Driving style feature extraction method of variational auto-encoder based on improved LSTM

Through the improved LSTM variational autoencoder, combined with bidirectional LSTM and autoregressive LSTM, the problem of insufficient timing information analysis in driving style analysis is solved, and more accurate driving style recognition and clustering effects are achieved.

CN119004073AActive Publication Date: 2024-11-22DALIAN MARITIME UNIVERSITY

Patent Information

Application Number
CN202411092853.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-22
Estimated Expiration
2044-08-09

AI Technical Summary

Technical Problem

The prior art is difficult to fully capture timing information and maintain the integrity of feature extraction information in driving style analysis, resulting in insufficient accuracy in driving style recognition.

Method used

Adopting an improved LSTM variational autoencoder, combined with a bidirectional LSTM encoder and an autoregressive LSTM decoder, the reconstructed error is corrected by the gated unit, providing more comprehensive timing information analysis and feature extraction to ensure the integrity and consistency of information. .

Benefits of technology

It improves the accuracy and reliability of driving style clustering, and can more accurately identify the operating strategies and behavioral characteristics of different drivers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004073A_ABST
    Figure CN119004073A_ABST
Patent Text Reader

Abstract

The invention discloses a driving style feature extraction method of a variational auto-encoder based on an improved LSTM, and belongs to the technical field of vehicle engineering and data science. The method comprises the steps of preprocessing data set information; constructing a model based on the improved encoder and decoder, and training the model; inputting the preprocessed data, performing driving style feature extraction, feature clustering and driving style analysis, and outputting a corresponding driving style; and verifying the feature extraction performance of the model by using feature clustering and feature analysis results. According to the method, feature extraction is improved, and the accuracy and reliability of driving style clustering are effectively improved while the algorithm complexity is optimized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of vehicle engineering and data science technology, and in particular to a driving style feature extraction method based on an improved LSTM variational autoencoder. Background Art

[0002] Driving style refers to the behavior and operating habits of drivers during driving. Different drivers have different subjective attitudes, experiences and psychological states, which lead to different operating methods and driving characteristics. Analyzing driving style helps to improve driving safety awareness, reduce traffic accidents, optimize driving training, and improve passenger comfort and fuel economy. Traditional driving style identification mainly relies on questionnaires, but this method relies on the subjective answers of drivers and lacks accuracy. With the development of Internet of Vehicles technology and algorithms, data-driven driving style clustering analysis has gradually become mainstream. This method is more objective by collecting natural data from drivers during driving. K-means and hierarchical clustering are often used in the prior art to analyze driving styles and obtain the distribution of behavioral characteristics of different drivers. By constructing an integrated semi-supervised model based on the k-nearest neighbor algorithm (kNN), a high recognition rate can be achieved even with a small sample size. Or based on lateral and longitudinal position changes, hierarchical clustering is used to evaluate the safety of driver behavior. However, in actual application, it is found that the driving data collected by the Internet of Vehicles is of many types and large scale. Therefore, in order to build an accurate clustering model and improve the recognition of driving styles, feature extraction and information compression of high-dimensional data are very important.

[0003] Existing feature extraction methods include statistical-based methods and change-based methods. The statistical-based method describes the driving style by extracting the mean and variance of data such as speed and acceleration, but this method can only reflect the performance of macro data and lacks the description of micro features such as sudden acceleration and deceleration. To solve this problem, the principal component analysis (PCA) method based on linear transformation can be used to improve the clustering or classification effect through dimensionality reduction. However, the above method rarely uses the time dimension information of the driver's style, resulting in the inability to fully obtain the driver's dynamic driving style, and the PCA linear dimensionality reduction method is prone to cause the loss of extracted style information.

[0004] Therefore, there is a need for a feature extraction method that can provide more comprehensive temporal information analysis and improve the integrity of feature extraction information. Summary of the invention

[0005] In view of this, the present invention provides a driving style feature extraction method based on improved LSTM variational autoencoder to maintain the integrity and consistency of feature extraction information, thereby improving the accuracy of driving style analysis.

[0006] To this end, the present invention provides the following technical solutions:

[0007] A driving style feature extraction method based on an improved LSTM variational autoencoder comprises the following steps:

[0008] Preprocess the data set information;

[0009] A driving style feature extraction model is constructed based on the variational autoencoder of the improved LSTM, and the driving style feature extraction model is trained; the driving style feature extraction model uses the variational autoencoder framework as the main framework; and includes: a bidirectional LSTM encoder for feature extraction and an autoregressive LSTM decoder for feature fusion;

[0010] The preprocessed data is input into the driving style feature extraction model to extract reconstructed driving style feature data, including: the preprocessed data is input into a bidirectional LSTM encoder to extract features; the extracted features are re-parameterized to obtain corresponding distributions; the extracted features and the obtained distribution are respectively input into a mixed feature autoregressive LSTM decoder to reconstruct the data.

[0011] Furthermore, the preprocessing of the data set information includes:

[0012] Filter data based on analysis requirements;

[0013] After screening, the data were randomly cut to obtain continuous sequences;

[0014] Convert data units in the dataset to international units;

[0015] Smooth the horizontal and vertical coordinates through Savitzky-Golay filtering;

[0016] According to the smoothed horizontal and vertical coordinates, the velocity acceleration as well as the horizontal and vertical velocity acceleration are recalculated.

[0017] Furthermore, the autoregressive LSTM decoder for fusion features includes a gating unit.

[0018] Furthermore, the autoregressive LSTM decoder for fusion features, the fusion feature process includes:

[0019] The features extracted by the encoder are used as the initial hidden state, and the distribution obtained after reparameterization is used as the input at the initial moment;

[0020] At each time step, the features extracted by the encoder are compared with the h output at the previous moment. t-1 Combined, the features of the encoder are reintroduced for each time step, where the data update formula in the gating unit is:

[0021] g t-1 =σ(W g ·[h t-1 ,h0]+b g );

[0022] h′ t-1 =g t-1 ·h0+(1-g t-1 )·h t-1 ;

[0023] Among them, h t-1 is the hidden state of the previous moment, h0 is the feature extracted by the encoder, σ is the sigmoid activation function, g t-1 is the gating signal, h′ t-1 is the context vector; based on g t-1 To weight the fusion and h t-1 and h0, passed into the next moment as the new hidden state.

[0024] Furthermore, the loss function used by the training model includes reconstruction error and KL divergence:

[0025]

[0026] Among them, the reconstruction error is calculated using the mean square error; x i It is real data. is the reconstructed data, N is the number of data samples, μ j and are the mean and variance of the latent variable z, and J is the dimension of the latent space.

[0027] Furthermore, the driving style feature extraction model further includes:

[0028] Use K-means++ to cluster the reconstructed driving feature data;

[0029] The silhouette coefficient, CH index and SSE evaluation model are used to obtain the optimal number of clusters to divide driving styles.

[0030] The output driving style is divided according to the optimal number of clusters.

[0031] Furthermore, the clustering of the reconstructed driving feature data using K-means++ is also used for:

[0032] The reconstructed driving feature data were clustered using K-means++, and models of different feature extraction methods were constructed. The silhouette coefficient, CH index, and SSE were used to verify the performance of the model feature extraction.

[0033] Furthermore, the outputting of driving styles according to the optimal number of clusters is also used for:

[0034] According to the divided driving styles, the driving characteristics of each driving style are analyzed, including speed, acceleration, and headway characteristics; each feature is further analyzed in more detail, including average speed, frequency of sudden acceleration and deceleration, average headway, and proportion of safe headway characteristics; the accuracy of driving style feature clustering is verified, and the performance of model feature extraction is further verified.

[0035] Advantages and positive effects of the present invention:

[0036] The present invention adopts a variational autoencoder as the main architecture, which optimizes the continuity and interpretability of the latent space while providing effective data reconstruction, thereby improving the data representation capability in the unsupervised learning process; the introduction of a bidirectional LSTM can comprehensively capture the forward and backward dependencies of driving data, and provide richer temporal information analysis; the hybrid feature autoregressive LSTM at the decoder end performs data recovery, and the accumulated error of the autoregressive LSTM reconstruction is corrected by a gating unit, thereby maintaining the integrity and consistency of the information and making data recovery more accurate; through an improved LSTM variational autoencoder method, while optimizing the algorithm complexity, the accuracy and reliability of driving style clustering are effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0038] Figure 1 This is a flow chart of a driving style analysis model according to an embodiment of the present invention;

[0039] Figure 2 A comparison diagram of the lateral and longitudinal positions, speed and acceleration before and after the vehicle data preprocessing in an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of the autoregressive LSTM decoder model of an embodiment of the present invention;

[0041] Figure 4 A schematic diagram of a gate control unit with mixed features according to an embodiment of the present invention;

[0042] Figure 5 It is an overall block diagram of the model in the embodiment of the present invention;

[0043] Figure 6 Determine the optimal number of clusters for SSE in an embodiment of the present invention;

[0044] Figure 7 Determine the optimal number of clusters for the CH index in an embodiment of the present invention;

[0045] Figure 8 Determining the optimal number of clusters for the silhouette coefficient in an embodiment of the present invention;

[0046] Fig. 9 The driving characteristics corresponding to the three clusters in the embodiment of the present invention;

[0047] Fig.10 : is the headway distribution of three clusters in the embodiment of the present invention;

[0048] Fig.11 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0050] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0051] The present invention provides a driving style feature extraction method based on a variational autoencoder of an improved LSTM long short-term memory network (E-LSTM, Enhanced-Long Short-Term Memory), and the concept is as follows: a variational autoencoder is used as the main architecture, so that the model can optimize the continuity and interpretability of the potential space while providing effective reconstruction of data, thereby improving the data representation capability in the unsupervised learning process; a bidirectional LSTM is used on the encoder side, so that the forward and backward dependencies in the driving data can be captured simultaneously, and a more comprehensive temporal information analysis can be provided; a hybrid feature autoregressive LSTM is used on the decoder side for data recovery, and the model gradually recovers data using the characteristics of the autoregressive LSTM, and the features extracted by the initial encoder are mixed through a gating unit, so as to correct the accumulated errors generated by the autoregressive LSTM reconstruction, and maintain the integrity and consistency of the information. Thus, the accuracy and reliability of driving style clustering are effectively improved.

[0052] The process of the present invention is described in conjunction with the driving style analysis model flow chart shown in Figure 1:

[0053] S1. Filter the data in the data set, remove outliers through SG filtering, and recalculate the remaining driving feature information;

[0054] In this embodiment, the NGSIM data set is selected, and the driving style characteristics of the expressway section are taken into consideration without considering the ramps; the driving data with a driving time longer than 30 seconds in lanes 1-5 of US-101 and I-80 in the data set are screened, and the screened data are cut into a continuous sequence of 30 seconds in length, and the data units in the original data set are converted to international units;

[0055] However, although the NGSIM dataset is recognized as the most commonly used dataset for studying driving style and driving intention, recent studies have also shown that there are a lot of inaccurate values ​​of speed and acceleration recorded in the NGSIM dataset. Therefore, the SG filter is used to smooth the lateral and longitudinal coordinates. At the same time, the smoothed coordinates and time information are used to update the speed, acceleration, lateral speed, lateral acceleration, longitudinal speed, and longitudinal acceleration features. The features after filtering and recalculation are different from the original features. For example, Figure 2 shown.

[0056] S2. Use the proposed improved LSTM variational autoencoder to extract driving features;

[0057] Build an improved LSTM variational autoencoder model architecture and train the model;

[0058] Combination Figure 3 , Figure 4 , Figure 5 , Fig.11Explain the model feature extraction part:

[0059] Specifically, the main framework is a variational autoencoder framework, the encoder side is constructed using a bidirectional LSTM, and the decoder side is constructed using an autoregressive LSTM with fused features.

[0060] The preprocessed data is input into the bidirectional LSTM encoder to extract features; the features extracted by the encoder are then reparameterized to obtain the corresponding distribution; the extracted features and the obtained distribution are respectively input into the initial input and initial state of the hybrid feature autoregressive LSTM decoder for data reconstruction; the most critical part of the fusion feature autoregressive LSTM is that, first, the features extracted by the encoder are used as the initial hidden state, and the distribution obtained after reparameterization is used as the input at the initial moment. Second, this study adds a gating unit to the autoregressive LSTM decoder to compare the features extracted by the encoder with the h output at the previous moment at each time step. t-1 By reintroducing the features of the encoder at each time step, the integrity and consistency of the information can be maintained, and the errors can be corrected to ensure that the output of each step is closer to the true pattern of the original data. The data update formulas in the gating unit are:

[0061] g t-1 =σ(W g ·[h t-1 ,h0]+b g );

[0062] h′ t-1 =g t-1 ·h0+(1-g t-1 )·h t-1 ;

[0063] Among them, h t-1 is the hidden state of the previous moment, h0 is the feature extracted by the encoder, σ is the sigmoid activation function, g t-1 is the gating signal, h′ t-1 is the context vector, based on g t-1 To weight the fusion and h t-1 and h0, passed into the next moment as the new hidden state.

[0064] Finally, the model is trained and the best model is obtained through the silhouette coefficient, CH index and SSE evaluation model.

[0065] In this embodiment, a 2-layer bidirectional LSTM is selected as the encoder, and a 1-layer autoregressive LSTM with mixed features is used as the decoder. The driver style representation feature output by the encoder is 10, which is obtained by averaging the bidirectional features extracted last. During the training process, the learning rate is set to 0.001, the batchsize is set to 32, the activation function is the LeakyRelu function, and the Adam optimization algorithm is used to update the weight parameter matrix. The loss function is:

[0066]

[0067] where x i It is real data. is the reconstructed data, N is the number of data samples, μ j and are the mean and variance of the latent variable z, and J is the dimension of the latent space.

[0068] In this embodiment, the experiment is implemented using the PyTorch framework;

[0069] The number of driver style categories is determined by training a data-driven approach, and the model is evaluated by the silhouette coefficient and CH index, specifically:

[0070] After the model training is completed, all high-dimensional time series are input, and the bidirectional LSTM encoder is used to reduce the dimension and extract features. In this embodiment, K-means++ is used to determine the optimal number of clusters, and SSE, CH index and silhouette coefficient are used to determine the optimal number of clusters.

[0071] S3, using K-means++ to perform cluster analysis on the features; and performing driving behavior feature analysis on drivers with different driving styles to verify the accuracy of feature extraction by the method of the present invention. Specifically:

[0072] The fixed clustering method is K-means++, and the corresponding models using different dimensionality reduction methods are compared. The dimensionality reduction methods include: statistical-based methods, PCA dimensionality reduction methods, LSTM-AE, LSTM-DAE, Bi-LSTM-DAE, Bi-LSTM-DAE-AR, and the improved LSTM denoising variational autoencoder of the present invention. When the number of clusters is the optimal number, the contour coefficient and CH index are compared. The comparison results are shown in Table 1, where the decoders of LSTM-AE, LSTM-DAE, and Bi-LSTM-DAE all use a method of reconstructing the features extracted by the encoder by inputting them into each time step instead of generating them step by step, and Bi-LSTM-DAE-AR uses autoregressive LSTM as a decoder. Through the experimental results, it can be found that the model E-LSTM-DVAE proposed by the present invention has an optimal value of the contour coefficient of 0.6824 and an optimal value of the CH index of 26913.34, which is better than the other compared models.

[0073] Table 1

[0074]

[0075]

[0076] In the study of driving style, it is usually divided into 2-4 categories. In this embodiment, experiments are conducted with 2-9 clusters, and the results are as follows: Figure 6 , Figure 7 , Figure 8 The experimental results show that the method of the present invention obtains the best clustering effect when K=3 in the experiments of the three indicators, so the driving styles of the drivers are divided into three categories, namely, aggressive, general and cautious.

[0077] The detailed comparison of the mean speed, the mean absolute value of acceleration and the mean headway of the three clusters is shown in the figure below. Fig. 9 As shown in the figure, the aggressive type of drivers has the highest mean values ​​of speed and absolute acceleration, while the conservative type of drivers has the lowest value among the three, and the number of general type of drivers is in the middle.

[0078] Based on the above three characteristics, more detailed driving behavior characteristics are analyzed;

[0079] In this embodiment, the frequency of sudden acceleration and deceleration and the headway time th are selected for analysis; specifically, the frequency of sudden acceleration and deceleration behaviors of aggressive drivers is often higher, while the frequency of cautious drivers is lower than the general value. The acceleration thresholds are set as follows: 2m / s 2 , 1.5m / s 2 , 1m / s 2 、-2m / s2 、-1.5m / s 2 、-1m / s 2 The frequency of sudden acceleration and deceleration is calculated for all cluster samples, and the results are shown in Table 2. The experimental results show that the frequency of aggressive drivers, whether they accelerate or decelerate suddenly, is the maximum among the three driving styles in the selected 6 thresholds, while the frequency of conservative drivers is the minimum. This experimental result shows that in terms of sudden acceleration and deceleration, the aggressive type is more likely to accelerate and decelerate suddenly than the other two types, while the frequency of conservative drivers is lower than the other two types, which verifies the results of style clustering.

[0080] Table 2

[0081] Acceleration threshold Radical General Conservative 2 0.043 0.030 0.028 1.5 0.090 0.072 0.064 1 0.179 0.154 0.135 -2 0.043 0.030 0.029 -1.5 0.074 0.068 0.058 -1 0.140 0.137 0.112

[0082] For the headway time th, according to previous studies, when the value of th is less than 2s, it indicates that the probability of an accident is high, while when it is greater than 2s, driving is safer. Therefore, this study uses 2s as the threshold for the headway time study and calculates the proportion of drivers with different styles whose th values ​​exceed 2s and are less than 2s, as shown in the following figure: Fig.10 As shown in the figure. The results show that more than 55% of aggressive drivers are in an unsafe driving state, while only 20% of conservative drivers are in an unsafe driving state. This shows the characteristics of the three types of drivers in terms of headway. There are large differences between drivers of different styles. The proportion of aggressive drivers driving with a safe headway is the lowest among the three, while the proportion of conservative drivers driving with a safe headway is much higher than the other two. Therefore, the effectiveness of style clustering is also verified from the headway characteristics.

[0083] Therefore, the present invention has the advantages that the model processes driving trajectory data through a bidirectional LSTM encoder, obtains a continuous potential distribution using a reparameterization technique, and then reconstructs the data through an autoregressive LSTM decoder that fuses features. K-means++ is applied to the features extracted by the encoder for cluster analysis, and the effect of the model is verified using the US-101 and I-80 road section data in the NGSIM data set. It is thus verified that when the features extracted by the model are used for cluster analysis, the CH coefficient can reach 28115.66, the silhouette coefficient can reach 0.6752, and when the number of clusters is determined by the three methods of CH index, silhouette coefficient and SSE, the optimal number of clusters can be obtained as 3, and it is verified that the drivers of the three driving styles have obvious differences in driving behavior characteristics such as speed acceleration and headway, and the present invention can accurately identify different driving styles and reflect the driver's operating strategy.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A driving style feature extraction method based on improved LSTM variational autoencoder, characterized in that: Includes steps: Preprocess the data set information; Building a driving style feature extraction model based on an improved LSTM variational autoencoder, and training the driving style feature extraction model; The driving style feature extraction model uses a variational autoencoder framework as the main framework; including: a bidirectional LSTM encoder for feature extraction and an autoregressive LSTM decoder for feature fusion; The preprocessed data is input into the driving style feature extraction model to extract reconstructed driving style feature data, including: the preprocessed data is input into a bidirectional LSTM encoder to extract features; the extracted features are re-parameterized to obtain corresponding distributions; the extracted features and the obtained distribution are respectively input into a mixed feature autoregressive LSTM decoder to reconstruct the data.

2. The driving style feature extraction method based on improved LSTM variational autoencoder according to claim 1 is characterized in that: The preprocessing of the data set information includes: Filter data based on analysis requirements; After screening, the data were randomly cut to obtain continuous sequences; Convert data units in the dataset to international units; Smooth the horizontal and vertical coordinates through Savitzky-Golay filtering; According to the smoothed horizontal and vertical coordinates, the velocity acceleration as well as the horizontal and vertical velocity acceleration are recalculated.

3. The driving style feature extraction method based on improved LSTM variational autoencoder according to claim 1 is characterized in that: The autoregressive LSTM decoder for fusing features includes a gating unit.

4. The driving style feature extraction method based on improved LSTM variational autoencoder according to claim 1 is characterized in that: The autoregressive LSTM decoder for fusion features, the fusion feature process includes: The features extracted by the encoder are used as the initial hidden state, and the distribution obtained after reparameterization is used as the input at the initial moment; At each time step, the features extracted by the encoder are compared with the h output at the previous moment. t-1 Combined, the features of the encoder are reintroduced for each time step, where the data update formula in the gating unit is: g t-1 =σ(W g ·[h t-1 ,h0]+b g ); h' t-1 =g t-1 ·h0+(1-g t-1 ) h t-1 ; Among them, h t-1 is the hidden state of the previous moment, h0 is the feature extracted by the encoder, σ is the sigmoid activation function, g t-1 is the gating signal, h′ t-1 is the context vector; based on g t-1 To weight the fusion and h t-1 and h0, passed into the next moment as the new hidden state.

5. The driving style feature extraction method based on improved LSTM variational autoencoder according to claim 1 is characterized in that: The loss functions used in the training model include reconstruction error and KL divergence: Among them, the reconstruction error is calculated using the mean square error; x i It is real data. is the reconstructed data, N is the number of data samples, μ j and are the mean and variance of the latent variable z, and J is the dimension of the latent space.

6. The driving style feature extraction method based on improved LSTM variational autoencoder according to claim 1 is characterized in that: The driving style feature extraction model further includes: Use K-means++ to cluster the reconstructed driving feature data; The silhouette coefficient, CH index and SSE evaluation model are used to obtain the optimal number of clusters to divide driving styles. The output driving style is divided according to the optimal number of clusters.

7. The driving style feature extraction method based on improved LSTM variational autoencoder according to claim 6 is characterized in that: The use of K-means++ to cluster the reconstructed driving feature data is also used for: The reconstructed driving feature data were clustered using K-means++, and models of different feature extraction methods were constructed. The silhouette coefficient, CH index, and SSE were used to verify the performance of the model feature extraction.

8. The driving style feature extraction method based on improved LSTM variational autoencoder according to claim 6 is characterized in that: The output driving style divided according to the optimal number of clusters is also used for: According to the divided driving styles, the driving characteristics of each driving style are analyzed, including speed, acceleration, and headway characteristics; each feature is further analyzed in more detail, including average speed, frequency of sudden acceleration and deceleration, average headway, and proportion of safe headway characteristics; the accuracy of driving style feature clustering is verified, and the performance of model feature extraction is further verified.

Citation Information

Patent Citations

  • Variational auto-encoder for text generation

    CN111931461A

  • Industrial internet time series data anomaly detection method and device

    CN116522265A

  • Driving style identification method for driving risk assessment, electronic equipment and medium

    CN117473282A

  • Abnormal traceability method and system for electricity utilization information acquisition system

    CN117648215A

  • Method and device for training a style encoder of a neural network and method for generating a driving style representation representing a driving style of a driver

    US20220327812A1

Cited By

  • Lane change interaction behavior feature analysis method based on Copula

    CN120183207A

  • A Copula-based lane-changing interaction behavior feature analysis method

    CN120183207B