Car-following speed prediction method and device considering driver style and vehicle heterogeneity

Abstract

The invention relates to a car-following speed prediction method and device considering driver styles and vehicle heterogeneity, and the method comprises the steps: building a corresponding model between driver style feature data in a data set and the probability of driver style types based on a full-connection neural network; building a prediction model of historical traffic data, driver style feature data and target speed data of the following vehicle in the data set by using a bidirectional long-short-term memory network and the trained driver style prediction network; carrying out heterogeneity classification on the following vehicle and the following vehicle by utilizing the prediction model, and establishing a prediction neural network model of the following vehicle according to a vehicle heterogeneity combination; and training a prediction neural network model by using the data in the data set until the prediction neural network model meets a training stop condition, and generating a target prediction neural network model for predicting the speed of the following vehicle. According to the invention, the speed of the following vehicle can be predicted more accurately, and the accuracy of road traffic state pre-judgment is improved.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle trajectory prediction, and in particular to a method and device for predicting following speed taking into account driver style and vehicle heterogeneity. Background Art

[0002] Speed ​​prediction, as a core technology in intelligent transportation systems, provides fundamental support for traffic management, vehicle navigation, and accident prevention. Accurate speed prediction allows traffic management systems to optimize traffic signal timing strategies and improve road capacity. Navigation systems can leverage speed prediction to provide real-time data support, helping vehicles select optimal routes based on current traffic conditions, thereby improving driving efficiency and reducing fuel consumption. Furthermore, with the increasing adoption of ADAS (Advanced Driving Assistance Systems), accurate car-following speed prediction has become a critical component in ensuring system safety and reliability. Assisted driving technology uses real-time perception and prediction of surrounding vehicle speeds and distances to help vehicles maintain a reasonable following distance, smoothly adjust speed, and avoid frequent braking and acceleration, thereby enhancing the driving experience and safety. Accurate speed prediction can significantly reduce the risk of sudden braking and collisions, particularly on highways and in heavy traffic, providing drivers with a smoother driving environment. During normal driving, vehicles often switch between car-following and free-range driving. Free travel refers to driving without a preceding vehicle or at a considerable distance. Due to the inherent desire for efficiency, human drivers tend to accelerate to their desired speed and then maintain a constant speed during free travel, making speed prediction relatively straightforward. However, car-following, when driving under the influence of the preceding vehicle, requires adjusting the vehicle's speed to maintain a safe distance. This driving state is subject to significant uncertainty, making speed prediction more challenging. On urban roads and congested highways, vehicles spend a disproportionately large proportion of their time in car-following mode. Therefore, accurate car-following speed prediction can aid the application of intelligent transportation systems and holds significant research value.

[0003] Current car-following speed prediction models are mainly divided into two types: rule-based prediction and historical data-based prediction. Rule-based prediction, such as the Intelligent Driver Model (IDM), uses predefined fixed mathematical relationships or empirical rules to calculate vehicle speed or acceleration to predict car-following behavior. This type of model is usually based on a set of known physical or behavioral laws, which clearly stipulate how the driver should adjust the vehicle speed and maintain the distance under different traffic conditions. Prediction based on historical data relies on statistical analysis, time series analysis, or machine learning algorithms to learn the speed change patterns of the vehicle from historical patterns and automatically discover potential driving patterns and laws.

[0004] Related technologies assume that all driver behaviors and vehicle types are consistent. This assumption is overly idealistic and ignores differences in driver driving styles and vehicle heterogeneity. Because driver style is stable over time and does not change with changes in road traffic conditions, the classification and identification of driver style is highly robust. Driver style can be categorized in various ways, such as aggressive versus cautious, skilled versus unskilled, responsive versus insensitive, and so on. A driver's style significantly affects their driving behavior and, consequently, their following speed. For example, aggressive drivers tend to maintain a closer following distance and higher following speeds, while skilled drivers maintain a longer collision distance to ensure safety.

[0005] At the same time, vehicles use different following strategies when following large commercial vehicles and small passenger cars. Considering driving safety, when the speed of the preceding vehicle is the same, the following distance of the commercial vehicle is often greater than that of the passenger car, so the speed selection is also different.

[0006] Furthermore, both of the aforementioned prediction methods essentially only consider the impact of the preceding vehicle's driving state on the following vehicle. However, during the same time period, without considering lane changes, the driving state of the preceding vehicle will affect the preceding vehicle, and the following vehicle's state will in turn affect the following vehicle. However, current perception range based on single-vehicle intelligence is limited to vehicles surrounding the vehicle itself. Even if millimeter-wave radar can sense the preceding vehicle, the perception data is not accurate enough. With the development of vehicle-road-cloud integrated technology, the fusion of perception through road test equipment has expanded the vehicle's perception range. Simultaneously considering the driving state of both the preceding and preceding vehicles can improve the accuracy of following speed prediction. Summary of the Invention

[0007] The present application provides a method and device for predicting car-following speed that takes into account driver style and vehicle heterogeneity, in order to address the problem that related technologies assume that all driver behaviors and vehicle types are consistent. This assumption is too idealistic and ignores the differences in drivers' driving styles and vehicle heterogeneity.

[0008] The first aspect of the present application provides a method for predicting a following speed that takes into account driver style and vehicle heterogeneity, which is applied to the model building stage and includes the following steps: based on a fully connected neural network, establishing a correspondence model between driver style feature data in a data set and the probability of driver style type; based on the correspondence model, using a bidirectional long short-term memory network and a trained driver style prediction network, building a prediction model of historical traffic data in the data set, the driver style feature data and target speed data of the following vehicle; using the prediction model to perform heterogeneity classification on the following vehicle and the following vehicle to generate a vehicle heterogeneity combination, and establishing a prediction neural network model for the following vehicle based on the vehicle heterogeneity combination; using the data in the data set to train the prediction neural network model until the prediction neural network model meets the training stop condition, and then generating a target prediction neural network model for predicting the speed of the following vehicle.

[0009] Optionally, in one embodiment of the present application, before establishing a correspondence model between the driver style feature data in the data set and the probability of the driver style type, it also includes: judging the type of the preceding vehicle; if the type of the preceding vehicle is a commercial vehicle, generating the historical traffic data based on the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle; if the type of the preceding vehicle is a passenger car, generating the historical traffic data based on the speed of the preceding vehicle before following, the following distance between the preceding vehicle before following, the speed of the preceding vehicle before following, the following distance between the preceding vehicle before following, and the speed of the following vehicle.

[0010] Optionally, in one embodiment of the present application, the corresponding model between the driver style feature data in the data set and the probability of the driver style type is established based on the fully connected neural network, including: extracting the driver style feature data including headway, collision time, own vehicle speed, own vehicle acceleration and following distance from the data set; generating sample features based on the average value of the inverse collision time, the average value and standard deviation of the headway, the average value and standard deviation of the acceleration, and the standard deviation of the own vehicle speed and the mean of the following distance; performing dimensionality reduction processing on the features of the samples to generate reduced-dimensionality data, and clustering the features of the samples based on the reduced-dimensionality data to generate the driver style type; building the fully connected neural network, extracting feature data corresponding to the vehicle identification number at preset time intervals, using the feature data as the input of the fully connected neural network, and using the probability of the corresponding type label after clustering as the output of the fully connected neural network, establishing a driver style identification neural network with a fully connected structure, and generating the corresponding model based on the driver style identification neural network.

[0011] Optionally, in one embodiment of the present application, the construction of a prediction model of the historical traffic data, the driver style feature data and the target speed data of the following vehicle in the data set includes: based on the historical traffic data and the driver style feature data, using the bidirectional long short-term memory network to process the historical traffic time series sequence to generate a processed traffic time series sequence; based on the processed traffic time series sequence, using the driver style identification neural network to output the probability of the driver style, and based on the style probability, using the fully connected neural network to output the target speed data of the following vehicle, so as to build the prediction model according to the target speed data of the following vehicle.

[0012] Optionally, in one embodiment of the present application, the calculation formula of the headway is:

[0013]

[0014] Wherein, d is the relative distance between the following vehicle and the following vehicle, v e is the longitudinal speed of the vehicle;

[0015] The calculation formula of the collision time distance is:

[0016]

[0017] Wherein, d is the relative distance between the following vehicle and the following vehicle, v e is the longitudinal velocity of the vehicle, v p is the longitudinal speed of the preceding vehicle.

[0018] Optionally, in one embodiment of the present application, the use of the prediction model to perform heterogeneous classification on the following vehicle and the following vehicle to generate a vehicle heterogeneous combination includes: performing heterogeneous classification on the following vehicle and the following vehicle based on the speed, position information of the following vehicle and the speed, position, and driver characteristic information of the following vehicle to generate the vehicle heterogeneous combination.

[0019] A second aspect of the present application provides a method for predicting the following speed taking into account driver style and vehicle heterogeneity, which is applied in the model application stage and includes the following steps: obtaining a vehicle heterogeneity combination; inputting the vehicle heterogeneity combination into a pre-constructed prediction neural network model to predict the speed of the following vehicle, wherein the prediction neural network model is constructed by the vehicle heterogeneity combination.

[0020] In a third aspect, an embodiment of the present application provides a device for predicting a following speed that takes into account driver style and vehicle heterogeneity, which is applied to a model building stage and includes: an establishment module for establishing a correspondence model between driver style feature data in a data set and the probability of a driver style type based on a fully connected neural network; a construction module for building a prediction model of historical traffic data, the driver style feature data, and target speed data of a following vehicle in the data set using a bidirectional long short-term memory network and a trained driver style prediction network based on the correspondence model; a generation module for using the prediction model to perform heterogeneity classification on a following vehicle and a following vehicle to generate a vehicle heterogeneity combination, and to establish a prediction neural network model for the following vehicle based on the vehicle heterogeneity combination; a training module for training the prediction neural network model using the data in the data set until the prediction neural network model meets a training stop condition, and then generating a target prediction neural network model for predicting the speed of the following vehicle.

[0021] Optionally, in one embodiment of the present application, it further includes: a judgment module for judging the type of the preceding vehicle before establishing a correspondence model between the driver style feature data and the probability of the driver style type in the data set; a first data generation module for generating the historical traffic data according to the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle when the type of the preceding vehicle is a commercial vehicle; and a second data generation module for generating the historical traffic data according to the speed of the preceding vehicle, the following distance between the preceding vehicle and the preceding vehicle, the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle when the type of the preceding vehicle is a passenger car.

[0022] Optionally, in one embodiment of the present application, the establishment module includes: an extraction unit for extracting the driver style feature data including headway, collision time, own vehicle speed, own vehicle acceleration and following distance from the data set; a generation unit for generating sample features based on the average value of the inverse collision time, the average value and standard deviation of the headway, the average value and standard deviation of the acceleration, and the standard deviation of the own vehicle speed and the mean of the following distance; a clustering unit for performing dimensionality reduction processing on the features of the samples to generate reduced-dimensionality data, and clustering the features of the samples based on the reduced-dimensionality data to generate the driver style type; an establishment unit for building the fully connected neural network, extracting feature data corresponding to the vehicle identification number at preset time intervals, using the feature data as the input of the fully connected neural network, and using the probability of the corresponding type label after clustering as the output of the fully connected neural network, establishing a driver style identification neural network with a fully connected structure, and generating the corresponding model based on the driver style identification neural network.

[0023] Optionally, in one embodiment of the present application, the building module includes: a processing unit for processing historical traffic time series sequences using the bidirectional long short-term memory network based on the historical traffic data and the driver style feature data to generate a processed traffic time series sequence; a building unit for outputting the probability of the driver style using the driver style identification neural network based on the processed traffic time series sequence, and outputting the target speed data of the following vehicle using the fully connected neural network based on the style probability, so as to build the prediction model according to the target speed data of the following vehicle.

[0024] Optionally, in one embodiment of the present application, the calculation formula of the headway is:

[0025]

[0026] Wherein, d is the relative distance between the following vehicle and the following vehicle, v e is the longitudinal speed of the vehicle;

[0027] The calculation formula of the collision time distance is:

[0028]

[0029] Wherein, d is the relative distance between the following vehicle and the following vehicle, v e is the longitudinal velocity of the vehicle, v p is the longitudinal speed of the preceding vehicle.

[0030] Optionally, in one embodiment of the present application, the generation module includes: a combination generation unit, which is used to perform heterogeneity classification on the following vehicle and the following vehicle based on the speed, position information of the following vehicle and the speed, position, and driver characteristic information of the following vehicle to generate the vehicle heterogeneity combination.

[0031] The fourth aspect of the present application provides a device for predicting the following speed taking into account the driver's style and vehicle heterogeneity, which is applied to the model application stage and includes: an acquisition module for acquiring the vehicle heterogeneity combination; a prediction module for inputting the vehicle heterogeneity combination into a pre-built prediction neural network model to predict the speed of the following vehicle, wherein the prediction neural network model is constructed by the vehicle heterogeneity combination.

[0032] The fifth aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the following speed prediction method considering driver style and vehicle heterogeneity as described in the above embodiments.

[0033] This embodiment comprehensively considers differences in driver driving styles, vehicle heterogeneity, and the influence of the preceding vehicle on following behavior, as well as their coupling effects. This allows for more accurate prediction of the speed of the following vehicle, improving the accuracy of road traffic status prediction. This addresses the overly idealistic assumption in related technologies that all driver behaviors and vehicle types are consistent, ignoring differences in driver driving styles and vehicle heterogeneity.

[0034] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0036] Figure 1 This is a flowchart of a method for predicting car-following speed considering driver style and vehicle heterogeneity according to an embodiment of the present application applied to a model building stage;

[0037] Figure 2 This is an overall flow chart of a method for predicting following speed considering driver style and vehicle heterogeneity according to one embodiment of the present application;

[0038] Figure 3 A following relationship diagram of a following speed prediction method considering driver style and vehicle heterogeneity according to one embodiment of the present application;

[0039] Figure 4 This is a diagram showing the results of a principal component analysis method for predicting a following speed taking into account driver style and vehicle heterogeneity according to one embodiment of the present application;

[0040] Figure 5 This is a clustering result diagram of a following speed prediction method considering driver style and vehicle heterogeneity according to one embodiment of the present application;

[0041] Figure 6 This is a diagram showing a prediction model structure when the preceding vehicle is a commercial vehicle according to a method for predicting following speed that considers driver style and vehicle heterogeneity according to one embodiment of the present application;

[0042] Figure 7 This is a diagram showing a prediction model structure when the preceding vehicle is a passenger car according to a method for predicting following speed that considers driver style and vehicle heterogeneity according to one embodiment of the present application;

[0043] Figure 8 A flow chart of a method for predicting car-following speed considering driver style and vehicle heterogeneity according to an embodiment of the present application applied to a model application stage;

[0044] Figure 9 This is a schematic structural diagram of a car-following speed prediction device considering driver style and vehicle heterogeneity according to an embodiment of the present application, applied to a model building stage;

[0045] Figure 10 This is a schematic structural diagram of a car-following speed prediction device considering driver style and vehicle heterogeneity according to an embodiment of the present application, applied to a model application stage;

[0046] Figure 11 A schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0047] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0048] The following describes, with reference to the accompanying drawings, a method and device for predicting car-following speed that considers driver style and vehicle heterogeneity, according to an embodiment of the present application. In response to the related art mentioned in the background art, which assumes that all driver behaviors and vehicle types are consistent, this assumption is overly idealistic and ignores the differences in driver driving styles and vehicle heterogeneity. This application provides a car-following speed prediction method that considers driver style and vehicle heterogeneity. In this method, a K-means clustering method and an FCN neural network are used to establish a correspondence model between driver style feature data in a dataset and the probability of a given driver style type. A BiLSTM neural network is then integrated with a trained driver style prediction network to build a prediction model based on historical traffic data, driver style feature data, and future speed data of the following vehicle. The prediction model is then classified based on the heterogeneity of the vehicle and the preceding vehicle, and multiple corresponding prediction neural network models are established. Multiple training runs are performed using data from the HighD dataset to improve the accuracy of model predictions. This application can more accurately predict the speed of the following vehicle, improving the accuracy of road traffic status prediction. This method addresses the problem that related art assumes that all driver behaviors and vehicle types are consistent, which is overly idealistic and ignores the differences in driver driving styles and vehicle heterogeneity.

[0049] Specifically, Figure 1 A flow chart of a method for predicting car-following speed that takes into account driver style and vehicle heterogeneity, provided in an embodiment of the present application.

[0050] like Figure 1 As shown, the following speed prediction method considering driver style and vehicle heterogeneity includes the following steps:

[0051] In step S101 , a correspondence model between driver style feature data in a data set and the probability of driver style types is established based on a fully connected neural network.

[0052] In the actual implementation process, Figure 2 As shown, the embodiment of the present application can establish a correspondence model between the driver style feature data in the data set and the probability of the driver style type based on the K-means clustering method and the FCN fully connected neural network.

[0053] The HighD dataset was selected as the dataset, which contains trajectory data for over 110,000 vehicles. This data includes each vehicle's spatial position (x,y coordinates), velocity, acceleration, and other information at a temporal resolution of 25 frames per second. Considering the computational complexity and the need for real-time prediction, this application uses 7 seconds of historical data to predict the future velocity of the following vehicle within 4 seconds, with a step size of 0.2 seconds for each data point.

[0054] Optionally, in one embodiment of the present application, before establishing a correspondence model between the driver style feature data in the data set and the probability of the driver style type, it also includes: judging the type of the preceding vehicle; if the preceding vehicle is a commercial vehicle, generating historical traffic data based on the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle; if the preceding vehicle is a passenger car, generating historical traffic data based on the speed of the preceding vehicle before following, the following distance between the preceding vehicle before following, the speed of the preceding vehicle before following, the following distance between the preceding vehicle before following, and the speed of the preceding vehicle before following.

[0055] In the actual implementation process, the embodiment of the present application can first clean and classify the data set before building a prediction model of historical traffic data, driver style feature data and target speed data of the following vehicle in the data set. The present application can predict the future speed of all vehicles in the following state, where the following vehicles can be divided into two categories according to the number of following vehicles in front. One situation is that there is only one vehicle following the following vehicle in front, that is, the following vehicle in front is in a free driving state (hereinafter referred to as situation 1), and the other situation is that there are two vehicles following the following vehicle in front, that is, the following vehicle in front is also in a following state (hereinafter referred to as situation 2). The following relationship can be represented by Figure 3 express.

[0056] The embodiment of the present application can determine the type of the preceding vehicle. When the preceding vehicle is a commercial vehicle, historical traffic data is generated based on the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle. When the preceding vehicle is a passenger car, historical traffic data is generated based on the speed of the preceding vehicle before following, the following distance between the preceding vehicle before following, the speed of the preceding vehicle before following, the following distance between the preceding vehicle before following, and the speed of the following vehicle after following.

[0057] Optionally, in one embodiment of the present application, a correspondence model between driver style feature data in a data set and the probability of a driver style type is established based on a fully connected neural network, including: extracting driver style feature data including headway, collision time, own vehicle speed, own vehicle acceleration and following distance from the data set; generating sample features based on the average value of the inverse collision time, the average value and standard deviation of the headway, the average value and standard deviation of the acceleration, and the standard deviation of the own vehicle speed and the mean of the following distance; performing dimensionality reduction processing on the sample features to generate reduced-dimensionality data, and clustering the sample features based on the reduced-dimensionality data to generate the driver style type; building a fully connected neural network, extracting feature data corresponding to the vehicle identification number at preset time intervals, using the feature data as the input of the fully connected neural network, and using the probability of the corresponding type label after clustering as the output of the fully connected neural network, establishing a driver style identification neural network with a fully connected structure, and generating a corresponding model based on the driver style identification neural network.

[0058] It can be understood that the preset time interval in the embodiment of the present application can be an interval of 7 seconds, and the characteristic data corresponding to the vehicle identification number can be the characteristic data corresponding to the vehicle ID.

[0059] In the actual implementation process, the embodiment of the present application can build a driver style recognition network. The data describing the driver's characteristics corresponding to each vehicle is extracted from the HighD data set, including the time to collision (THW), time to collision (TTC), vehicle speed, vehicle acceleration and following distance. After the data is sorted according to the vehicle ID, the average value of the inverse time to collision (the inverse of the time to collision TTC, TTCi), the average value and standard deviation of the time to collision, the average value and standard deviation of the acceleration, the standard deviation of the vehicle speed and the mean of the following distance are calculated, and each value is used as a feature of the sample.

[0060] After the features are sorted, they need to be clustered to summarize the driver style type. Due to the large number of features and the large dimensionality of the data, the distances between data points become increasingly similar, making it difficult to distinguish different clusters. At the same time, they are also more affected by noise, so dimensionality reduction is required. Figure 4 As shown, the dimensionality reduction method selected by this application is principal component analysis. This method projects the original high-dimensional data into a low-dimensional space through three steps: data centering, calculating the covariance matrix, and eigenvalue decomposition to retain the main information of the data and reduce the dimension of the data so as to perform K-means clustering and obtain clustering results. After principal component analysis, the proportion of the first two principal components has exceeded 65%, indicating that these two dimensions can already represent the characteristics of the original data. The proportion of each original feature in the principal component is Figure 4 As shown, the larger the value, the greater the degree of positive correlation with the principal component. Figure 4It can be seen that the features with a relatively large proportion of the first principal component are TTCi mean, TTCi standard deviation, acceleration mean, acceleration standard deviation and speed mean. The collision time interval is mainly related to whether the driver is skilled. Skilled drivers can effectively avoid collisions, so the mean collision time interval will be larger and the reverse collision time interval will be smaller. At the same time, skilled drivers have more stable driving behavior, smaller fluctuations in collision time intervals, and smaller standard deviations, so the standard deviation of reverse collision time intervals is smaller. Acceleration symbolizes a comfort index. Skilled drivers are more comfortable driving, so the acceleration mean and standard deviation are smaller. Figure 5 As can be seen from the figure, the first principal component is positively correlated with the mean and standard deviation of the inverse collision time interval and the mean and standard deviation of the acceleration. Therefore, the first principal component represents the proficiency factor, with larger values ​​indicating a more proficient driver. Similarly, the second principal component is characterized by the mean, standard deviation, and mean headway interval. Headway interval and following distance indicate the driver's cautiousness in following a vehicle. More cautious drivers have smaller mean headway intervals and following distance intervals. Therefore, the second principal component represents the aggressive-cautious factor, with larger values ​​indicating a more cautious driver.

[0061] After principal component analysis of the above samples, the driver styles are divided into 5 categories. The clustering results are projected onto the three principal components as shown in the following figure: Figure 5 As shown in the figure, Cluster 1 is negatively correlated with the first principal component, indicating a skilled type, and positively correlated with the second principal component, indicating a cautious type. Therefore, Cluster 1 can be defined as a cautious-skilled driver. Similarly, Cluster 4 is negatively correlated with both the first and second principal components, so Cluster 4's driver style can be defined as an aggressive-skilled driver. Cluster 5 is significantly positively correlated with the first principal component, indicating an unskilled type, but its relationship with the second principal component is not significant, so Cluster 5 can be defined as an unskilled driver. Clusters 2 and 3 fall between Clusters 1, 4, and 5 on the first principal component, that is, between skilled and unskilled, and can be defined as standard. Cluster 2 is positively correlated on the second principal component, so Cluster 2 can be defined as a cautious-standard driver, while Cluster 1 is negatively correlated, indicating an aggressive-standard driver. The clustering of all driver styles is complete.

[0062] The primary data used to describe driver style during clustering is time to collision (TTC), headway (THW), and acceleration. Therefore, these feature data in the HighD dataset are matched one-to-one with the vehicle ID and clustered type number, and then organized. Since this application uses 7 seconds of historical data to predict the future speed of the following vehicle within 4 seconds, consistency is essential when identifying driver style. Therefore, feature data corresponding to each vehicle ID is extracted at 7-second intervals to form the neural network dataset. This 7-second feature data serves as the neural network input, and the probability of the corresponding type label after clustering serves as the neural network output. A fully connected neural network for driver style identification is constructed. With a step size of 0.2 seconds and three features, there are 105 input data points and 105 input neurons. The network has four hidden layers, each containing 100 neurons. The network outputs the probabilities of five types, resulting in five output neurons. By dividing the dataset into training and test sets in an 8:2 ratio, the recognition accuracy reaches 95%, demonstrating good accuracy.

[0063] This application comprehensively considers driver style and vehicle heterogeneity, using clustering to classify driver style characteristics. After dimensionality reduction using principal factor analysis, Kmeans clustering is used to identify features related to proficiency and aggressiveness, classifying driver styles into five categories. An FCN neural network is constructed, and historical feature data from a driver's driving over a 7-second period is input to determine the probability of the driver belonging to each of the five different styles.

[0064] In step S102, based on the corresponding model, a prediction model of historical traffic data, driver style feature data and target speed data of the following vehicle in the dataset is constructed using a bidirectional long short-term memory network and a trained driver style prediction network.

[0065] During the actual implementation process, the embodiment of the present application can be based on the corresponding model, using a bidirectional long short-term memory network (BiLSTM neural network) and a trained driver style prediction network to fuse together to build a prediction model based on historical traffic data, driver style feature data, and target speed data of the following vehicle in the dataset.

[0066] Optionally, in one embodiment of the present application, a prediction model is constructed based on historical traffic data, driver style feature data and target speed data of the following vehicle in a data set, including: based on the historical traffic data and driver style feature data, using a bidirectional long short-term memory network to process the historical traffic time series sequence to generate a processed traffic time series sequence; based on the processed traffic time series sequence, using a driver style recognition neural network to output the probability of the driver style, and based on the style probability, using a fully connected neural network to output the target speed data of the following vehicle, so as to build a prediction model based on the target speed data of the following vehicle.

[0067] It is understandable that the prediction model in the embodiment of the present application may be a complete car-following speed prediction network.

[0068] As a possible implementation method, the embodiment of the present application can build a complete car-following speed prediction network to identify the driver's style probability using historical data. The network structure is as follows: Figure 2 As shown. There are two types of inputs in the prediction network, one of which is historical traffic data, which mainly reflects the impact of historical traffic conditions on future following speeds. Since the main parameters that affect the speed of the following vehicle in the following process are the speed of the leading vehicle, the following distance between the leading vehicle and the following vehicle, and the speed of the following vehicle. By analogy, the following vehicle will be affected by the preceding vehicle, so the main parameters that affect the leading vehicle are the speed of the preceding vehicle before following and the following distance between the preceding vehicle before following. According to the above classification based on the heterogeneity of the following vehicle, when the following vehicle is a commercial vehicle, the preceding vehicle has little impact on the following vehicle. Therefore, when the following vehicle is a commercial vehicle, only the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle are selected as the input of historical traffic data. A total of 7s of data with a step size of 0.2s are used, with a total of 105 input neurons. The specific structure is as follows Figure 6 When the preceding vehicle is a passenger car, five parameters are selected as the historical traffic data input: the preceding vehicle speed, the following distance between the preceding vehicle and the preceding vehicle, the preceding vehicle speed, the following distance between the preceding vehicle and the following vehicle, and the following vehicle speed. A total of 7 seconds of data is used, with a step size of 0.2 seconds, and a total of 175 input neurons. The specific structure is as follows: Figure 7 The main network structure used is Bi-LSTM (Bidirectional Long Short-Term Memory Network). Bi-LSTM is an enhanced recurrent neural network (RNN) architecture that processes time series data in both directions (forward and backward) of the input sequence. In other words, it uses a Bi-LSTM network to process historical traffic time series and generate processed traffic time series, thereby capturing long-term dependencies in the sequence.

[0069] Another set of inputs is used for driver style recognition. These include driver style feature data: time to collision (TTC), headway (THW), and acceleration. The output is the probability of each driver style. These five probabilities are concatenated with the output of the Bi-LSTM network as input to build a fully connected network. The output is the speed of the following vehicle four seconds into the future, with a step size of 0.2 seconds and a total of 20 output neurons.

[0070] In one embodiment of the present application, the driver style feature data is used to identify the driver style probability. The headway and collision headway are required in the process of driver style identification. The calculation formula of the headway is:

[0071]

[0072] Where d is the relative distance between the following vehicle and the preceding vehicle, v e is the longitudinal speed of the vehicle;

[0073] The calculation formula for collision time is:

[0074]

[0075] Where d is the relative distance between the following vehicle and the preceding vehicle, v e is the longitudinal velocity of the vehicle, v p is the longitudinal speed of the preceding vehicle.

[0076] According to the calculation formula of headway and collision distance, it is only necessary to obtain the longitudinal speed of the vehicle and the preceding vehicle and the relative distance between the two.

[0077] Therefore, the inputs to this model are the longitudinal velocity of the following vehicle, the longitudinal acceleration of the following vehicle, the vehicle type of the following vehicle, the relative distance between the following vehicle and the leading vehicle, the longitudinal velocity of the leading vehicle, the vehicle type of the following vehicle, the relative distance between the following vehicle and the preceding vehicle, the relative velocity of the preceding vehicle, and the vehicle type of the preceding vehicle. This totals 280 data points. These are fed into the model as input, and the output is the following velocity for the next 4 seconds.

[0078] In step S103, the prediction model is used to perform heterogeneity classification on the following vehicle and the leading vehicle to generate a vehicle heterogeneity combination, and a prediction neural network model for the following vehicle is established based on the vehicle heterogeneity combination.

[0079] Specifically, since this application studies car-following speed, the primary approach is to use data-driven historical data from the preceding vehicle to predict the future following speed of the following vehicle. Therefore, it is necessary to ensure that the following vehicle engages in following behavior and that the following behavior lasts for a sufficient period of time. In case 2, the preceding vehicle must maintain a sufficiently long following relationship with the preceding vehicle. Based on this, all data meeting the requirements is cleaned and filtered from the dataset. As mentioned above, vehicle heterogeneity during car-following can affect the predicted following speed. Therefore, the types of the following vehicle, the preceding vehicle, and the preceding vehicle must be considered separately. In the HighD dataset, each vehicle is classified as either "Car" or "Truck," corresponding to the heterogeneous categories of passenger cars and commercial vehicles. In computer science, neural networks are algorithmic models designed based on the structure and function of the human brain, simulating the predictive judgments made by drivers. Therefore, the driver's car-following psychology must be considered to analyze the impact of vehicle heterogeneity. When the following vehicle is a commercial vehicle, due to its large mass and high safety risk, the following vehicle will focus on maintaining a safe distance and speed to avoid danger. Furthermore, commercial vehicles are large, and the driving status of the preceding vehicle is often obscured when following a commercial vehicle. Based on these two factors, the driving status of the preceding vehicle is less influential when following a commercial vehicle, so the influence of the preceding vehicle does not need to be considered. When the following vehicle is a passenger car, the safety risk is relatively low, and the driving status of the preceding vehicle can be more clearly observed. Changes in the preceding vehicle's driving status, such as acceleration or deceleration, can significantly affect the following vehicle's speed control, leading to changes in future driving status. Therefore, when the following vehicle is a passenger car, the influence of both the preceding vehicle and the preceding vehicle should be considered. Given that neural network methods require a sufficient amount of data for training, considering the heterogeneity of each vehicle will result in too small a amount of cleaned data, making it difficult to train an accurate prediction network. Therefore, the heterogeneity of only two vehicles is considered. Therefore, when the leading vehicle is a commercial vehicle, the heterogeneity between the following vehicle and the preceding vehicle is considered. When the leading vehicle is a passenger car, the heterogeneity between the preceding vehicle and the preceding vehicle is considered. After classifying the extracted data based on heterogeneity, Case 1 corresponds to four categories: passenger car-commercial vehicle, commercial vehicle-commercial vehicle, passenger car-passenger car, and commercial vehicle-passenger car. Case 2 corresponds to four categories: passenger car-commercial vehicle, commercial vehicle-commercial vehicle, following vehicle-passenger car-passenger car, and following vehicle-passenger car-commercial vehicle.

[0080] This application uses a prediction model to classify the heterogeneity of the following vehicle and the following vehicle to generate a vehicle heterogeneity combination, and establishes a prediction neural network model for the following vehicle based on the vehicle heterogeneity combination. That is, the vehicle and the leading vehicle are classified according to their heterogeneity, and multiple corresponding prediction neural network models are established, which comprehensively considers the differences in drivers' driving styles, vehicle heterogeneity, and the influence of the leading vehicle on the following behavior and their coupling effect.

[0081] Optionally, in one embodiment of the present application, a prediction model is used to perform heterogeneous classification on the following vehicle and the following vehicle to generate a vehicle heterogeneous combination, including: performing heterogeneous classification on the following vehicle and the following vehicle based on the speed, position information of the following vehicle and the speed, position, and driver characteristic information of the following vehicle to generate a vehicle heterogeneous combination.

[0082] During actual implementation, the embodiments of the present application can perform heterogeneous classification of the following vehicle and the preceding vehicle based on the speed, position information of the preceding vehicle and the speed, position, and driver characteristic information of the following vehicle to generate a vehicle heterogeneity combination. Considering the different driving states of drivers when following commercial vehicles and passenger cars, the following vehicle and the preceding vehicle are classified according to their heterogeneity and neural networks are constructed and trained separately. Historical traffic data and driver characteristic data are input separately, and the historical traffic time series are processed using a Bi-LSTM network. The established driver style recognition network is used to output the probability of each style. The outputs of the two are then spliced ​​together and the FCN network is used to output the future predicted following vehicle speed.

[0083] In step S104, the prediction neural network model is trained using the data in the data set until the prediction neural network model meets the training stop condition, and a target prediction neural network model for predicting the speed of the following vehicle is generated.

[0084] As you can understand, there are two cases (Case 1 and Case 2) based on the number of vehicles ahead of the following vehicle. Each case is further divided into four cases based on vehicle heterogeneity, for a total of eight cases. Data matching each case was extracted from the HighD dataset to create a separate dataset for testing. Each dataset was divided into training and test sets in an 8:2 ratio, and training and validation were performed separately.

[0085] During the actual execution process, the embodiment of the present application can use the data in the data set to train the prediction neural network model until the prediction neural network model meets the training stop condition, stop training, and generate a target prediction neural network model for predicting the speed of the following vehicle.

[0086] The embodiment of the present application can use the data in the HighD dataset for multiple trainings to improve the accuracy of model predictions, more accurately predict the speed of the following vehicle, and improve the accuracy of road traffic status prediction.

[0087] Figure 8 A following speed prediction method considering driver style and vehicle heterogeneity is proposed for the model application stage, which includes the following steps:

[0088] In step S801 , a vehicle heterogeneity combination is obtained.

[0089] The speed and position information of the leading vehicle and the speed, position and driver characteristic information of the trailing vehicle are used to classify the trailing vehicle and the leading vehicle according to their heterogeneity to obtain a vehicle heterogeneity combination.

[0090] In step S802, the vehicle heterogeneity combination is input into a pre-built prediction neural network model to predict the speed of the following vehicle, wherein the prediction neural network model is constructed by the vehicle heterogeneity combination.

[0091] The embodiments of this application can build multiple neural networks to predict the future speed of the following vehicle. Compared with rule-based speed prediction, this method significantly improves prediction accuracy and reduces prediction error. Furthermore, since the prediction is based on the speed for a period of time in the future, compared to outputting only the speed for the next step, this application has advantages in both short-term and long-term prediction, providing assistance in the fields of vehicle decision-making, control, and traffic management.

[0092] According to the following speed prediction method that takes into account the driver style and vehicle heterogeneity proposed in the embodiment of the present application, the K-means clustering method and the FCN neural network are used to establish a corresponding model between the driver style feature data in the data set and the probability of setting the driver style type. The BiLSTM neural network is used and integrated with the trained driver style prediction network to build a prediction model of the historical traffic data, driver style feature data and future speed data of the following vehicle in the data set. The vehicle is classified according to the heterogeneity of the vehicle itself and the preceding vehicle, and multiple corresponding prediction neural network models are established. The data in the HighD data set is used for multiple training to improve the accuracy of the model prediction. The present invention can more accurately predict the speed of the following vehicle and improve the accuracy of the road traffic status prediction. Thus, it solves the problem that the related art assumes that all driver behaviors and vehicle types are consistent. This assumption is too idealistic and ignores the differences in drivers' driving styles and the heterogeneity of vehicles.

[0093] Next, a device for predicting a following speed taking into account driver style and vehicle heterogeneity according to an embodiment of the present application will be described with reference to the accompanying drawings.

[0094] Figure 9 3 is a schematic structural diagram of a device for predicting a following speed taking into account driver style and vehicle heterogeneity according to an embodiment of the present application.

[0095] like Figure 9 As shown, the following speed prediction device 10 considering driver style and vehicle heterogeneity includes: an establishment module 100 , a construction module 200 , a generation module 300 and a training module 400 .

[0096] Specifically, the establishment module 100 is used to establish a correspondence model between the driver style feature data in the data set and the probability of the driver style type based on a fully connected neural network.

[0097] Building module 200 is used to build a prediction model based on the corresponding model, using a bidirectional long short-term memory network and a trained driver style prediction network, based on the historical traffic data, driver style feature data and target speed data of the following vehicle in the dataset.

[0098] The generation module 300 is used to perform heterogeneity classification on the following vehicle and the leading vehicle using the prediction model to generate a vehicle heterogeneity combination, and establish a prediction neural network model for the following vehicle based on the vehicle heterogeneity combination.

[0099] The training module 400 is used to train the prediction neural network model using the data in the data set until the prediction neural network model meets the training stop condition, thereby generating a target prediction neural network model for predicting the speed of the following vehicle.

[0100] Optionally, in one embodiment of the present application, the establishment module 100 includes: an extraction unit, a generation unit, a clustering unit and an establishment unit.

[0101] The extraction unit is used to extract driver style feature data including headway, collision distance, own vehicle speed, own vehicle acceleration and following distance from the data set.

[0102] The generating unit is used to generate the features of the sample according to the average value of the inverse collision time distance, the average value and standard deviation of the headway time distance, the average value and standard deviation of the acceleration, the standard deviation of the vehicle speed and the mean value of the following vehicle distance.

[0103] The clustering unit is used to perform dimensionality reduction processing on the features of the samples to generate reduced-dimensionality data, and cluster the features of the samples based on the reduced-dimensionality data to generate driver style types.

[0104] A unit is established to build a fully connected neural network, extract feature data corresponding to the vehicle identification number at preset time intervals, use the feature data as the input of the fully connected neural network, use the probability of the corresponding type label after clustering as the output of the fully connected neural network, establish a driver style recognition neural network with a fully connected structure, and generate a corresponding model based on the driver style recognition neural network.

[0105] Optionally, in one embodiment of the present application, the building module 200 includes: a processing unit and a building unit.

[0106] The processing unit is used to process the historical traffic time series sequence based on the historical traffic data and the driver style feature data using a bidirectional long short-term memory network to generate a processed traffic time series sequence.

[0107] A construction unit is used to output the probability of the driver's style using a driver style recognition neural network based on the processed traffic time series, and based on the style probability, a fully connected neural network is used to output the target speed data of the following vehicle, so as to build a prediction model based on the target speed data of the following vehicle.

[0108] Optionally, in one embodiment of the present application, the calculation formula for the headway is:

[0109]

[0110] Where d is the relative distance between the following vehicle and the preceding vehicle, v e is the longitudinal speed of the vehicle;

[0111] The calculation formula for collision time is:

[0112]

[0113] Where d is the relative distance between the following vehicle and the preceding vehicle, v e is the longitudinal velocity of the vehicle, v p is the longitudinal speed of the preceding vehicle.

[0114] Optionally, in one embodiment of the present application, the following speed prediction device 10 considering driver style and vehicle heterogeneity further includes: a judgment module, a first data generation module and a second data generation module.

[0115] Among them, the judgment module is used to judge the type of the following vehicle before building a prediction model based on historical traffic data, driver style feature data and target speed data of the following vehicle in the data set.

[0116] The first data generating module is configured to generate historical traffic data according to the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle when the preceding vehicle is a commercial vehicle.

[0117] The second data generation module is used to generate historical traffic data based on the speed of the preceding vehicle, the following distance between the preceding vehicle and the preceding vehicle, the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle if the type of the preceding vehicle is a passenger car.

[0118] Optionally, in one embodiment of the present application, the generation module 300 includes: a combination generation unit.

[0119] The combination generating unit is used to perform heterogeneity classification on the following vehicle and the leading vehicle based on the speed and position information of the leading vehicle and the speed, position, and driver characteristic information of the following vehicle to generate a vehicle heterogeneity combination.

[0120] Figure 10 A car-following speed prediction device 20 considering driver style and vehicle heterogeneity is provided, which is applied in a model application stage and includes: an acquisition module 500 and a prediction module 600.

[0121] Specifically, the acquisition module 500 is used to acquire a vehicle heterogeneity combination.

[0122] The prediction module 600 is used to input the vehicle heterogeneity combination into a pre-built prediction neural network model to predict the speed of the following vehicle, wherein the prediction neural network model is constructed by the vehicle heterogeneity combination.

[0123] It should be noted that the aforementioned explanation of the embodiment of the following speed prediction method considering driver style and vehicle heterogeneity is also applicable to the following speed prediction device considering driver style and vehicle heterogeneity of this embodiment, and will not be repeated here.

[0124] According to the embodiment of the present application, a device for predicting the following speed that takes into account the driver style and vehicle heterogeneity is proposed. The K-means clustering method and the FCN neural network are used to establish a corresponding model between the driver style feature data in the data set and the probability of setting the driver style type. The BiLSTM neural network is used and integrated with the trained driver style prediction network to build a prediction model of the historical traffic data, driver style feature data and future speed data of the following vehicle in the data set. The vehicle is classified according to the heterogeneity of the vehicle itself and the preceding vehicle, and multiple corresponding prediction neural network models are established. The data in the HighD data set is used for multiple training to improve the accuracy of the model prediction. The present invention can more accurately predict the speed of the following vehicle and improve the accuracy of the road traffic status prediction. Thus, the problem that the related art assumes that all driver behaviors and vehicle types are consistent is solved. This assumption is too idealistic and ignores the differences in drivers' driving styles and the heterogeneity of vehicles.

[0125] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0126] A memory 1101 , a processor 1102 , and a computer program stored in the memory 1101 and executable on the processor 1102 .

[0127] When the processor 1102 executes the program, the following speed prediction method considering the driver style and vehicle heterogeneity provided in the above embodiment is implemented.

[0128] Furthermore, the electronic device further includes:

[0129] The communication interface 1103 is used for communication between the memory 1101 and the processor 1102 .

[0130] The memory 1101 is used to store computer programs that can be run on the processor 1102 .

[0131] The memory 1101 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0132] If the memory 1101, processor 1102, and communication interface 1103 are implemented independently, the communication interface 1103, memory 1101, and processor 1102 can be connected to each other via a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 11 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0133] Optionally, in a specific implementation, if the memory 1101, the processor 1102 and the communication interface 1103 are integrated on a chip, the memory 1101, the processor 1102 and the communication interface 1103 can communicate with each other through an internal interface.

[0134] The processor 1102 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0135] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0136] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0137] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0138] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or N wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program can be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing it in other suitable ways as necessary, and then storing it in a computer memory.

[0139] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0140] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0141] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0142] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A following speed prediction method considering driver style and vehicle heterogeneity, characterized in that: Applied to the model building phase, it includes the following steps: Based on a fully connected neural network, a correspondence model is established between the driver style feature data in the dataset and the probability of the driver style type; Based on the corresponding model, a prediction model is constructed using a bidirectional long short-term memory network and a trained driver style prediction network to construct the historical traffic data, the driver style feature data, and the target speed data of the following vehicle in the dataset; Using the prediction model to perform heterogeneity classification on the following vehicle and the leading vehicle to generate a vehicle heterogeneity combination, and establishing a prediction neural network model for the following vehicle based on the vehicle heterogeneity combination; The prediction neural network model is trained using the data in the data set until the prediction neural network model meets the training stop condition, thereby generating a target prediction neural network model for predicting the speed of the following vehicle.

2. The method according to claim 1, characterized in that Before establishing a correspondence model between the driver style feature data in the data set and the probability of the driver style type, it also includes: Determining the type of the preceding vehicle; If the type of the preceding vehicle is a commercial vehicle, generating the historical traffic data according to the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle; If the type of the preceding vehicle is a passenger car, the historical traffic data is generated according to the speed of the preceding vehicle, the following distance between the preceding vehicle and the preceding vehicle, the speed of the preceding vehicle, the following distance between the preceding vehicle and the following vehicle, and the speed of the following vehicle.

3. The method according to claim 1, characterized in that The method of establishing a correspondence model between the driver style feature data in the data set and the probability of the driver style type based on a fully connected neural network includes: Extracting the driver style feature data including headway, collision distance, own vehicle speed, own vehicle acceleration and following distance from the data set; The characteristics of the samples are generated based on the mean value of the inverse collision time distance, the mean value and standard deviation of the headway time distance, the mean value and standard deviation of the acceleration, the standard deviation of the vehicle speed, and the mean value of the following distance; performing dimensionality reduction processing on the features of the samples to generate dimensionality-reduced data, and clustering the features of the samples based on the dimensionality-reduced data to generate the driver style type; The fully connected neural network is constructed, and feature data corresponding to the vehicle identification number is extracted at preset time intervals. The feature data is used as the input of the fully connected neural network, and the probability of the corresponding type label after clustering is used as the output of the fully connected neural network. A driver style recognition neural network with a fully connected structure is established, and the corresponding model is generated based on the driver style recognition neural network.

4. The method according to claim 3, characterized in that The step of constructing a prediction model based on the historical traffic data, the driver's style characteristic data, and the target speed data of the following vehicle in the dataset includes: Based on the historical traffic data and the driver style feature data, using the bidirectional long short-term memory network to process the historical traffic time series sequence to generate a processed traffic time series sequence; Based on the processed traffic time series, the driver style recognition neural network is used to output the probability of the driver style. Based on the style probability, the fully connected neural network is used to output the target speed data of the following vehicle, so as to build the prediction model based on the target speed data of the following vehicle.

5. The method according to claim 3, characterized in that The calculation formula of the headway is: Wherein, d is the relative distance between the following vehicle and the following vehicle, v e is the longitudinal speed of the vehicle; The calculation formula of the collision time distance is: Wherein, d is the relative distance between the following vehicle and the following vehicle, v e is the longitudinal velocity of the vehicle, v p is the longitudinal speed of the preceding vehicle.

6. The method according to claim 1, characterized in that The method of using the prediction model to perform heterogeneity classification on the following vehicle and the leading vehicle to generate a vehicle heterogeneity combination includes: The heterogeneity of the following vehicle and the following vehicle is classified based on the speed and position information of the leading vehicle and the speed, position, and driver characteristic information of the following vehicle to generate the vehicle heterogeneity combination.

7. A method for predicting following speed considering driver style and vehicle heterogeneity, characterized in that: Applied to the model application phase, it includes the following steps: Obtain vehicle heterogeneity combinations; The vehicle heterogeneity combination is input into a pre-built prediction neural network model to predict the speed of the following vehicle, wherein the prediction neural network model is constructed based on the vehicle heterogeneity combination.

8. A following speed prediction device considering driver style and vehicle heterogeneity, characterized in that: Applied in the model building phase, including: Establishing a module for establishing a correspondence model between driver style feature data in a data set and the probability of driver style types based on a fully connected neural network; A building module for building a prediction model based on the corresponding model using a bidirectional long short-term memory network and a trained driver style prediction network to build a prediction model for the historical traffic data in the dataset, the driver style feature data, and the target speed data of the following vehicle; a generation module for performing heterogeneity classification on the following vehicle and the leading vehicle using the prediction model to generate a vehicle heterogeneity combination, and establishing a prediction neural network model for the following vehicle based on the vehicle heterogeneity combination; A training module is used to train the prediction neural network model using the data in the data set until the prediction neural network model meets the training stop condition, thereby generating a target prediction neural network model for predicting the speed of the following vehicle.

9. A following speed prediction device considering driver style and vehicle heterogeneity, characterized in that: Applied in the model application phase, including: An acquisition module, used to obtain vehicle heterogeneity combinations; A prediction module is used to input the vehicle heterogeneity combination into a pre-built prediction neural network model to predict the speed of the following vehicle, wherein the prediction neural network model is constructed by the vehicle heterogeneity combination.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the following speed prediction method considering driver style and vehicle heterogeneity according to any one of claims 1 to 6 or the following speed prediction method considering driver style and vehicle heterogeneity according to claim 7.

Images