectory according to different time series distribution characteristics, so as to classify the driver under the progressive variable speed limit; The speed evolution model construction module is used to construct the driver's global speed evolution model under the progressive lane-level variable speed limit control, divide the driver's vehicle global speed into segments, and obtain the state transition probability between different trajectory segments, so as to effectively judge their abnormal driving behavior and actively perceive the occurrence of abnormal driving behavior of the driver under the progressive lane-level variable speed limit; The speed evolution prediction model construction module is used to construct a driver's global speed evolution prediction model under lane-level variable speed limit control. According to the driver's global speed distribution characteristics under progressive lane-level variable speed limit control, its speed evolution trend is predicted, thereby predicting its driving behavior and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limit.

6. The classification and prediction system for the global speed evolution pattern of a driver under a progressive variable speed limit according to claim 5 is characterized by: The information release device is a lane-level variable speed limit information release device; The vehicle trajectory data acquisition device includes a radar and / or a camera to obtain continuous and uninterrupted sub-meter-level time-series vehicle trajectories, with a speed acquisition accuracy of no less than 1km / h and an acquisition frequency of 10Hz, providing a data basis for the acquisition and analysis of the driver's real-time time-series driving behavior characteristics; The vehicle time-series trajectory data and lane-level speed limit information release data are all collected in real time and use UTC time, and the locations where the data is released use unified stake number coordinates; The speed evolution prediction model construction module, based on the classification of the driver's global speed evolution pattern, further constructs the driver's global speed evolution prediction model based on the driver's global speed evolution pattern architecture constructed by the graph neural network, and uses the graph neural network combined with CNN-LSTM, that is, the MTK-GNN-CNN-LSTM global speed evolution prediction model, to predict the driver's speed evolution trend, thereby predicting the driver's driving behavior, and actively predicting the occurrence of abnormal driving behavior of the driver under progressive lane-level variable speed limits.

7. The classification and prediction system for the global speed evolution pattern of a driver under a progressive variable speed limit according to claim 5 is characterized by: The classification module adopts the K-Shape clustering algorithm, eliminates the amplitude influence of the time series by normalization, realizes local alignment by moving the time series, and introduces the fast Fourier algorithm in calculating similarity to reduce the calculation cost of high-dimensional time series. The K-Shape algorithm combines the concepts of DTW distance and shape features and can effectively capture the intrinsic structure of the time series. The classification process of the classification module includes the following steps: Step S41, unifying the length of vehicle time series trajectory data and the distribution of acquisition point coordinates; since the radar acquisition frequency is fixed, vehicles of different speeds take different time lengths to pass through the same road section, resulting in different lengths of speed time series data. In order to achieve matching based on vehicle position in the clustering process of different time series, the speed time series data of each vehicle uses the longitudinal position as the "time series" and performs interpolation processing at a spacing of 1m, and processes the speed time series data of each vehicle into equal-length time series data; Step S42, initializing cluster centers: randomly selecting K time series from the data set as initial cluster centers. These sequences will become representative shapes of the clusters; Set the number of clusters to 3; Step S43, shape distance measurement: using DTW or other shape distance measurement to calculate the distance between each time series and each cluster center; Step S44, assigning data points: assigning each time series to the cluster with the closest cluster center; thus, each series will be classified into the cluster with the most similar shape to it; Step S45, updating cluster centers: updating cluster centers using the average shape of the time series assigned to each cluster. In this way, the cluster center will be continuously adjusted to better represent the shape characteristics of the cluster; Step S46, repeat iteration: Repeat the above steps until the cluster center no longer changes and reaches a predetermined number of iterations; eventually, the obtained cluster will contain all time series in the data set, and each cluster represents a group of sequences with similar shapes.

8. The classification and prediction system for the global speed evolution pattern of a driver under a progressive variable speed limit according to claim 5 is characterized by: The model building process of the velocity evolution model building module includes the following steps: Step S51: discretize and perform quantile discretization processing on the vehicle speed time series to obtain vehicle speed discretization partitions. Step S52: Establish a Markov transition matrix for the partitioned time series data to obtain a matrix M1: Step S53, normalize the values ​​and calculate the conversion probability matrix M′1: Step S54, calculating the Markov transition field; Step S55, the time-series segmented subsequences processed by the Markov transition field are encoded and mapped by a graph neural network, and a graph neural network is further constructed to describe different speed fluctuation states and the relationship between them; Step S56, use the module partitioning algorithm in the tsia graph neural network analysis tool to perform community partitioning on the obtained graph neural network; the driver's global speed state is divided into different state intervals, and the edges between the neural vertices represent the conversion probability between different speed state intervals; the process of the driver making decisions in the face of speed limits is divided into several state intervals as needed, so as to more finely describe the driver's global speed evolution process.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images