A vehicle trajectory prediction and behavior decision method and system considering driving style

Abstract

The present invention provides a vehicle trajectory prediction and behavior decision method and system considering driving style, and relates to the field of autonomous driving technology. The method comprises: obtaining a driving behavior data set, and performing data processing to obtain a balanced data set for vehicle trajectory prediction, wherein the balanced data set also includes a driving style vector label obtained based on a driving style classifier; constructing a long short-term memory network based on Bayesian optimization fused with a discrete cosine transform attention mechanism for vehicle trajectory prediction; wherein the long short-term memory network is trained and tested using the balanced data set; combining the safety factor, comfort factor, efficiency factor, gain factor of a multi-lane lane change scenario, as well as the driving style vector label and vehicle trajectory prediction information, and using a deep learning model to implement lane change behavior decision. The present invention can improve the accuracy, safety and real-time performance of vehicle behavior decision-making.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a vehicle trajectory prediction and behavior decision-making method and system taking driving style into consideration. Background Art

[0002] Autonomous driving technology has been one of the hot topics in the automotive industry and information technology field in recent years. The development and application of this technology can not only improve traffic safety and increase road utilization, but also is expected to completely change people's travel methods and urban transportation systems. In a relatively mature autonomous driving technology system, the environmental perception module can be compared to the sensory organ of the vehicle, while the decision-making and planning module can be regarded as the intelligent center of the vehicle. Although the existing behavioral decision-making algorithms can ensure driving safety and efficiency to a certain extent, they usually lack the flexibility to handle complex road conditions.

[0003] At present, the existing behavioral decision-making technology still has the following problems:

[0004] 1. Existing behavioral decision-making algorithms are often too idealistic. Most decision-making methods are based on a priori rules. However, in actual vehicle driving, a series of problems will be encountered, including limited coverage of traffic scenarios, difficulty in rule design, and limited scalability, which are difficult to handle with existing technologies.

[0005] 2. Currently, there are still situations where self-driving cars and manually driven cars coexist on the road. In order to ensure the safety and smoothness of road traffic, it is necessary to ensure that the behavior between self-driving cars and manually driven cars is more consistent. When faced with the same driving situation, drivers with different styles will make different behavioral decisions. Some drivers choose to change lanes, while others choose to continue driving in the current lane. However, existing technologies do not take into account the personalized preferences of various types of drivers or passengers.

[0006] 3. To achieve safe, efficient and comfortable driving in complex scenarios such as cities, accurately predicting the driving trajectory of surrounding vehicles is one of the key issues that must be solved. Existing technologies do not consider the complex interactive relationship between the self-vehicle and surrounding vehicles, and do not incorporate the predicted trajectory of the vehicle into the decision-making system, which affects the accuracy of decision-making.

[0007] Therefore, it is particularly important to develop a behavioral decision-making system that can flexibly respond to various traffic environments while ensuring driving safety. Summary of the invention

[0008] In view of the above problems, the purpose of the present invention is to provide a vehicle trajectory prediction and behavior decision-making method and system taking driving style into consideration, which fully considers the driver's personalized driving style, predicts the trajectories of surrounding vehicles, and makes full use of the above information to make behavior decisions, thereby improving the accuracy, safety and real-time performance of vehicle behavior decisions.

[0009] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0010] On the one hand, a vehicle trajectory prediction and behavior decision method considering driving style is provided, the method comprising the following steps:

[0011] S1. Acquire a driving behavior data set, and perform data processing on the driving behavior data set to obtain a balanced data set for vehicle trajectory prediction;

[0012] The balanced data set also includes a driving style vector label obtained by classifying the driving behavior data set based on a driving style classifier;

[0013] S2. Build a long short-term memory network based on Bayesian optimization and discrete cosine transform attention mechanism for vehicle trajectory prediction;

[0014] wherein the balanced data set is used to train and test the long short-term memory network;

[0015] S3. Combining the safety factor, comfort factor, efficiency factor, gain factor of multi-lane lane change scenarios, as well as the driving style vector label and vehicle trajectory prediction information, a deep learning model is used to realize lane change behavior decision.

[0016] Optionally, in step S1, data processing is performed on the driving behavior data set, specifically including:

[0017] Filtering the driving behavior data set based on a wavelet transform denoising algorithm;

[0018] Screen the filtered driving behavior dataset and remove useless features;

[0019] Add new features, including: vehicle heading angle, left lane label and right lane label, lane change label, vehicle speed and acceleration, and driving style vector label obtained by driving style classifier;

[0020] The sliding window method is used to align the driving behavior dataset after adding new features;

[0021] The driving behavior dataset after data alignment is divided into a training set and a test set, wherein 80% of the data is allocated to the training set and 20% of the data is allocated to the test set, thereby obtaining the balanced dataset.

[0022] Optionally, in step S1, classifying the driving behavior dataset based on a driving style classifier to obtain a driving style vector label specifically includes:

[0023] Selecting multiple driving behavior characteristic parameters from the driving behavior data set, including: maximum speed, average speed, speed standard deviation, maximum lateral speed, mean lateral speed, standard deviation of lateral speed, maximum absolute acceleration, mean absolute acceleration, standard deviation of acceleration, maximum absolute lateral acceleration, average absolute lateral acceleration, standard deviation of lateral acceleration, minimum following distance, average following distance, standard deviation of following distance, minimum headway, mean headway, standard deviation of headway, and driving distance;

[0024] Using principal component analysis to reduce the dimension of the driving behavior characteristic parameters;

[0025] Based on the K-means clustering algorithm, a cluster analysis is performed on the driving behavior characteristic parameters after dimensionality reduction, and the driving styles are clustered into two categories, which are labeled with vector 1 and vector 2 respectively to obtain the driving style vector label; among which, vector 1 and vector 2 correspond to aggressive driving style and conservative driving style respectively.

[0026] Optionally, in step S2, constructing a long short-term memory network based on Bayesian optimization fused with discrete cosine transform attention mechanism specifically includes:

[0027] A single-layer convolutional neural network is used to extract potential features from the input target vehicle state observation feature vector;

[0028] The extracted latent features are input into the long short-term memory network, and a frequency-enhanced channel attention mechanism based on discrete cosine transform is introduced;

[0029] The Bayesian optimization algorithm is used to optimize the hyperparameters of the model, and the mean square error function is defined as the objective function minimized by the Bayesian optimization algorithm.

[0030] Optionally, a single-layer convolutional neural network is used to extract potential features, and the target vehicle state observation feature vector Input (t) It is expressed as follows:

[0031] Input (t) =[X (t) ,Y (t) ,v (t) ,a (t) ,D (t) ,T (t) ,L,W]

[0032] Among them, X (t) and Y (t)They represent the lateral and longitudinal coordinates of the predicted target vehicle trajectory, v (t) and a (t) Denote the predicted instantaneous speed and acceleration of the target vehicle, respectively, and D (t) represents the following distance between the surrounding vehicles and the predicted target vehicle, T (t) It represents the headway between the surrounding vehicles and the predicted target vehicle, and L and W are the length and width of the predicted target vehicle, respectively.

[0033] Optionally, the hyperparameters of the model are optimized using a Bayesian optimization algorithm, and the objective function is:

[0034]

[0035] Among them, N t represents the number of samples, represents the true value of the tth sample, Represents the model's predicted value for the tth sample.

[0036] Optionally, in step S3, a deep learning model is used to implement lane change behavior decision, specifically including:

[0037] Combined with the multi-lane lane change scenario, the following factors are considered comprehensively: Safety factor S safe , Comfort Factor C comfort , efficiency factor E efficiency , gain factor G gain , Driving style vector label T style and surrounding vehicle trajectory prediction information P predicted , construct the lane change decision vector as shown below:

[0038] F=f LC (S safe ,C comfort ,E efficiency ,G gain ,T style ,P predicted )

[0039] where F is a three-element lane change decision vector corresponding to the probabilities of lane keeping, left lane change, and right lane change, and f LC (·) is the lane change probability function based on the above input factors.

[0040] Optionally, the safety factor S safe It is expressed as:

[0041] S safety =f S (d LP ,d RP ,d CP ,dth )

[0042] Among them, d LP Indicates the longitudinal distance between the vehicle and the left front vehicle, d RP Indicates the longitudinal distance between the vehicle and the right front vehicle, d CP Indicates the longitudinal distance between the vehicle and the preceding vehicle in the current lane, d th Indicates the minimum safe distance;

[0043] The comfort factor C comfort It is expressed as:

[0044]

[0045] In the formula, x(t) and y(t) represent the horizontal coordinate and vertical coordinate corresponding to the current vehicle driving time t, respectively. x (t) and a y (t) respectively represent the lateral acceleration and longitudinal acceleration corresponding to the current vehicle driving time t;

[0046] The efficiency factor E efficiency It is expressed as:

[0047] E efficiency =f E (s(t),v(t))

[0048] In the formula, s(t) and v(t) represent the driving distance and driving speed corresponding to the current vehicle driving time t respectively;

[0049] The gain factor G gain It is expressed as:

[0050] G gain =fG((v CP -v E ),(v LP -v E ),(v RP -v E ),d LP )-d CP ),(d RP -d CP ))

[0051] Among them, v E represents the vehicle speed, v CP is the speed of the vehicle in front of the current lane, v LP and v RP Represents the speed of the front vehicle in the left lane and the right lane respectively; d CP is the headway between the vehicle and the preceding vehicle in the current lane, d LP and d RPRespectively represent the headway time between the vehicle and the preceding vehicle in the left lane and the right lane;

[0052] Driving style vector label T style and surrounding vehicle trajectory prediction information P predicted It is cascaded with the parameter vectors in the above factors and then input into a fully connected neural network, which ultimately outputs one of three types of behavioral decision results, namely lane keeping, left lane change, or right lane change.

[0053] On the other hand, a vehicle trajectory prediction and behavior decision system considering driving style is provided, which is used to implement any of the above methods, and the system includes:

[0054] An environment perception module, used to obtain a driving behavior data set, perform data processing on the driving behavior data set, and obtain a balanced data set for vehicle trajectory prediction;

[0055] The balanced data set also includes a driving style vector label obtained by classifying the driving behavior data set based on a driving style classifier;

[0056] The trajectory prediction module is used to build a long short-term memory network based on Bayesian optimization and discrete cosine transform attention mechanism to predict vehicle trajectories;

[0057] wherein the balanced data set is used to train and test the long short-term memory network;

[0058] The behavior decision module is used to combine the safety factor, comfort factor, efficiency factor, gain factor, driving style vector label and vehicle trajectory prediction information of multi-lane lane change scenarios, and use a deep learning model to realize lane change behavior decision.

[0059] In another aspect, an electronic device is provided, the electronic device comprising:

[0060] processor;

[0061] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are loaded and executed by the processor, the steps of the vehicle trajectory prediction and behavior decision-making method as described above are implemented.

[0062] On the other hand, a computer-readable storage medium is provided, in which a program code is stored. The program code can be called by a processor to execute the steps of the above-mentioned vehicle trajectory prediction and behavior decision-making method.

[0063] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0064] 1. This invention proposes a new decision-making framework based on deep learning, which can accurately imitate and predict human behavioral decisions. This framework not only fully considers the driver's personalized driving style, but also comprehensively considers the uncertainty of the trajectory of surrounding driving vehicles. Compared with traditional methods, this invention can more accurately simulate the human lane change decision process.

[0065] 2. This paper proposes a new driving style classifier to classify driving trajectories in the driving behavior dataset and convert the classification results into driving style vector labels, which are added to the original dataset as one of the important parameters for trajectory prediction and behavior decision network training. This method helps to more effectively meet the decision-making needs of drivers with different driving styles.

[0066] 3. The present invention designs and trains a long short-term memory network based on the discrete cosine transform attention mechanism for trajectory prediction of surrounding vehicles, and uses the Bayesian optimization algorithm to optimize its parameters. The network can significantly improve the accuracy of vehicle trajectory prediction, enabling the behavior decision system to more efficiently adapt to complex surrounding traffic conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0068] Figure 1 It is a schematic diagram of the overall framework of a vehicle trajectory prediction and behavior decision method considering driving style provided by an embodiment of the present invention;

[0069] Figure 2 is a schematic diagram of the NGSIM data processing flow provided by an embodiment of the present invention;

[0070] Figure 3 is a schematic diagram of a driving style classification process provided by an embodiment of the present invention;

[0071] Figure 4 Schematic diagram of the DCTAM-BOLSTM algorithm framework provided by an embodiment of the present invention;

[0072] Figure 5 is a schematic diagram of a multi-lane complex lane change scenario provided by an embodiment of the present invention;

[0073] Figure 6 It is a factor loading matrix heat map provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] 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 "exemplary" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the word "exemplary" is used to present concepts in a specific way.

[0076] The embodiment of the present invention provides a vehicle trajectory prediction and behavior decision method considering driving style, which can be implemented by an electronic device, which can be a terminal or a server. Figure 1 As shown, the method comprises the following steps:

[0077] S1. Acquire a driving behavior data set, perform data processing on the driving behavior data set, and obtain a balanced data set for vehicle trajectory prediction. The balanced data set also includes a driving style vector label obtained by classifying the driving behavior data set based on a driving style classifier.

[0078] As an optional implementation of the present invention, in order to ensure the reliability of the data, the public data set NGSIM (Next Generation Simulation) is selected as the driving behavior data set for verification. This data set covers hot areas of vehicle-road collaboration research such as structured roads, intersections, and on- and off-ramps on highways. Researchers can use this data set to study driving behaviors such as vehicle following and lane changing, traffic flow analysis, micro-traffic model construction, vehicle motion trajectory prediction, driver intention recognition, and autonomous driving decision planning. For the convenience of subsequent use, the present invention exports the data set to CSV format.

[0079] Furthermore, if Figure 2 As shown, data processing is performed on the driving behavior data set, specifically including:

[0080] In the first step, the driving behavior data set is filtered based on a wavelet transform denoising algorithm.

[0081] Since the NGSIM dataset has certain errors in the acquisition process, especially the lateral velocity error is obvious, it is necessary to smooth it. The present invention uses a wavelet transform denoising operation with 5 cycles to perform trajectory data filtering on the NGSIM original dataset.

[0082] The second step is to screen the filtered driving behavior dataset and remove useless features.

[0083] The NGSIM dataset is mainly based on lane keeping. To ensure the balance and diversity of the dataset (left lane change, right lane change and lane keeping), the present invention labels the trajectory data in advance, filters out the required balanced dataset, and removes useless features.

[0084] The third step is to add new features, mainly including: vehicle heading angle, left lane label and right lane label, lane change label, vehicle speed and acceleration, and driving style vector label obtained by the driving style classifier.

[0085] The fourth step is to use the sliding window method to align the driving behavior dataset after adding new features.

[0086] Since the model requires data sequences with the same length, the present invention uses a sliding window method to obtain the required 8s trajectory data sequence set, where the data sequence length is 80 and the sliding step is 1. The present invention uses the 3s historical vehicle trajectory to predict the vehicle's trajectory for the next 5s, converts the original trajectory data into feature data, and uses it for later model training and prediction.

[0087] In the fifth step, the aligned driving behavior dataset is divided into a training set and a test set, wherein 80% of the data is allocated to the training set and 20% of the data is allocated to the test set, thereby obtaining the balanced dataset.

[0088] The data set is divided into a training set and a test set for research, aiming to provide a balanced data set for the model to ensure the accuracy and reliability of model training and evaluation. The present invention uses random sampling technology to divide the data set. Specifically, 80% of the data is assigned to the training set, and the remaining 20% ​​of the data is assigned to the test set.

[0089] Furthermore, if Figure 3 As shown, the driving behavior dataset is classified based on the driving style classifier to obtain a driving style vector label, which specifically includes:

[0090] A plurality of driving behavior characteristic parameters are selected from the driving behavior data set. As shown in Table 1, 19 characteristic parameters are selected, including maximum speed, average speed, speed standard deviation, maximum lateral speed, mean lateral speed, standard deviation of lateral speed, maximum absolute value of acceleration, mean absolute value of acceleration, standard deviation of acceleration, maximum absolute lateral acceleration, average absolute lateral acceleration, standard deviation of lateral acceleration, minimum following distance, average following distance, standard deviation of following distance, minimum headway, mean headway, standard deviation of headway, and driving distance.

[0091] Table 1: Selection of driving behavior characteristic parameters

[0092]

[0093] In order to improve the effect and efficiency of driving style clustering, the embodiment of the present invention adopts the principal component analysis (PCA) method to reduce the dimension of the driving behavior characteristic parameters, ensure the accuracy and reliability of the clustering results, and thus avoid the adverse effects on the clustering results caused by too many characteristic parameters.

[0094] In order to perform cluster analysis on the data after dimensionality reduction, in the embodiment of the present invention, a cluster analysis is performed on the driving behavior characteristic parameters after dimensionality reduction based on the K-means clustering algorithm, and the driving styles are clustered into two categories, which are respectively labeled with vector 1 and vector 2 to obtain a driving style vector label, and add it to the driving behavior data set; wherein vector 1 and vector 2 correspond to an aggressive driving style and a conservative driving style, respectively.

[0095] S2. Construct a Long Short-Term Memory Network with Discrete Cosine Transform Attention Mechanism and Bayesian Optimization Algorithm (DCTAM-BOLSTM) for vehicle trajectory prediction. The balanced data set is used to train and test the Long Short-Term Memory Network.

[0096] In order to improve the safety of traffic management, it is necessary to accurately predict the lane change trajectories of surrounding vehicles and incorporate them into the vehicle's decision-making process. Therefore, the present invention constructs a vehicle trajectory prediction method based on a long short-term memory network combined with a discrete cosine transform frequency enhanced channel attention mechanism, and uses a Bayesian hyperparameter optimization algorithm to adjust the model parameters to further improve the accuracy of vehicle trajectory prediction.

[0097] The algorithm framework of DCTAM-BOLSTM is as follows Figure 4As shown. First, by exploring the potential relationship between feature vectors, the potential features of the target vehicle state observation feature vector are extracted. The present invention uses a single-layer convolutional neural network (CNN) to apply a series of convolution operations to extract the potential features of the input target vehicle state observation feature vector to obtain spatial and temporal features. In order to ensure that the model captures the potential features in real traffic data, the input data should include parameters such as the predicted target vehicle type, operating status, and spatiotemporal relationship with surrounding vehicles.

[0098] Specifically, a single-layer convolutional neural network is used to extract potential features. The target vehicle state observation feature vector Input (t) It is expressed as follows:

[0099] Input (t) =[X (t) ,Y (t) ,v (t) ,a (t) ,D (t) ,T (t) ,L,W]

[0100] Among them, X (t) and Y (t) They represent the lateral and longitudinal coordinates of the predicted target vehicle trajectory, v (t) and a (t) Denote the predicted instantaneous speed and acceleration of the target vehicle, respectively, and D (t) represents the following distance between the surrounding vehicles and the predicted target vehicle, T (t) It represents the headway between the surrounding vehicles and the predicted target vehicle, and L and W are the length and width of the predicted target vehicle, respectively.

[0101] Next, due to the significant advantages of the Long Short Term Memory Network (LSTM) in trajectory prediction, the present invention inputs the potential features extracted by the convolutional neural network into the LSTM network, and introduces a frequency-enhanced channel attention mechanism based on discrete cosine transform (DCT) in the prediction network. This mechanism can essentially avoid the high-frequency noise caused by the Gibbs phenomenon in the Fourier transform process, thereby effectively improving the accuracy of the prediction.

[0102] Finally, in the model training process, the optimization of hyperparameters is crucial, and the optimal hyperparameter combination can improve the predictive performance and generalization ability of the model. Traditional hyperparameter optimization methods rely on experienced human experts in the field. In contrast, the Bayesian optimization algorithm can search the hyperparameter space more intelligently and help the model find the optimal hyperparameter combination faster. Therefore, the present invention uses the Bayesian optimization algorithm to optimize the hyperparameters of the model, and defines the mean square error (MSE) function as the objective function minimized by the Bayesian optimization algorithm:

[0103]

[0104] Among them, N t represents the number of samples, represents the true value of the tth sample, Represents the model's predicted value for the tth sample. The smaller the mean square error (MSE) value, the smaller the difference between the model's predicted value and the true value, which reflects the better performance of the model.

[0105] The algorithm steps of DCTAM-BOLSTM are shown in Table 2:

[0106] Table 2 DCTAM-BOLSTM algorithm steps

[0107]

[0108]

[0109] S3. Combining the safety factor, comfort factor, efficiency factor, gain factor of multi-lane lane change scenarios, as well as the driving style vector label and vehicle trajectory prediction information, a deep learning model is used to realize lane change behavior decision.

[0110] Vehicle behavior decision-making involves multi-parameter nonlinear problems, which are difficult to solve effectively with traditional mathematical methods. In recent years, deep learning has shown great potential in solving such problems. Therefore, the present invention adopts a deep learning model to solve this problem. The goal of learning is to find an f θ (·) model to minimize the long-term average loss, where θ is a set of model parameters and is designed to take into account Figure 5 The multi-lane complex lane change scenario shown.

[0111] In the embodiment of the present invention, the following factors are comprehensively considered in combination with the multi-lane lane change scenario: safety factor S safe , Comfort Factor C comfort , efficiency factor E efficiency , gain factor G gain , Driving style vector label Tstyle and surrounding vehicle trajectory prediction information P predicted , construct the lane change decision vector as shown below:

[0112] F=f LC (S safe ,C comfort ,E efficiency ,G gain ,T style ,P predicted )

[0113] where F is a three-element lane change decision vector corresponding to the probabilities of lane keeping, left lane change, and right lane change, and f LC (·) is the lane change probability function based on the above input factors.

[0114] Next, the present invention will delve into four key considerations in lane change decision-making: safety, comfort, driving efficiency, and lane change gain.

[0115] 1. Safety factor (S safety ): To ensure safe lane change, it is necessary to evaluate the collision risk with the preceding vehicle in the current lane and the preceding vehicle in the target lane. This evaluation is consistent with the current minimum safe following distance (d th ) is closely related to the distance to the vehicle ahead.

[0116] Safety factor S safe It is expressed as:

[0117] S safety =f S (d LP ,d RP ,d CP ,d th )

[0118] Among them, d LP Indicates the longitudinal distance between the vehicle and the left front vehicle, d RP Indicates the longitudinal distance between the vehicle and the right front vehicle, d CP Indicates the longitudinal distance between the vehicle and the preceding vehicle in the current lane, d th Indicates the minimum safe distance.

[0119] 2. Comfort factor (C comfort ): The comfort factor reflects the driver's requirement for driving comfort, mainly considering the absolute values ​​of the longitudinal and lateral acceleration of the vehicle.

[0120] Comfort Factor C comfort It is expressed as:

[0121]

[0122] In the formula, x(t) and y(t) represent the horizontal coordinate and vertical coordinate corresponding to the current vehicle driving time t, respectively. x (t) and a y (t) represent the lateral acceleration and longitudinal acceleration corresponding to the current vehicle driving time t respectively.

[0123] 3. Efficiency factor (E efficiency ): Driving efficiency reflects the driver's desire to reach the destination quickly, mainly considering the driving speed of the vehicle.

[0124] Efficiency Factor E efficiency It is expressed as:

[0125] E efficiency =f E (s(t),v(t))

[0126] Where s(t) and v(t) represent the driving distance and driving speed corresponding to the current vehicle driving time t, respectively.

[0127] 4. Gain factor (G gain ): The gain factor is mainly reflected in the consideration of speed and headway. Specifically, it mainly considers the speed and distance advantage of the target lane during the left and right lane changes.

[0128] Gain factor G gain It is expressed as:

[0129] G gain =f G ((v CP -v E ),(v LP -v E ),(v RP -v E ),(d LP -d CP ),(d RP -d CP ))

[0130] Among them, v E represents the vehicle speed, v CP is the speed of the vehicle in front of the current lane, v LP and v RP Represents the speed of the front vehicle in the left lane and the right lane respectively; d CP is the headway between the vehicle and the preceding vehicle in the current lane, d LP and d RP Respectively represent the headway time between the vehicle and the vehicle ahead in the left lane and the right lane.

[0131] Driving style vector label T styleand surrounding vehicle trajectory prediction information P predicted It is cascaded with the 13-dimensional parameter vectors in the above factors and then input into a fully connected neural network (FCNN), which ultimately outputs one of three types of behavioral decision results, namely lane keeping, left lane change or right lane change.

[0132] The vehicle trajectory prediction and behavior decision-making method considering driving style proposed in the present invention has the following advantages:

[0133] 1. A new decision-making framework based on deep learning is proposed, which can accurately imitate and predict human behavioral decisions. This framework not only fully considers the driver's personalized driving style, but also comprehensively considers the uncertainty of the trajectory of surrounding driving vehicles. Compared with traditional methods, this invention can more accurately simulate the human lane change decision process.

[0134] 2. A new driving style classifier is proposed to classify the driving trajectories in the driving behavior dataset and convert the classification results into driving style vector labels, which are added to the original dataset as one of the important parameters for trajectory prediction and behavior decision network training. This method helps to more effectively meet the decision-making needs of drivers with different driving styles.

[0135] 3. A long short-term memory network based on discrete cosine transform attention mechanism is designed and trained to predict the trajectory of surrounding vehicles, and the Bayesian optimization algorithm is used to optimize its parameters. This network can significantly improve the accuracy of vehicle trajectory prediction, enabling the behavior decision system to more efficiently adapt to the complex traffic conditions around it.

[0136] In order to study the performance of the proposed vehicle trajectory prediction and behavior decision method, first, the vehicle trajectory is classified using a driving style classifier. Next, the trajectory prediction module will combine the different styles of drivers and the surrounding driving environment to predict the vehicle's trajectory in the future. Finally, the behavior decision module will make reasonable behavior decisions based on the current driving situation, the predicted trajectory of surrounding vehicles, and the driving style of the driver or passengers of the vehicle. The implementation case of the present invention will be carried out on an NVIDIA GeForce RTX 4060 GPU to ensure computational efficiency and performance.

[0137] First, principal component analysis (PCA) was used to reduce the dimension of the selected features and reduce the features to 5 principal components to reduce the dimension and complexity of the data while retaining as much information as possible. The contribution of each principal component was obtained through PCA to evaluate the interpretability of the data after dimensionality reduction. The contribution of each principal component is shown in Table 3.

[0138] Table 3: Contribution of each principal component

[0139]

[0140] In principal component analysis, the contribution value indicates the proportion of the total variance explained by each principal component. Specifically, for a given principal component, the contribution value indicates how much variance it explains in the original data. It can be seen that the contribution of the first principal component is 0.82139042, indicating that the first principal component can explain about 82.14% of the total variance. The contribution of the second principal component is 0.16333355, indicating that the second principal component can explain about 16.33% of the total variance. Similarly, the contributions of the third, fourth, and fifth principal components are 0.00919988, 0.00531227, and 0.00037623, respectively, indicating that they can explain about 0.92%, 0.53%, and 0.04% of the total variance, respectively. The first principal component has the largest contribution, and the contribution decreases in turn, indicating that this method can effectively reduce the dimension and complexity of the data while retaining as much information as possible. Doing so helps to simplify data analysis and visualization, and can better understand the structure and characteristics of the data.

[0141] Next, we conduct factor analysis and use heat maps to visualize the factor loading matrix. Figure 6 As shown, its rows represent factors, columns represent original variables, and the depth of color represents the size of factor loading, which intuitively shows the degree of correlation between variables and factors. By drawing the rotated factor loading matrix, we can get the linear combination of the original indicator variables and find out which component is most correlated with a certain factor. In this way, we can explore the potential factor structure in the data to understand the intrinsic relationship between features.

[0142] In the figure, the horizontal axis represents the five components, the vertical axis represents the 19 characteristic parameters selected above, and the color depth of the factor loading matrix shows the degree of correlation between each principal component and the original variable. It can be seen that the range of values ​​in the figure is usually between -1 and 1, close to 1 indicates a high positive correlation, close to -1 indicates a high negative correlation, and close to 0 indicates a low correlation or no correlation. The higher load value in the heat map indicates that there is a strong correlation between the principal component and the original variable. If the load value of a principal component and a variable is very high, such as the first factor contributes 79.80% to the total variance, and the cumulative contribution of the first two factors is 81.32%, it means that the principal component can well explain the variance of the variable.

[0143] Next, the K-means clustering method is used, and the number of clusters is 2, that is, the samples are divided into two clusters. Then, the fit method is used to fit the data and obtain the clustering results, and the cluster center and the cluster label of each sample are obtained. Finally, the number of samples in each category is counted and printed out. According to the clustering results, there are two clusters, of which the first cluster contains 300 samples and the second cluster contains 20 samples. The average silhouette coefficient (Silhouette-Score) of the cluster is: 0.9342949098434978. The closer the value is to 1, the better the clustering effect is; and the closer it is to -1, the more suitable the sample is to be assigned to different clusters. The CH (Calinski-Harabasz) index is: 442.51558243139345, and the DB (Davies-Bouldin) index is: 0.46819286842833047. Both indicate that the clustering effect is good, the similarity between the data points in the cluster is high, and the distinction between clusters is good.

[0144] After the driving style is effectively classified, the present invention will combine the vehicle driving style vector and the surrounding traffic environment vector, and use the designed DCTAM-BOLSTM model to predict the trajectory of surrounding vehicles. The present invention uses the 3s historical trajectory to predict the trajectory of the next 5s, and the evaluation index is the terminal displacement error of 1 to 5. The experiment was carried out on the NVIDIA GeForce RTX4060GPU, and the code was implemented under the deep learning framework PyTorch. The number of training iterations is 600, the Batch Size is 1024, the optimizer is Adam, the learning rate is 0.001, and the weight decay is 0.0001. Five experiments were conducted on 10 sets of data sets, for a total of 50 sets of experiments. In order to verify the advancedness of the proposed model, the present invention will be compared with the current trajectory prediction model with better performance. The comparison results of the terminal displacement error predicted within 5 seconds by different models are shown in Table 4. Among them, S-LSTM is an LSTM encoding and decoding model using a social pooling layer; M-LSTM is an LSTM encoding and decoding model that takes into account the driving behavior of surrounding vehicles; MHA-LSTM is an LSTM encoding and decoding model that integrates a multi-head attention mechanism; and DCTAM-BOLSTM is the LSTM encoding and decoding model proposed in the present invention.

[0145] Table 4: Comparison of the error of different models predicting the end point displacement within 5 seconds

[0146]

[0147] It can be seen that the model proposed in the present invention has a root mean square error of the end point displacement within 5 seconds within 1.35m in terms of vehicle trajectory prediction, and its prediction ability is significantly better than other models. Therefore, the vehicle trajectory prediction model proposed in the present invention can significantly improve the model's ability to learn real trajectory data.

[0148] Finally, the vehicle driving style vector label and the surrounding vehicle trajectory prediction information obtained by the trajectory prediction module are cascaded with the traffic information vector contained in the safety, comfort, driving efficiency and lane change gain mentioned above, and then input into the fully connected neural network (FCNN). In order to verify the decision accuracy of the proposed model, the present invention will be compared with the support vector machine (SVM) model based on the supervised learning algorithm, the rule-based model MOBIL, and the learning-based LSTM decision model. The comparison results are shown in Table 5.

[0149] Table 5: Comparison of decision accuracy of different behavioral decision models

[0150]

[0151]

[0152] From the numerical results, it can be seen that the behavior decision-making model proposed in the present invention, which takes into account the personalized driving style and the driving trajectory of surrounding vehicles, has effectively improved the prediction accuracy compared with other models.

[0153] Accordingly, an embodiment of the present invention further provides a vehicle trajectory prediction and behavior decision system considering driving style, the system comprising:

[0154] An environment perception module, used to obtain a driving behavior data set, perform data processing on the driving behavior data set, and obtain a balanced data set for vehicle trajectory prediction;

[0155] The balanced data set also includes a driving style vector label obtained by classifying the driving behavior data set based on a driving style classifier;

[0156] The trajectory prediction module is used to build a long short-term memory network based on Bayesian optimization and discrete cosine transform attention mechanism to predict vehicle trajectories;

[0157] wherein the balanced data set is used to train and test the long short-term memory network;

[0158] The behavior decision module is used to combine the safety factor, comfort factor, efficiency factor, gain factor, driving style vector label and vehicle trajectory prediction information of multi-lane lane change scenarios, and use a deep learning model to realize lane change behavior decision.

[0159] The system of this embodiment can be used to execute Figure 1 The technical solution of the method embodiment shown has similar implementation principles and technical effects, which will not be repeated here.

[0160] The vehicle trajectory prediction and behavior decision-making method and system considering driving style proposed in the present invention can better imitate the driving operation of human drivers by considering the predicted trajectories of vehicles with different driving styles in a highly interactive, dynamic and complex traffic environment, so as to achieve accurate and efficient behavior decision-making.

[0161] In an exemplary embodiment, the present invention further provides an electronic device, the electronic device comprising:

[0162] processor;

[0163] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are loaded and executed by the processor, the steps of the vehicle trajectory prediction and behavior decision-making method as described above are implemented.

[0164] In an exemplary embodiment, the present invention further provides a computer-readable storage medium, wherein at least one instruction is stored in the computer-readable storage medium, and the at least one instruction is loaded and executed by a processor to implement the steps of the vehicle trajectory prediction and behavior decision method as described above. For example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0165] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0166] References in the specification to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments", etc. indicate that the described embodiments may include a particular feature, structure, or characteristic, but not every embodiment may include the particular feature, structure, or characteristic. In addition, when a particular feature, structure, or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the relevant art to implement such feature, structure, or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0167] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0168] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0169] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0170] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0171] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0173] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0174] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0175] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A vehicle trajectory prediction and behavior decision method considering driving style, characterized in that: The following steps are involved: S1. Acquire a driving behavior data set, and perform data processing on the driving behavior data set to obtain a balanced data set for vehicle trajectory prediction; The balanced data set also includes a driving style vector label obtained by classifying the driving behavior data set based on a driving style classifier; S2. Build a long short-term memory network based on Bayesian optimization and discrete cosine transform attention mechanism for vehicle trajectory prediction; wherein the balanced data set is used to train and test the long short-term memory network; In step S2, a long short-term memory network based on Bayesian optimization fused with discrete cosine transform attention mechanism is constructed, which specifically includes: A single-layer convolutional neural network is used to extract potential features from the input target vehicle state observation feature vector; the target vehicle state observation feature vector Input(t) is expressed as follows: Input (t) =[X (t) ,Y (t) ,v (t) ,a (t) ,D (t) ,T (t) ,L,W] Where X(t) and Y(t) represent the lateral and longitudinal coordinates of the predicted target vehicle trajectory, respectively; v(t) and a(t) represent the instantaneous velocity and acceleration of the predicted target vehicle, respectively; D(t) represents the following distance between the surrounding vehicles and the predicted target vehicle; T(t) represents the headway between the surrounding vehicles and the predicted target vehicle; L and W represent the length and width of the predicted target vehicle, respectively; The extracted latent features are input into the long short-term memory network, and a frequency-enhanced channel attention mechanism based on discrete cosine transform is introduced; The Bayesian optimization algorithm is used to optimize the model's hyperparameters, and the mean square error function is defined as the objective function minimized by the Bayesian optimization algorithm; S3. Combining the safety factor, comfort factor, efficiency factor, gain factor of multi-lane lane change scenarios, as well as the driving style vector label and vehicle trajectory prediction information, a deep learning model is used to realize lane change behavior decision.

2. The vehicle trajectory prediction and behavior decision method according to claim 1, characterized in that: In step S1, data processing is performed on the driving behavior data set, specifically including: Filtering the driving behavior data set based on a wavelet transform denoising algorithm; Screen the filtered driving behavior dataset and remove useless features; Add new features, including: vehicle heading angle, left lane label and right lane label, lane change label, vehicle speed and acceleration, and driving style vector label obtained by driving style classifier; The sliding window method is used to align the driving behavior dataset after adding new features; The driving behavior dataset after data alignment is divided into a training set and a test set, wherein 80% of the data is allocated to the training set and 20% of the data is allocated to the test set, thereby obtaining the balanced dataset.

3. The vehicle trajectory prediction and behavior decision method according to claim 1, characterized in that: In the step S1, the driving behavior data set is classified based on the driving style classifier to obtain a driving style vector label, which specifically includes: Selecting multiple driving behavior characteristic parameters from the driving behavior data set, including: maximum speed, average speed, speed standard deviation, maximum lateral speed, mean lateral speed, standard deviation of lateral speed, maximum absolute acceleration, mean absolute acceleration, standard deviation of acceleration, maximum absolute lateral acceleration, average absolute lateral acceleration, standard deviation of lateral acceleration, minimum following distance, average following distance, standard deviation of following distance, minimum headway, mean headway, standard deviation of headway, and driving distance; Using principal component analysis to reduce the dimension of the driving behavior characteristic parameters; Based on the K-means clustering algorithm, a cluster analysis is performed on the driving behavior characteristic parameters after dimensionality reduction, and the driving styles are clustered into two categories, which are labeled with vector 1 and vector 2 respectively to obtain the driving style vector label; among which, vector 1 and vector 2 correspond to aggressive driving style and conservative driving style respectively.

4. The vehicle trajectory prediction and behavior decision method according to claim 1, characterized in that: The Bayesian optimization algorithm is used to optimize the hyperparameters of the model, and the objective function is: Among them, N t represents the number of samples, represents the true value of the tth sample, Represents the model's predicted value for the tth sample.

5. The vehicle trajectory prediction and behavior decision method according to claim 1, characterized in that: In step S3, a deep learning model is used to implement lane change behavior decision, which specifically includes: Combined with the multi-lane lane change scenario, the following factors are considered comprehensively: Safety factor S safe , Comfort Factor C comfort , efficiency factor E efficiency , gain factor G gain , Driving style vector label T style and surrounding vehicle trajectory prediction information P predicted , construct the lane change decision vector as shown below: F=f LC (S safe ,C comfort ,E efficiency ,G gain ,T style ,P predicted ) where F is a three-element lane change decision vector corresponding to the probabilities of lane keeping, left lane change, and right lane change, and f LC (·) is the lane change probability function based on the above input factors.

6. The vehicle trajectory prediction and behavior decision method according to claim 5, characterized in that: The safety factor S safe It is expressed as: S safety =f S (d LP ,d RP ,d CP ,d th ) Among them, d LP Indicates the longitudinal distance between the vehicle and the left front vehicle, d RP Indicates the longitudinal distance between the vehicle and the right front vehicle, d CP Indicates the longitudinal distance between the vehicle and the preceding vehicle in the current lane, d th Indicates the minimum safe distance; The comfort factor C comfort It is expressed as: In the formula, x(t) and y(t) represent the horizontal coordinate and vertical coordinate corresponding to the current vehicle driving time t, respectively. x (t) and a y (t) respectively represent the lateral acceleration and longitudinal acceleration corresponding to the current vehicle driving time t; The efficiency factor E efficiency It is expressed as: E efficiency =f E (s(t),v(t)) In the formula, s(t) and v(t) represent the driving distance and driving speed corresponding to the current vehicle driving time t respectively; The gain factor G gain It is expressed as: G gain =f G ((v CP -v E ),(v LP -v E ),(v RP -v E ),(d LP -d CP ),(d RP -d CP )) Among them, v E represents the vehicle speed, v CP is the speed of the vehicle in front of the current lane, v LP and v RP Represents the speed of the front vehicle in the left lane and the right lane respectively; d CP is the headway between the vehicle and the preceding vehicle in the current lane, d LP and d RP Respectively represent the headway time between the vehicle and the preceding vehicle in the left lane and the right lane; Driving style vector label T style and surrounding vehicle trajectory prediction information P predicted It is cascaded with the parameter vectors in the above factors and then input into a fully connected neural network, which ultimately outputs one of three types of behavioral decision results, namely lane keeping, left lane change, or right lane change.

7. A vehicle trajectory prediction and behavior decision system considering driving style, the system being used to implement the method as claimed in any one of claims 1 to 6, characterized in that: The system comprises: An environment perception module, used to obtain a driving behavior data set, perform data processing on the driving behavior data set, and obtain a balanced data set for vehicle trajectory prediction; The balanced data set also includes a driving style vector label obtained by classifying the driving behavior data set based on a driving style classifier; The trajectory prediction module is used to build a long short-term memory network based on Bayesian optimization and discrete cosine transform attention mechanism to predict vehicle trajectories; wherein the balanced data set is used to train and test the long short-term memory network; The behavior decision module is used to combine the safety factor, comfort factor, efficiency factor, gain factor, driving style vector label and vehicle trajectory prediction information of multi-lane lane change scenarios, and use a deep learning model to realize lane change behavior decision.

8. An electronic device, characterized in that: The electronic device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are loaded and executed by the processor, the method according to any one of claims 1 to 6 is implemented.

Images