A vehicle trajectory prediction method and device adapted to different driving styles

Abstract

The present invention provides a vehicle trajectory prediction method and device adaptable to different driving styles, relating to the field of autonomous driving technology. The vehicle trajectory prediction method adaptable to different driving styles includes: obtaining an original sample data set of vehicle trajectories; performing filtering processing on the sample data set to obtain a filtered sample data set; performing dimensionality reduction processing by using the principal component analysis method to obtain dimensionally reduced data; performing soft clustering processing by using the fuzzy C-means clustering algorithm to obtain hard labels; constructing an initial global-local attention mechanism residual bidirectional long short-term memory network; training the initial network to obtain a trained global-local attention mechanism residual bidirectional long short-term memory network; obtaining the state information of a target vehicle within an observation time domain; and inputting the state information into the trained network to obtain the future trajectory of the target vehicle within a prediction time domain. By using the present invention, the accuracy of vehicle trajectory prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and particularly to a vehicle trajectory prediction method and device adapted to different driving styles. Background Art

[0002] The different driving habits of traffic participants and their complex interactions with the environment pose a huge challenge to the accurate prediction of future trajectories. It is worth noting that human drivers usually observe surrounding traffic participants, predict their future states, and then perform appropriate driving operations, including accelerating, braking, or changing lanes. Therefore, how to establish an efficient and accurate trajectory prediction system in a highly complex driving environment has become one of the main challenges faced by autonomous vehicles.

[0003] Currently, the methods for vehicle trajectory prediction include: physical model-based methods, classical machine learning-based methods, and deep learning-based methods; among them, the physical model-based methods can perform short-term trajectory prediction, but do not consider the influence of driving style on the vehicle driving trajectory; the classical machine learning-based methods use data-driven methods for trajectory prediction, but usually require pre-determining or providing relevant features, which limits their effectiveness in dealing with tasks in complex environments; the deep learning-based methods train and predict vehicle trajectories by using a large amount of data, but cannot perform vehicle trajectory prediction for different driving styles, and when dealing with long-time sequences, the calculation time of the deep learning-based methods is relatively long, resulting in poor prediction accuracy. Summary of the Invention

[0004] In order to solve the technical problems in the prior art that vehicle trajectory prediction cannot be performed for different driving styles, and the calculation time of the deep learning-based method is relatively long when dealing with long-time sequences, resulting in poor prediction accuracy, the embodiments of the present invention provide a vehicle trajectory prediction method and device adapted to different driving styles. The technical solutions are as follows:

[0005] On the one hand, a vehicle trajectory prediction method adapted to different driving styles is provided. This method is implemented by a vehicle trajectory prediction device adapted to different driving styles, and the method includes:

[0006] S1. Obtain the sample data set of the original vehicle trajectory;

[0007] S2. Perform filtering processing on the sample data set to obtain the filtered sample data set;

[0008] S3. According to the filtered sample data set, perform dimensionality reduction processing by using the principal component analysis method to obtain the dimensionality-reduced data;

[0009] S4. Use the fuzzy C - means clustering algorithm to perform soft clustering on the dimension - reduced data, obtain the membership degrees of the dimension - reduced data for each driving style, determine the hard labels of the dimension - reduced data according to the maximum membership degree, and add the hard labels to the sample data set after filtering processing;

[0010] S5. Construct an initial global - local attention mechanism residual bidirectional long - short - term memory network;

[0011] S6. According to the sample data set with hard labels added, train the initial global - local attention mechanism residual bidirectional long - short - term memory network to obtain a trained global - local attention mechanism residual bidirectional long - short - term memory network;

[0012] S7. Obtain the state information of the target vehicle within the observation time domain; input the state information of the target vehicle at the observation time into the trained global - local attention mechanism residual bidirectional long - short - term memory network to obtain the future trajectory of the target vehicle within the prediction time domain.

[0013] On the other hand, a vehicle trajectory prediction device adapted to different driving styles is provided. This device is applied to the vehicle trajectory prediction method adapted to different driving styles, and the device includes:

[0014] The first acquisition unit is used to acquire the sample data set of the original vehicle trajectory;

[0015] The pre - processing unit is used to perform filtering processing on the sample data set to obtain the sample data set after filtering processing;

[0016] The second acquisition unit is used to perform dimension - reduction processing on the sample data set after filtering processing by using the principal component analysis method to obtain the dimension - reduced data;

[0017] The third acquisition unit uses the fuzzy C - means clustering algorithm to perform soft clustering on the dimension - reduced data, obtain the membership degrees of the dimension - reduced data for each driving style, determine the hard labels of the dimension - reduced data according to the maximum membership degree, and add the hard labels to the sample data set after filtering processing;

[0018] The construction unit is used to construct an initial global - local attention mechanism residual bidirectional long - short - term memory network;

[0019] The training unit is used to train the initial global - local attention mechanism residual bidirectional long - short - term memory network according to the sample data set with hard labels added to obtain a trained global - local attention mechanism residual bidirectional long - short - term memory network;

[0020] A prediction unit, configured to obtain the state information of a target vehicle within an observation time domain; and input the state information of the target vehicle at the observation time into the trained global-local attention mechanism residual bidirectional long short-term memory network to obtain the future trajectory of the target vehicle within a prediction time domain.

[0021] On the other hand, a vehicle trajectory prediction device adapted to different driving styles is provided. The vehicle trajectory prediction device adapted to different driving styles includes: a processor; a memory storing computer-readable instructions thereon, and when the computer-readable instructions are executed by the processor, any one of the methods in the vehicle trajectory prediction method adapted to different driving styles as described above is implemented.

[0022] On the other hand, a computer-readable storage medium is provided, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement any one of the methods in the vehicle trajectory prediction method adapted to different driving styles as described above.

[0023] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include:

[0024] Obtain a sample data set of the original vehicle trajectory; perform filtering processing on the sample data set to obtain a filtered sample data set; according to the filtered sample data set, perform dimensionality reduction processing using the principal component analysis method to obtain the data after dimensionality reduction; use the fuzzy C-means clustering algorithm to perform soft clustering processing on the data after dimensionality reduction to obtain the membership degrees of the data after dimensionality reduction for each driving style, determine the hard labels of the data after dimensionality reduction according to the maximum membership degree, and add the hard labels to the filtered sample data set; construct an initial residual bidirectional long short-term memory network based on the global-local attention mechanism; according to the sample data set with added hard labels, train the initial residual bidirectional long short-term memory network based on the global-local attention mechanism to obtain a trained residual bidirectional long short-term memory network based on the global-local attention mechanism; obtain the state information of the target vehicle within the observation time domain; input the state information of the target vehicle at the observation time into the trained residual bidirectional long short-term memory network based on the global-local attention mechanism to obtain the future trajectory of the target vehicle within the prediction time domain. In the embodiment of the present invention, the fuzzy C-means algorithm is used to perform soft clustering on different driver styles, and the clustering results are converted into clear hard labels; the vehicle trajectory prediction network proposed by the present invention comprehensively considers different driving styles and environmental uncertainties, improving the accuracy and adaptability of the vehicle trajectory prediction network; in order to effectively extract long-term sequence features, the embodiment of the present invention models the vehicle as a waveform to better understand long-term sequence features; in addition, the embodiment of the present invention proposes a new attention mechanism to aggregate global features and local features to ensure the effective and proper utilization of features. The trajectory prediction network model proposed by the present invention combines the long short-term memory network LSTM encoder-decoder structure with the residual structure, simplifies the training process, and improves the stability and accuracy of the prediction. Description of the Drawings

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0026] Figure 1 It is a process diagram of the implementation of vehicle trajectories with different driving styles provided by the embodiment of the present invention;

[0027] Figure 2 It is a flowchart of a vehicle trajectory prediction method that adapts to different driving styles provided by the embodiment of the present invention;

[0028] Figure 3 It is a schematic diagram of a vehicle trajectory annotation algorithm provided by the embodiment of the present invention;

[0029] Figure 4 is a comparison diagram of lane change trajectories of drivers with different styles provided by an embodiment of the present invention;

[0030] Figure 5 is a schematic diagram of a local attention module provided by an embodiment of the present invention;

[0031] Figure 6 is a schematic diagram of a global attention module provided by an embodiment of the present invention;

[0032] Figure 7 It is a schematic diagram of internal variables of an LSTM unit provided by an embodiment of the present invention;

[0033] Figure 8 It is a schematic diagram of a residual bidirectional long short-term memory network structure based on a global-local attention mechanism provided by an embodiment of the present invention;

[0034] Figure 9 is a schematic diagram of a trajectory prediction scenario provided by an embodiment of the present invention;

[0035] Figure 10 is a left lane change prediction result diagram provided by an embodiment of the present invention;

[0036] Figure 11 is a right lane change prediction result diagram provided by an embodiment of the present invention;

[0037] Figure 12 is a lane keeping prediction result graph provided by an embodiment of the present invention;

[0038] Figure 13 It is a block diagram of a vehicle trajectory prediction device adapted to different driving styles provided by an embodiment of the present invention;

[0039] Figure 14 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0041] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0042] In the embodiments of the present invention, the terms "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, their intended meanings are the same. The terms "of", "corresponding", and "corresponding to" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, their intended meanings are the same.

[0043] In the embodiments of the present invention, sometimes subscripts such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, their intended meanings are the same.

[0044] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0045] The embodiments of the present invention provide a vehicle trajectory prediction method adapted to different driving styles. This method can be implemented by a vehicle trajectory prediction device adapted to different driving styles, and the vehicle trajectory prediction device adapted to different driving styles can be a terminal or a server. As Figure 1 shown in the implementation process diagram of vehicle trajectories with different driving styles, Figure 1 it shows the overall framework of vehicle trajectory prediction, including: Wavelet Denoising (WD); Sliding Window Method (SWM); Principal Component Analysis (PCA); Fuzzy c-Means Clustering (FCM); Residual Structure (Res). As Figure 2 shown in the flowchart of the vehicle trajectory prediction method adapted to different driving styles, the processing flow of this method can include the following steps:

[0046] S1. Obtain the sample data set of the original vehicle trajectory.

[0047] Among them, the sample data set of the original vehicle trajectory can adopt the public traffic safety data set (Next Generation Simulation, NGSIM) launched by the Federal Highway Administration of the United States, which is widely used in the field of transportation and is used to study driving behaviors such as vehicle following and lane changing, vehicle motion trajectory prediction, and decision-making planning. The data set includes areas where vehicles and roads are coordinated, including: intersections on structured roads and highway ramps, etc.; the data set has been exported in.csv format for subsequent use.

[0048] Among them, the driving style is the driving method selected or habitual by the driver, including styles such as slow and steady driving, fast driving, and smooth driving.

[0049] S2. Perform filtering processing on the sample data set to obtain the filtered sample data set.

[0050] Optionally, the specific implementation process of S2 may include S21 - S22:

[0051] S21. Denoise the sample data set using the five - iteration wavelet transform denoising algorithm to obtain the denoised sample data set;

[0052] S22. According to the denoised sample data set, calculate the heading angle as a metric for trajectory annotation, and perform balancing processing on the denoised sample data set to obtain the balanced sample data set.

[0053] Among them, using the trajectory label to balance the denoised data set and performing driving style label annotation on the denoised data set can enhance the adaptability of the network to diverse driving behaviors.

[0054] In a feasible implementation manner, the embodiment of the present invention considers that the sample data set is concentrated on the lane - keeping trajectory and there are few left - and - right lane - change trajectories. Balance the sample data set through trajectory label annotation, and use the heading angle as a metric for trajectory annotation; among them, the heading angle can be expressed by the following formula (1):

[0055] (1)

[0056] Among them, represents the lateral position of the trajectory point; represents the longitudinal position of the trajectory point; represents the current time step; represents the next time step; represents the heading angle.

[0057] In a feasible implementation manner, the present invention regards the lane - change process as a continuous action of crossing the lane boundary, specifically including:

[0058] Filter the trajectory features of the target vehicle according to the vehicle ID information; as Figure 3 shown in the schematic diagram of the vehicle trajectory annotation algorithm, where the lane - change is confirmed by the lane ID information, and the point is defined as the position where the trajectory crosses the lane boundary; calculate the heading angle at the intersection point. If the heading angles of three consecutive points are between 0 degrees and 90 degrees, the first point that meets the condition is marked as the starting point of the right lane - change , and the last point that meets the condition is marked as the ending point of the right lane - change , at the starting point and the end point All the trajectory points between them are marked as part of a right lane change; Similarly, the left lane change is also marked: If the heading angle of three consecutive points is between -90 degrees and 0 degrees, the first point that meets the condition is marked as the starting point of the left lane change and the last point that meets the condition is marked as the end point of the left lane change , at the starting point and the end point All the trajectory points between them are marked as part of the left lane change. Among them, the points not recognized as lane changes are marked as lane keeping. Optionally, Table 1 shows the steps of the trajectory annotation algorithm for the NGSIM dataset.

[0059] Table 1

[0060]

[0061] Among them, the final labeled dataset is presented in the first row of Table 2. Obviously, the number of samples for right lane changes is relatively small. To ensure data balance, random sampling is performed on the lane keeping and left lane change data so that their sample sizes are equal to the right lane change data. The resulting balanced dataset is presented in the second row of Table 2.

[0062] Among them, Table 2 is the balanced data obtained after the trajectory label algorithm processes the sample dataset.

[0063] Table 2

[0064]

[0065] S3. According to the filtered sample dataset, use the principal component analysis method for dimensionality reduction processing to obtain the data after dimensionality reduction.

[0066] In a feasible implementation manner, the embodiment of the present invention selects driving behavior parameters as the characteristic parameters of the sample data, and performs dimensionality reduction processing on the filtered sample dataset by using the principal component analysis method; among them, the driving behavior parameters include: average speed, speed standard deviation, average absolute acceleration, and average headway distance, etc.

[0067] Among them, using principal component analysis to reduce the dimensionality of the dataset can enable the vehicle trajectory prediction network to better understand the structure and pattern of the data.

[0068] S4. Use the fuzzy C-means clustering algorithm to perform soft clustering processing on the data after dimensionality reduction, obtain the membership degrees of the data after dimensionality reduction for each driving style, determine the hard labels of the data after dimensionality reduction according to the maximum membership degree, and add the hard labels to the filtered sample dataset.

[0069] Among them, the clustering results are usually deterministic, such as distinguishing between men and women. Driving style represents a fuzzy concept, and the boundaries between different driving behaviors are not clear. Traditional hard clustering algorithms, such as k-means and Gaussian mixture models, face challenges in dealing with the uncertainty of driving styles. In contrast, the fuzzy C-means clustering algorithm effectively solves these problems by utilizing membership degrees; among them, the membership degree represents the degree of belonging of each data point to each cluster center. This method can accurately capture the ambiguity of driving styles and can well adapt to the complex structure of trajectory data.

[0070] Optionally, the specific implementation process of S4 may include S41 - S43:

[0071] S41. Calculate the fuzzy cluster centers for the dimensionality-reduced data using the fuzzy C-means clustering algorithm;

[0072] In a feasible implementation, in the fuzzy C-means clustering algorithm, the fuzzy cluster centers can be determined by minimizing the objective function, where the objective function can be represented by the following formula (2):

[0073] (2)

[0074] Among them, represents the fuzzy coefficient that controls the degree of fuzziness in the clustering process and affects the way of assigning membership values to each driving style in the trajectory sample data ; represents the number of vehicle trajectory samples; represents the number of driving style categories, that is, the number of cluster centers; represents the th cluster center; represents the sample belonging to the cluster center with a probability; represents the dimension of the sample; represents the sample at the th dimension value; represents the cluster center at the th dimension value; represents the Euclidean norm used to measure the distance between the driving sample and the cluster center.

[0075] Among them, the membership degree can be represented by the following formula (3):

[0076] (3)

[0077] Among them, the cluster center can be represented by the following formula (4):

[0078] (4)

[0079] S42. Iteratively calculate the membership degree distribution of the data after dimensionality reduction in different driving styles according to the fuzzy clustering center and the preset iteration conditions;

[0080] In a feasible implementation, iteratively calculate to obtain the membership degree distribution of each sample data in different driving styles. When the iteration termination condition is reached, stop the iteration to obtain the clustering result. Among them, the preset iteration termination condition can be expressed by the following formula (5):

[0081] (5)

[0082] Among them, represents the number of iterations, is a small constant representing the error threshold; when the maximum change in consecutive iterations does not exceed , the iteration stops.

[0083] S43. According to the membership degree distribution, calculate the membership degree of each data point in all clusters, and assign the data points to the corresponding clustering centers according to the maximum membership degree to obtain the hard labels of the sample data, and add the hard labels to the sample data set after filtering processing.

[0084] Among them, obtain the maximum membership degree, average the maximum membership degrees of all data points to obtain the fuzzy partition coefficient. Among them, the maximum membership degree can be expressed by the following formula (6):

[0085] (6)

[0086] Among them, represents the total number of data points; the closer the FPC value is to 1, the better the clustering effect.

[0087] Among them, the process of obtaining the hard labels of the sample data can be expressed by the following (7):

[0088] (7)

[0089] Among them, represents the hard label of the sample .

[0090] Among them, the fuzzy C-means clustering algorithm performs soft clustering processing on the sample data set, enhancing the accuracy and applicability of the vehicle trajectory prediction network.

[0091] The fuzzy C - means clustering algorithm is used to evaluate the influence of different numbers of clusters on the clustering performance; Table 3 shows the influence results of different numbers on the clustering effect; It can be seen from the results shown in Table 3 that the clustering reaches the best performance when there are three categories. Among them, in order to assign accurate driving style labels, each data point is assigned to the cluster center corresponding to its maximum membership degree. The fuzzy clustering results are transformed into accurate hard labels by using the concepts of soft clustering and hard labels, and the accurate hard labels are recorded in the dataset.

[0092] Table 3

[0093]

[0094] Among them, as Figure 4 shows the comparison of lane - changing trajectories from three different driving style categories. Obviously, the trajectory curve of Cluster 1 is relatively smooth and the lane - changing angle is small, which is classified as a conservative driving style. Cluster 2 shows a moderate lane - changing angle and a medium trajectory length, showing a neutral driving style. Cluster 3 shows fast lane - changing actions and a short trajectory length, which is defined as an aggressive driving style.

[0095] Among them, the data is divided into a training set, a test set and a validation set according to the ratio of 6:2:2. In order to improve the prediction ability and training efficiency of the model, data difference and feature scaling regularization methods are adopted to reduce the noise and fluctuations in the sequence.

[0096] S5. Construct an initial global - local attention mechanism - based residual bidirectional long - short - term memory network.

[0097] Optionally, the initial global - local attention mechanism - based residual bidirectional long - short - term memory network in S5 includes: a global attention module, a local attention module, and a residual bidirectional long - short - term memory network;

[0098] Among them, the global attention module adopts an improved multi - head self - attention mechanism, a residual gating mechanism, and a global context attention mechanism to enhance the extracted global features;

[0099] The local attention module is used to enhance the extracted local features;

[0100] The residual bidirectional long - short - term memory network is used to determine the long - term dependencies in the input sequence data.

[0101] S6. According to the sample dataset after adding hard labels, train the initial global - local attention mechanism - based residual bidirectional long - short - term memory network to obtain a trained global - local attention mechanism - based residual bidirectional long - short - term memory network.

[0102] Optionally, the specific implementation process of S6 may include S61 - S65:

[0103] S61. Input the sample data set with hard tags added into the initial global-local attention mechanism residual bidirectional long short-term memory network. According to the principles of quantum mechanics, model the vehicle as a waveform. By analyzing the superposition effect of vehicle waves, determine the amplitude and phase of the vehicle wave, and obtain the feature data with vehicle waves and phases.

[0104] In a feasible implementation, when the waves of two or more vehicles meet at the same point, the superposition principle of waves is followed; at this point, the superposition effect of vehicle waves can be represented by the resultant vector sum of their respective displacements, and the superposition effect can be expressed by the following formula (8):

[0105] (8)

[0106] where represents the imaginary unit; represents the absolute value operation; represents element-wise multiplication; represents the amplitude, represents the phase, represents a selected positive integer.

[0107] where the amplitude of the vehicle wave is affected by the speed difference, and the phase of the wave represents the vehicle position. When two vehicle waves exhibit the same displacement in the same direction at a certain point, the phase difference at which constructive interference occurs can be expressed by the following formula (9):

[0108] (9)

[0109] where represents the corresponding selected integer multiple; represents the phase difference between two vehicle waves.

[0110] where, when two vehicle waves have opposite displacements at a certain point, the phase difference between the two vehicle waves can determine the occurrence of destructive interference, which can be expressed by the following formula (10):

[0111] (10)

[0112] where, when vehicles with different speeds interfere, the superposition effect of waves can show how waves with different amplitudes and phases interfere with each other. By analyzing the superposition effect of vehicles, the interaction between different vehicles can be inferred, and the dynamic behavior in the traffic scene can be understood.

[0113] where the amplitude and phase of the vehicle wave can be expressed by the following formula (11):

[0114] (11)

[0115] Among them, represents a standard fully connected layer; represents the first learning parameter; represents the second learning parameter; represents the input feature; represents the amplitude of the vehicle wave; represents the phase of the vehicle wave; represents a selected positive integer; among them, the first learning parameter and the second learning parameter can be adjusted according to the input feature to obtain the optimal weights for the amplitude and phase embeddings of the vehicle wave; among them, the representation of the phase and amplitude of the vehicle wave in the negative domain can be represented by the following formula (12):

[0116] (12)

[0117] In a feasible implementation, in the interaction of all vehicles in the driving scenario, n vehicle waves can be represented as , and the aggregation of vehicle waves can be represented by the following formula (13):

[0118] (13)

[0119] Among them, represents the complex domain representation after aggregating vehicle waves;

[0120] In a feasible implementation, the real number after aggregating vehicle waves can be obtained through weighted summation; among them, the real number after aggregating vehicle waves can be represented by the following formula (14):

[0121] (14)

[0122] Among them, represents the real number after aggregating vehicle waves; represents the first learning weight; represents the first learning weight.

[0123] S62. Input the feature data with vehicle waves and phases into the local attention module, and perform initial dimension conversion through the convolutional layer to obtain the converted feature map; use the adaptive convolutional layer and the weighting mechanism to process the converted feature map to obtain local features;

[0124] Among them, as Figure 5 shown in the schematic diagram of the local attention module framework structure; among them, input the input data into the local attention module, and through the processing of two layers of convolution, batch normalization function, and ReLU function, obtain the first processing result; use The convolution calculates the first processing result to obtain the second processing result; it is further processed through the softmax function and the Dropout regularization method to obtain local features.

[0125] S63. Input the feature data with vehicle waves and phases into the global attention module. Through the improved multi-head self-attention and global context attention, determine the feature relationships between time steps of the feature data with vehicle waves and phases to obtain global features.

[0126] Among them, as Figure 6 shown in the schematic diagram of the global attention module framework structure; among them, input the input data into the global attention module, process it through the multi-head attention mechanism to obtain the processing result; the residual gating mechanism further processes the processing result, extracts features through the global context attention to obtain global features. The global attention module includes the improved multi-head self-attention and global context attention. The improved multi-head self-attention adds a gating mechanism and a residual connection. By activating the GLU, the network can selectively enhance the input features or reduce the adverse effects of the input features; the residual connection helps to achieve efficient information transmission and integration between different attention heads; by processing multiple attention heads simultaneously, the network can effectively capture the feature relationships of each time step in the time series. Among them, the global context attention aggregates the global context information, so as to comprehensively understand the interdependent relationships between features in all time series.

[0127] Among them, the global attention module integrates the features of each local time window, and obtains the global structure information by considering the relationships between the features of the current time step and all other time steps.

[0128] In a feasible implementation manner, let the given first time step be p, and the feature of the first time step be ; let the given second time step be q, and the feature of the second time step be , map the above variables to a common feature space through the feature mapping function, which can be represented by the following formula (15):

[0129] (15)

[0130] Among them, represents the feature mapping function; represents the first time step and the second time step the relationship between features.

[0131] Among them, the relationship vector of the feature of the first time step p can be represented by the following formula (16):

[0132] (16)

[0133] Among them, represents the relationship vector between the first time step p and the feature .

[0134] In a feasible implementation, the features of the time step and the relationship vector are combined into a paired representation, and the global feature representation at the first time step p is obtained as ; the global feature representation at the second time step q can be obtained according to the above steps; among them, the specific implementation steps for obtaining the global feature representation at the second time step q are the same as those for obtaining the global feature representation at the first time step p, and will not be elaborated in the embodiments of the present invention here.

[0135] S64. Fuse the local feature and the global feature to obtain the fused feature;

[0136] In a feasible implementation, the fused feature can be represented by the following formula (17):

[0137] (17)

[0138] Among them, represents the fused feature; represents the local feature; represents the global feature.

[0139] Among them, by fusing the local feature and the global feature, the vehicle trajectory prediction network can comprehensively understand the overall traffic scenario and accurately determine the subtle changes between vehicles.

[0140] S65. Input the fused feature into the residual bidirectional long short-term memory network, and calculate the input fused feature by constructing a residual block through skip connections, and output the prediction result.

[0141] Among them, the bidirectional long short-term memory network is constructed on the basis of the long short-term memory network LSTM; among them, the long short-term memory network LSTM includes: a forget gate, an input gate, and an output gate; among them, as Figure 7 is a schematic diagram of the internal variables of the LSTM unit; among them, the relationship between the internal variables of the short-term memory network LSTM unit can be represented by the following formula (18):

[0142] (18)

[0143] Among them, represents the activation function; represents the hyperbolic tangent activation function; represents the time step The hidden state vector at includes the cumulative memory and information from the previous time step; Represents the current time step of the input vector, which is derived from ; Represents element-wise multiplication; Represents the first weight matrix, Represents the second weight matrix, Represents the third weight matrix, Represents the fourth weight matrix; Represents the bias term of the first weight matrix, Represents the bias term of the second weight matrix, Represents the bias term of the third weight matrix, Represents the bias term of the fourth weight matrix; The weight matrices and bias terms are used to calculate the forget gate, input gate, candidate cell state, and output gate.

[0144] Among them, the bidirectional long short-term memory network includes an additional reverse recurrent layer, which enables the network to access the forward and backward time step sequence data simultaneously. This bidirectional recursive structure can effectively integrate past and future information, enabling the model to more accurately determine the long-term dependencies in the sequence data.

[0145] In a feasible implementation, as shown in Figure 8 the schematic diagram of the residual bidirectional long short-term memory network structure; The residual bidirectional long short-term memory network adds a residual structure to the bidirectional long short-term memory network; Among them, the residual bidirectional long short-term memory network constructs a residual block using skip connections, enabling the network to utilize both the original input data and residual information from the previous layer simultaneously, enhancing the performance of the network; The calculation formula of the residual block can be represented by the following formula (19):

[0146] (19)

[0147] Among them, represents the input sequence, represents the output after network processing.

[0148] In a feasible implementation, when training the residual bidirectional long short-term memory network based on the global-local attention mechanism, it is carried out using the PyTorch deep learning framework on an NVIDIA GeForce RTX 4060 Ti GPU. The training process includes 500 iterations with a batch size of 1024; The loss function used is the mean squared error MSE, the optimizer is Adam, the learning rate is 0.001, and the weight decay is 0.0001; Among them, the encoder and decoder adopt the residual bidirectional long short-term memory network structure, each module includes 2 layers, the hidden dimension is 128, and the dropout rate is set to 0.2.

[0149] S7. Obtain the status information of the target vehicle within the observation time domain; input the status information of the target vehicle at the observation time into the trained global-local attention mechanism residual bidirectional long short-term memory network to obtain the future trajectory of the target vehicle within the prediction time domain.

[0150] Optionally, the status information of the target vehicle in the observation time domain in S7 includes: the type of the target vehicle, the running status of the target vehicle, and the spatio-temporal relationship between the target vehicle and surrounding obstacle vehicles.

[0151] Among them, as Figure 9 shown in the schematic diagram of the trajectory prediction scenario, it includes six obstacle vehicles with longitudinal displacements in the range of [-100, 100] meters; the goal of trajectory prediction is to predict the spatial coordinates of road participants including cars, pedestrians, and animals, etc., at different time periods, including short time intervals and long time intervals; where the short time interval is 1 - 3 seconds and the long time interval is 3 - 5 seconds.

[0152] Among them, in a complex traffic environment, the future trajectory of the target vehicle can be accurately predicted by using the surrounding environment information and the running status of the target vehicle itself.

[0153] Among them, the status information of the target vehicle in the observation time domain can be expressed by the following formula (20):

[0154] (20)

[0155]

[0156]

[0157] Among them, represents the input status of the target vehicle within the observation time domain ; represents the predicted length of the target vehicle; represents the predicted width of the target vehicle; represents the predicted lateral coordinate of the target;

[0158] represents the predicted longitudinal coordinate of the target; represents its driving style label. represents the predicted speed of the target; represents the predicted acceleration of the target; represents the headway between the predicted target vehicle and the th obstacle vehicle, represents the headway time between them, represents the observation time domain; Indicates the number of obstacle vehicles.

[0159] Among them, the future trajectory of the target vehicle within the prediction time domain can be represented by the following formula (21):

[0160] (21)

[0161]

[0162] Among them, represents the lateral coordinate of the predicted trajectory of the target vehicle from the current time to the prediction time domain within; represents the longitudinal coordinate of the predicted trajectory of the target vehicle from the current time to the prediction time domain within; represents the prediction time domain; represents the future trajectory of the target vehicle within the prediction time domain.

[0163] In a feasible implementation manner, the embodiment of the present invention uses the trajectory of the past 3 seconds to predict the trajectory of the next 5 seconds. Among them, the evaluation index is the root mean square error of the predicted trajectory within 1 to 5 seconds; the root mean square error evaluates the prediction accuracy of the entire trajectory sequence by calculating the average root of the mean square error between the predicted value and the actual value, and can reflect the overall accuracy of the vehicle trajectory prediction network; the calculation formula of the root mean square error can be represented by the following formula (22):

[0164] (22)

[0165] Among them, represents the number of samples, represents the coordinate of the th point of the true trajectory, represents the coordinate of the th point of the predicted trajectory.

[0166] Among them, to evaluate the performance of the proposed model, the present invention compares it with several advanced models proposed in recent years, including:

[0167] 1) GSTCN (Graph Space-Time Convolutional Network): Combines graph convolutional network (GCN) and CNN for time feature extraction.

[0168] 2) STDAN (Space-Time Dynamic Attention Network): Utilizes self-attention mechanism for multi-dimensional feature extraction to capture time and space interaction features.

[0169] 3) FHIFN (Historical Interaction Feature Fusion Network): Adopts multi-head attention mechanism to encode the interaction features between the target vehicle and surrounding vehicles.

[0170] 4) IHN (Intention-aware Hybrid Convolution and Attention Network): Utilize the intention convolution social pooling module and the hybrid attention mechanism module to consider the periodicity of trajectories affected by road geometry constraints.

[0171] 5) MTF-LSTM (Hybrid Teaching Force Long Short-Term Memory Network): A hybrid imitation learning decoding method based on the LSTM network for vehicle trajectory prediction.

[0172] 6) IMM (Interactive Multiple Model): Combines LSTM behavior recognition and Gaussian process motion modeling for accurate lane change trajectory prediction.

[0173] 7) GLEA-ResBiLSTM (Global-Local Attention Mechanism Residual Bidirectional Long Short-Term Memory Network): The vehicle trajectory prediction model proposed in the present invention.

[0174] Among them, as shown in the comparison result Table 4. It can be seen that the GLEA-ResBiLSTM model demonstrates excellent prediction ability. Especially in medium- and long-term prediction, the GSTCN, STDAN, and FHIFN methods show significant errors, and their RMSE values for 5-second prediction are 2.95, 3.57, and 3.63 respectively. In contrast, the IHN and MTF-LSTM methods show better performance. Although the IMM method has a slight advantage in the first 2-second prediction (RMSE less than 0.22), the proposed framework shows very stable prediction accuracy in the short term. As the prediction time extends, the prediction accuracy of the proposed GLEA-ResBiLSTM framework gradually emerges, and finally keeps the 5-second RMSE within 0.56. This result clearly demonstrates the robustness of the proposed GLEA-ResBiLSTM framework in dealing with complex scenario trajectory prediction.

[0175] Table 4

[0176]

[0177] To further intuitively demonstrate the prediction performance of the GLEA-ResBiLSTM model in different driving behaviors, the present invention selects trajectory samples of different driving behaviors from the balanced dataset shown in Table 2. Respectively shows the left lane change prediction result as shown in Figure 10 , the right lane change prediction result as shown in Figure 11 , and the prediction result of the lane keeping behavior as shown in Figure 12 . It can be observed again that the model proposed in the present invention shows excellent performance in predicting these behaviors, with high accuracy and stability.

[0178] An embodiment of the present invention obtains a sample data set of the original vehicle trajectory; performs filtering processing on the sample data set to obtain a filtered sample data set; performs dimensionality reduction processing on the filtered sample data set by using the principal component analysis method to obtain the data after dimensionality reduction; performs soft clustering processing on the data after dimensionality reduction by using the fuzzy C-means clustering algorithm to obtain the membership degree of the data after dimensionality reduction for each driving style, determines the hard label of the data after dimensionality reduction according to the maximum membership degree, and adds the hard label to the filtered sample data set; constructs an initial residual bidirectional long short-term memory network based on the global-local attention mechanism; trains the initial residual bidirectional long short-term memory network based on the global-local attention mechanism according to the sample data set added with the hard label to obtain a trained residual bidirectional long short-term memory network based on the global-local attention mechanism; obtains the state information of the target vehicle within the observation time domain; inputs the state information of the target vehicle at the observation time into the trained residual bidirectional long short-term memory network based on the global-local attention mechanism to obtain the future trajectory of the target vehicle within the prediction time domain.

[0179] An embodiment of the present invention uses the fuzzy C-means algorithm to perform soft clustering on different driver styles and converts the clustering results into clear hard labels; the vehicle trajectory prediction network proposed by the present invention comprehensively considers different driving styles and environmental uncertainties, improving the accuracy and adaptability of the vehicle trajectory prediction network; in order to effectively extract long-term sequence features, an embodiment of the present invention models the vehicle as a waveform to better understand long-term sequence features; in addition, an embodiment of the present invention proposes a new attention mechanism to aggregate global features and local features to ensure the effective and proper utilization of features. The trajectory prediction network model proposed by the present invention combines the long short-term memory network LSTM encoder-decoder structure with the residual structure, simplifies the training process, and improves the stability and accuracy of prediction.

[0180] Figure 13 is a block diagram of a vehicle trajectory prediction device adapted to different driving styles shown according to an exemplary embodiment. This device is used for the vehicle trajectory prediction method adapted to different driving styles. Refer to Figure 13 , this device includes a first acquisition unit 310, a preprocessing unit 320, a second acquisition unit 330, a third acquisition unit 340, a construction unit 350, a training unit 360, and a prediction unit 370. Among them:

[0181] The first acquisition unit 310 is used to acquire a sample data set of the original vehicle trajectory;

[0182] The preprocessing unit 320 is used to perform filtering processing on the sample data set to obtain a filtered sample data set;

[0183] A second acquisition unit 330, configured to perform dimensionality reduction processing on the filtered sample data set by using a principal component analysis method to obtain the data after dimensionality reduction;

[0184] A third acquisition unit 340, configured to perform soft clustering processing on the data after dimensionality reduction by using a fuzzy C-means clustering algorithm to obtain the membership degree of the data after dimensionality reduction for each driving style, determine the hard label of the data after dimensionality reduction according to the maximum membership degree, and add the hard label to the filtered sample data set;

[0185] A construction unit 350, configured to construct an initial global-local attention mechanism residual bidirectional long short-term memory network;

[0186] A training unit 360, configured to train the initial global-local attention mechanism residual bidirectional long short-term memory network according to the sample data set with hard labels added thereto to obtain a trained global-local attention mechanism residual bidirectional long short-term memory network;

[0187] A prediction unit 370, configured to obtain the state information of a target vehicle within an observation time domain; input the state information of the target vehicle at the observation time into the trained global-local attention mechanism residual bidirectional long short-term memory network to obtain the future trajectory of the target vehicle within a prediction time domain.

[0188] Figure 14 FIG. 16 is a schematic structural diagram of an electronic device 1400 provided by an embodiment of the present invention. The electronic device 1400 may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPUs) 1401 and one or more memories 1402. Among them, at least one instruction is stored in the memory 1402, and the at least one instruction is loaded and executed by the processor 1401 to implement the steps of the above vehicle trajectory prediction method adapted to different driving styles.

[0189] In an exemplary embodiment, a computer-readable storage medium is further provided, such as a memory including instructions, and the above instructions can be executed by a processor in a terminal to complete the above Chinese text spelling check method. For example, the computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0190] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium, and the above-mentioned storage medium can be a read-only memory, a magnetic disk or an optical disc, etc.

[0191] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A vehicle trajectory prediction method adaptable to different driving styles, characterized in that The method includes: S1. Obtain a sample data set of the original vehicle trajectory; S2. Perform filtering processing on the sample data set to obtain a filtered sample data set; Among them, the step of performing filtering processing on the sample data set in S2 to obtain a filtered sample data set includes: S21. Denoise the sample data set by using a five - iteration wavelet transform denoising algorithm to obtain a denoised sample data set; S22. According to the denoised sample data set, calculate the heading angle as a metric for trajectory annotation, and perform balancing processing on the denoised sample data set to obtain a balanced sample data set; Among them, the heading angle is represented by the following formula (1): Among them, x represents the horizontal position of the trajectory point; y represents the vertical position of the trajectory point; t represents the current time step; t+1 represents the next time step; θ head represents the heading angle; S3. According to the filtered sample data set, perform dimensionality reduction processing by using the principal component analysis method to obtain reduced - dimensional data; S4. Use the fuzzy C - means clustering algorithm to perform soft clustering on the reduced - dimensional data, obtain the membership degrees of the reduced - dimensional data for each driving style, determine the hard labels of the reduced - dimensional data according to the maximum membership degree, and add the hard labels to the filtered sample data set; Among them, the step of using the fuzzy C - means clustering algorithm in S4 to perform soft clustering on the reduced - dimensional data, obtain the membership degrees of the reduced - dimensional data for each driving style, determine the hard labels of the reduced - dimensional data according to the maximum membership degree, and add the hard labels to the filtered sample data set includes: S41. Use the fuzzy C - means clustering algorithm to calculate the fuzzy clustering centers of the reduced - dimensional data; Among them, in the fuzzy C - means clustering algorithm, the fuzzy clustering centers can be determined by minimizing the objective function, and the objective function is represented by the following formula (2): Among them, M represents the fuzzy coefficient that controls the degree of fuzziness in the clustering process, affecting the way of assigning membership values to each driving style in the trajectory sample data (1 ≤ M < ∞); N represents the number of vehicle trajectory samples; C represents the number of driving style categories, that is, the number of cluster centers; c j represents the j-th cluster center; u ij represents the probability that the sample x i belongs to the cluster center c j ; d represents the dimension of the sample; x ik represents the value of the sample x i at the k-th dimension; c jk represents the value of the cluster center c j at the k-th dimension; |·|2 represents the Euclidean norm used to measure the distance between the driving sample and the cluster center; where the membership degree u ij is represented by the following formula (3): Among them, the clustering center c j is represented by the following formula (4): S42. According to the fuzzy clustering centers and the preset iteration conditions, iteratively calculate the membership degree distributions of the reduced - dimensional data in different driving styles; S43. According to the membership degree distributions, calculate the membership degrees of each data point in all clusters, assign the data points to the corresponding clustering centers according to the maximum membership degree to obtain the hard labels of the sample data, and add the hard labels to the filtered sample data set; Among them, iteratively calculate to obtain the membership degree distributions of each sample data in different driving styles, and stop the iteration when the iteration termination condition is reached to obtain the clustering result; the preset iteration termination condition is represented by the following formula (5): where k' represents the number of iterations, and ε is a small constant representing the error threshold; when the maximum change in consecutive iterations does not exceed ε, the iteration stops; ij ​ Among them, the maximum membership degree is represented by the following formula (6): Among them, represents the total number of data points; the closer the FPC value is to 1, the better the clustering effect; Among them, the process of obtaining the hard labels of the sample data is represented by the following (7): where label(x i ) represents the hard label of sample x i ; S5. Construct an initial global - local attention mechanism residual bidirectional long - short - term memory network; S6. According to the sample data set with hard labels added, train the initial global - local attention mechanism residual bidirectional long - short - term memory network to obtain a trained global - local attention mechanism residual bidirectional long - short - term memory network; Among them, the specific implementation process of S6 includes S61 - S65: S61. Input the sample data set with hard tags added into the initial global-local attention mechanism residual bidirectional long short-term memory network. According to the principles of quantum mechanics, model the vehicle as a waveform. By analyzing the superposition effect of vehicle waves, determine the amplitude and phase of the vehicle wave, and obtain the feature data with vehicle waves and phases. Among them, when the waves of two or more vehicles meet at the same point, the superposition principle of waves is followed; at this point, the superposition effect of vehicle waves is represented by the resultant vector sum of their respective displacements, and the superposition effect is expressed by the following formula (8): where \(i\) m represents the imaginary unit; \(|\cdot|\) represents the absolute value operation; \(\odot\) represents element-wise multiplication; \(|z\) h | represents the amplitude, \(\theta\) h represents the phase, and \(h\) represents a selected positive integer; Among them, the amplitude of the vehicle wave is affected by the speed difference. The phase of the wave represents the vehicle position. When two vehicle waves show the same displacement in the same direction at a certain point, the phase difference at which constructive interference occurs is expressed by the following formula (9): Δθ h = θ0 + 2ηπ, η ∈ [0, ±2, ±4, …] (9) where η represents a corresponding selected integer multiple; Δθ h represents the phase difference between two vehicle waves; Among them, when two vehicle waves have opposite displacements at a certain point, the phase difference between the two vehicle waves determines the occurrence of destructive interference, which is expressed by the following formula (10): Δθ h = θ0 + ηπ, η ∈ [0, ±1, ±3, …] (10) Among them, the amplitude and phase of the vehicle wave are expressed by the following formula (11): Among them, FC represents a standard fully connected layer; ω z represents the first learning parameter; ω θ represents the second learning parameter; f g represents the input feature; θ h represents the phase of the vehicle wave; h represents a selected positive integer; among them, the first learning parameter and the second learning parameter are adjusted according to the input feature to obtain the optimal weights of the amplitude and phase embeddings of the vehicle wave; among them, the representations of the phase and amplitude of the vehicle wave in the complex domain are expressed by the following formula (12): Among the interactions of all vehicles in the driving scenario, n vehicle waves are represented as The aggregation of vehicle waves is represented by the following formula (13): Among them, represents the complex domain representation after aggregating vehicle waves; Among them, the real number after aggregating vehicle waves is obtained through weighted summation; the real number after aggregating vehicle waves is expressed by the following formula (14): Among them, o h represents a real number after aggregating vehicle waves; ω o represents the first learning weight; ω s represents the first learning weight; S62. Input the feature data with vehicle waves and phases into the local attention module, perform an initial dimension transformation through a 1×1 convolutional layer to obtain the transformed feature map; use an adaptive convolutional layer and a weighting mechanism to process the transformed feature map to obtain local features. Among them, input the input data into the local attention module, and obtain the first processing result through the processing of two layers of 1×1 convolution, batch normalization function, and ReLU function; perform calculations on the first processing result using a 1×1 convolution to obtain the second processing result; further process it through the softmax function and Dropout regularization method to obtain local features. S63. Input the feature data with vehicle waves and phases into the global attention module, and determine the feature relationship between time steps of the feature data with vehicle waves and phases through improved multi-head self-attention and global context attention to obtain global features. Among them, input the input data into the global attention module, process it through the multi-head attention mechanism to obtain the processing result; the residual gating mechanism further processes the processing result, and extracts features through global context attention to obtain global features. Among them, let the given first time step be p, and the feature of the first time step be T p ; let the given second time step be q, and the feature of the second time step be T q , and map the above variables to a common feature space through a feature mapping function, which is represented by the following formula (15): r p，q = f(T p , T q ) (15) where f(·) represents a feature mapping function; r p，q represents the relationship between features at the first time step p and the second time step q; Among them, the feature T at the first time step p p The relationship vector of is represented by the following formula (16): r p ＝[r p，1 ，r p，2 ，…，r p，q ，r p，N ，r 1，p ，r 2，p ，…，r q，p ，r N，p ] (16) where r p represents the relationship vector between the first time step p and the feature T p ; Among them, the features of the time step and the relationship vector are combined into a paired representation, and the global feature representation at the first time step p is obtained as The global feature representation at the second time step q is obtained according to the above steps; S64. Fuse the local features and global features to obtain the fused features. Among them, the fused features are expressed by the following formula (17): F C = F L + F G (17) Among them, F C represents the fused feature; F L represents the local feature; F G represents the global feature; S7. Obtain the state information of the target vehicle within the observation time domain; input the state information of the target vehicle at the observation time into the trained global-local attention mechanism residual bidirectional long short-term memory network to obtain the future trajectory of the target vehicle within the prediction time domain. Among them, the state information of the target vehicle within the observation time domain in S7 includes: the type of the target vehicle, the running state of the target vehicle, and the spatio-temporal relationship between the target vehicle and surrounding obstacle vehicles. Among them, the state information of the target vehicle within the observation time domain is represented by the following formula (18): m = 1, 2, 3…, T obs n＝1,2,3…,6 Among them, represents the input state of the target vehicle within the observation time domain T obs where L represents the length of the predicted target vehicle; W represents the width of the predicted target vehicle; represents the lateral coordinate of the predicted target; represents the longitudinal coordinate of the predicted target; represents its driving style label; represents the speed of the predicted target; represents the acceleration of the predicted target; represents the headway between the predicted target vehicle and the nth obstacle vehicle, represents the time headway between them, m represents the observation time domain; n represents the number of obstacle vehicles; Among them, the future trajectory of the target vehicle within the prediction time domain is represented by the following formula (19): m = 1, 2, 3…, T pred Among them, represents the lateral coordinate of the predicted trajectory of the target vehicle from the current time to the prediction time domain T pred within; represents the longitudinal coordinate of the predicted trajectory of the target vehicle from the current time to the prediction time domain T pred within; m represents the prediction time domain; represents the future trajectory of the target vehicle within the prediction time domain.

2. The vehicle trajectory prediction method for adapting to different driving styles according to claim 1, wherein The initial global-local attention mechanism residual bidirectional long short-term memory network of S5 includes: a global attention module, a local attention module, and a residual bidirectional long short-term memory network; Among them, the global attention module adopts an improved multi-head self-attention mechanism, a residual gating mechanism, and a global context attention mechanism to enhance the extracted global features; The local attention module is used to enhance the extracted local features; The residual bidirectional long short-term memory network is used to determine the long-term dependencies in the input sequence data.

3. The vehicle trajectory prediction method for adapting to different driving styles according to claim 2, wherein The S6 trains the initial global-local attention mechanism residual bidirectional long short-term memory network according to the sample data set after adding hard labels to obtain a trained global-local attention mechanism residual bidirectional long short-term memory network, including: S61. Input the sample data set after adding hard labels into the initial global-local attention mechanism residual bidirectional long short-term memory network. According to the principle of quantum mechanics, model the vehicle as a waveform. By analyzing the superposition effect of vehicle waves, determine the amplitude and phase of the vehicle wave, and obtain feature data with vehicle waves and phases; S62. Input the feature data with vehicle waves and phases into the local attention module, perform an initial dimension conversion through a 1×1 convolutional layer to obtain a converted feature map; use an adaptive convolutional layer and a weighting mechanism to process the converted feature map to obtain local features; S63. Input the feature data with vehicle waves and phases into the global attention module, and determine the feature relationship between time steps of the feature data with vehicle waves and phases through improved multi-head self-attention and global context attention to obtain global features; S64. Fuse the local features and the global features to obtain fused features; S65. Input the fused features into the residual bidirectional long short-term memory network, and calculate the input fused features through a residual block constructed by skip connections to output a prediction result.

4. A vehicle trajectory prediction device adapted to different driving styles, the vehicle trajectory prediction device adapted to different driving styles is used to implement the vehicle trajectory prediction method adapted to different driving styles according to any one of claims 2-3, characterized in that, The device includes: A first acquisition unit for acquiring a sample data set of the original vehicle trajectory; A preprocessing unit for performing filtering processing on the sample data set to obtain a filtered sample data set; Among them, the preprocessing unit is used for: Performing denoising on the sample data set using a five-iteration wavelet transform denoising algorithm to obtain a denoised sample data set; According to the denoised sample data set, calculating the heading angle as a metric for trajectory annotation, and performing balancing processing on the denoised sample data set to obtain a balanced sample data set; Among them, the heading angle is represented by the following formula (1): Among them, x represents the horizontal position of the trajectory point; y represents the vertical position of the trajectory point; t represents the current time step; t + 1 represents the next time step; θ head represents the heading angle; A second acquisition unit for performing dimensionality reduction processing on the filtered sample data set using the principal component analysis method to obtain dimensionally reduced data; A third acquisition unit, configured to perform soft clustering processing on the dimension-reduced data by using a fuzzy C-means clustering algorithm, obtain the membership degrees of the dimension-reduced data for each driving style, determine the hard labels of the dimension-reduced data according to the maximum membership degrees, and add the hard labels to the sample data set after filtering processing; Among them, the third acquisition unit is configured to: Use a fuzzy C-means clustering algorithm to calculate the fuzzy clustering centers of the dimension-reduced data; Among them, in the fuzzy C-means clustering algorithm, the fuzzy clustering centers are determined by minimizing the objective function, where the objective function is represented by the following formula (2): Among them, M represents the fuzzy coefficient that controls the degree of fuzziness in the clustering process, which affects the way of assigning membership values to each driving style in the trajectory sample data (1 ≤ M < ∞); N represents the number of vehicle trajectory samples; C represents the number of driving style categories, that is, the number of clustering centers; c j represents the j-th clustering center; u ij represents the probability that the sample x i belongs to the clustering center c j ; d represents the dimension of the sample; x ik represents the value of the sample x i at the k-th dimension; c jk represents the value of the clustering center c j at the k-th dimension; |·|2 represents the Euclidean norm used to measure the distance between the driving sample and the clustering center; Among them, the membership degree u ij is represented by the following formula (3): Among them, the clustering center c j is represented by the following formula (4): According to the fuzzy clustering centers and the preset iteration conditions, iteratively calculate the membership degree distributions of the dimension-reduced data in different driving styles; According to the membership degree distributions, calculate the membership degrees of each data point in all clusters, allocate the data points to the corresponding clustering centers according to the maximum membership degrees, obtain the hard labels of the sample data, and add the hard labels to the sample data set after filtering processing; Among them, iteratively calculate to obtain the membership degree distributions of each sample data in different driving styles, and stop iterating when reaching the iteration termination condition to obtain the clustering result; where the preset iteration termination condition is represented by the following formula (5): where k′ represents the number of iterations, and ε is a small constant representing the error threshold; when the maximum change in successive iterations does not exceed ε, the iteration stops; ij ​ Among them, the maximum membership degree is represented by the following formula (6): Among them, represents the total number of data points; the closer the FPC value is to 1, the better the clustering effect; Among them, the process of obtaining the hard labels of the sample data is represented by the following (7): Among them, label(x i ) represents the hard label of the sample x i ; A construction unit, configured to construct an initial global-local attention mechanism residual bidirectional long short-term memory network; A training unit, configured to train the initial global-local attention mechanism residual bidirectional long short-term memory network according to the sample data set after adding hard labels, and obtain a trained global-local attention mechanism residual bidirectional long short-term memory network; Among them, the training unit is configured to: Input the sample data set after adding hard labels into the initial global-local attention mechanism residual bidirectional long short-term memory network. According to the principle of quantum mechanics, model the vehicle as a waveform, and determine the amplitude and phase of the vehicle wave by analyzing the superposition effect of the vehicle waves to obtain the feature data with vehicle waves and phases; Among them, when the waves of two or more vehicles meet at the same point, the principle of wave superposition is followed; at this point, the superposition effect of the vehicle waves is represented by the composite vector sum of their respective displacements, and the superposition effect is represented by the following formula (8): where \(i\) m represents the imaginary unit; \(|\cdot|\) represents the absolute value operation; \(\odot\) represents the element-wise multiplication; \(|z\) h | represents the amplitude, \(\theta\) h represents the phase, and \(h\) represents a selected positive integer; Among them, the amplitude of the vehicle wave is affected by the vehicle speed difference, and the phase of the wave represents the vehicle position. When two vehicle waves show the same displacement in the same direction at a certain point, the phase difference at which constructive interference occurs is represented by the following formula (9): Δθ h = θ0 + 2ηπ, η ∈ [0, ±2, ±4, …] (9) where η represents the corresponding selected integer multiple; Δθ h represents the phase difference between two vehicle waves; Among them, when two vehicle waves have opposite displacements at a certain point, the phase difference between the two vehicle waves determines the occurrence of destructive interference, which is represented by the following formula (10): Δθ h = θ0 + ηπ, η ∈ [0, ±1, ±3, …] (10) Among them, the amplitude of the vehicle wave and the phase of the vehicle wave are represented by the following formula (11): Among them, FC represents a standard fully connected layer; ω z represents the first learning parameter; ω θ represents the second learning parameter; f g represents the input feature; θ h represents the phase of the vehicle wave; h represents a selected positive integer; among them, the first learning parameter and the second learning parameter are adjusted according to the input feature to obtain the optimal weights of the amplitude and phase embeddings of the vehicle wave; among them, the representations of the phase and amplitude of the vehicle wave in the negative number domain are represented by the following formula (12): Among the interactions of all vehicles in the driving scenario, the n vehicle waves are expressed as The aggregation of the vehicle waves is expressed by the following formula (13): Among them, represents the complex domain representation after aggregating vehicle waves; Among them, a real number after aggregating the vehicle waves is obtained by weighted summation; where the real number after aggregating the vehicle waves is represented by the following formula (14): Among them, o h represents the real number after aggregating vehicle waves; ω o represents the first learning weight; ω s represents the first learning weight; Input the characteristic data with vehicle wave and phase into the local attention module, perform an initial dimension transformation through a 1×1 convolutional layer to obtain the transformed feature map; use an adaptive convolutional layer and a weighting mechanism to process the transformed feature map to obtain local features; Among them, input the input data into the local attention module, and through the processing of two layers of 1×1 convolution, batch normalization function and ReLU function, obtain the first processing result; use 1×1 convolution to calculate the first processing result to obtain the second processing result; further process through the softmax function and Dropout regularization method to obtain local features; Input the characteristic data with vehicle wave and phase into the global attention module, and determine the feature relationship between time steps of the characteristic data with vehicle wave and phase through improved multi-head self-attention and global context attention to obtain global features; Among them, input the input data into the global attention module, and process it through the multi-head attention mechanism to obtain the processing result; the residual gating mechanism further processes the processing result, and extracts features through global context attention to obtain global features; Among them, let the given first time step be p, and the feature of the first time step be T p ; let the given second time step be q, and the feature of the second time step be T q , and map the above variables into a common feature space through a feature mapping function, which is represented by the following formula (15): r p，q = f(T p , T q ) (15) where f(·) represents a feature mapping function; r p，q represents the relationship between the features at the first time step p and the second time step q; Among them, the feature T at the first time step p p The relationship vector of is represented by the following formula (16): r p ＝[r p，1 ，r p，2 ，…，r p，q ，r p，N r 1，p ，r 2，p ，…，r q，p ，r N，p ] (16) where r p represents the relationship vector between the first time step p and the feature T p ; Among them, the features of the time step are combined with the relationship vector to form a paired representation, and the global feature representation at the first time step p is obtained as The global feature representation at the second time step q is obtained according to the above steps; Fuse the local features and the global features to obtain the fused features; Among them, the fused features are represented by the following formula (17): F C = F L + F G (17) Among them, F C represents the fused feature; F L represents the local feature; F G represents the global feature; A prediction unit, configured to obtain the state information of the target vehicle within the observation time domain; input the state information of the target vehicle at the observation time into the trained global-local attention mechanism residual bidirectional long short-term memory network to obtain the future trajectory of the target vehicle within the prediction time domain; Among them, the state information of the target vehicle within the observation time domain includes: the type of the target vehicle, the running state of the target vehicle, and the spatio-temporal relationship between the target vehicle and surrounding obstacle vehicles; Among them, the state information of the target vehicle within the observation time domain is represented by the following formula (18): m = 1, 2, 3…, T obs n＝1，2，3…，6 Among them, represents the input state of the target vehicle in the observation time domain T obs where L represents the predicted length of the target vehicle; W represents the predicted width of the target vehicle; represents the lateral coordinate of the predicted target; represents the longitudinal coordinate of the predicted target; represents its driving style label; represents the speed of the predicted target; represents the acceleration of the predicted target; represents the headway between the predicted target vehicle and the nth obstacle vehicle, represents the time headway between them, m represents the observation time domain; n represents the number of obstacle vehicles; Among them, the future trajectory of the target vehicle within the prediction time domain is represented by the following formula (19): m = 1, 2, 3…, T pred Among them, represents the lateral coordinate of the predicted trajectory of the target vehicle from the current time to the prediction time domain T pred within; represents the longitudinal coordinate of the predicted trajectory of the target vehicle from the current time to the prediction time domain T pred within; m represents the prediction time domain; represents the future trajectory of the target vehicle within the prediction time domain.

5. A vehicle trajectory prediction device adapted to different driving styles, characterized in that, The vehicle trajectory prediction device adapted to different driving styles includes: A processor; A memory, on which computer-readable instructions are stored, and when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that, Program code is stored in the computer-readable storage medium, and the program code can be called by the processor to execute the method according to any one of claims 1 to 3.

Images