Driving style identification method and device, computing equipment and vehicle

Abstract

The invention relates to the technical field of vehicles, in particular to a driving style recognition method and device, computing equipment and a vehicle. The driving style identification method comprises the steps that driving data of a user in at least one driving scene is acquired, an expert model corresponding to the at least one driving scene is determined, the expert model comprises a hybrid auto-encoder and a clustering network, and the driving style is identified based on the expert model corresponding to the at least one driving scene and the driving data in the at least one driving scene. The driving style of at least one driving scene is determined, the expert model carries out driving feature learning on the input driving data through a time convolution network TCN and a structured state space network Mamba, and the driving style is determined based on the learned driving features; and determining the driving style of the user based on the driving style of the at least one driving scene.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a driving style recognition method, device, computing equipment and vehicle. Background Art

[0002] Driving style recognition is the process of analyzing a driver's vehicle operating behavior, habits, and driving data, and using technical means to determine their driving style category. It is usually combined with sensors, algorithms, or artificial intelligence to quantitatively evaluate and classify driving behavior, and is applied to traffic safety, intelligent driving, insurance, and other fields.

[0003] In related technologies, driving operation data and the corresponding speed sequence are used as samples. A pre-trained driving style feature extraction model is then used to extract driving features. These features are then clustered, and the cluster with the highest score is selected as the optimal clustering result.

[0004] However, there are large differences between driving data in different driving scenarios. Directly processing the driving data in different driving scenarios in a unified manner cannot effectively highlight the differences between different driving scenarios, and the accuracy of the driving style evaluation results is not high. Summary of the Invention

[0005] The present invention provides a driving style recognition method, device, computing equipment and vehicle, which improve the recognition accuracy of driving style.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present application provides a method for identifying a driving style, which includes: obtaining driving data of a user in at least one driving scenario. Determining an expert model corresponding to at least one driving scenario, and determining the driving style of the at least one driving scenario based on the expert model corresponding to the at least one driving scenario and the driving data in the at least one driving scenario. Determining the user's driving style based on the driving style of the at least one driving scenario. The expert model includes a hybrid autoencoder and a clustering network. The expert model learns driving features of the input driving data through a temporal convolutional network (TCN) and a structured state space network (Mamba), and determines the driving style based on the learned driving features.

[0008] The technical solution provided by the embodiment of the present application is that multiple driving scenarios correspond to their own expert models. The expert models are used to perform feature extraction and cluster analysis on the driving data under multiple driving scenarios, and the driving characteristics and driving styles corresponding to different scenarios are learned in a targeted manner. This avoids interference between driving data of a single driving scenario and driving data of different driving scenarios, improves the generalization of driving style recognition, and thus improves the accuracy of driving style recognition results. At the same time, each expert model recognizes synchronously, improving the efficiency of driving style recognition. In addition, the hybrid autoencoder in the expert model, combined with TCN and Mamba, can deeply learn the low-dimensional representation of driving characteristics, retaining as much information as possible while reducing the dimensionality, making the extracted driving characteristics more accurate, and further improving the accuracy of driving style recognition results.

[0009] One possible implementation involves determining the driving style of at least one driving scenario based on an expert model corresponding to at least one driving scenario and driving data from the at least one driving scenario. This can be specifically accomplished by extracting features from the driving data for a target driving scenario within the at least one driving scenario to obtain a first feature. The driving style of the target driving scenario is then determined based on the first feature and the expert model corresponding to the target driving scenario. The first feature includes at least one of the following: a statistical feature, a frequency domain feature, and a time domain feature; the statistical dimensions of the statistical feature include, but are not limited to, mean, maximum, minimum, variance, and standard deviation. This initial feature extraction of the driving data reduces noise and redundant information. Simultaneously, the features of the driving data are decoupled from multiple dimensions, such as time domain features, frequency domain features, and statistical features, improving the processing efficiency of the expert model.

[0010] In one possible implementation, a hybrid autoencoder includes a TCN and Mamba. Based on a first feature and an expert model corresponding to the target driving scenario, the driving style of the target driving scenario is determined. Specifically, this can be achieved by using the TCN in the expert model corresponding to the target driving scenario to perform dimensionality reduction extraction on the first feature to obtain a second feature for the target driving scenario. Using Mamba in the expert model corresponding to the target driving scenario, the second feature is mapped into a latent space to obtain the driving characteristics of the target driving scenario. Based on the driving characteristics of the target driving scenario and the clustering network in the expert model corresponding to the target driving scenario, the driving style of the target driving scenario is determined. Combining TCN and Mamba for feature extraction not only efficiently processes long sequence data to extract multi-scale features, but also enables Mamba to better capture long-term dependencies in the sequence, improving the comprehensiveness of driving feature extraction and enabling the driving features to better represent information related to driving style.

[0011] One possible implementation involves determining a user's driving style based on the driving style of at least one driving scenario. This can be specifically accomplished by determining the user's driving time in at least one driving scenario, determining a weight for the at least one driving scenario based on the ratio of the user's driving time in the at least one driving scenario to the user's total driving time, and determining the user's driving style based on the weight for the at least one driving scenario and the driving style. By assigning weights to different driving scenarios based on driving time, the contribution of driving styles in different driving scenarios to the driving style recognition results is rationally distributed, further improving the accuracy of driving style recognition results.

[0012] In one possible implementation, driving data includes at least vehicle power data and operational data. Power data includes speed, longitudinal acceleration, lateral acceleration, motor speed, and motor torque, while operational data includes accelerator pedal position, brake pedal position, and steering angle. By assessing a user's driving style through multiple dimensions, including power data and operational data, the driving data used in driving style recognition is more comprehensive and diverse, avoiding errors caused by a single data source and further improving the accuracy of driving style recognition results.

[0013] In one possible implementation, the hybrid autoencoder further includes a decoding layer. The training process of the expert model corresponding to the target driving scenario includes: extracting features from the sample driving data in the target driving scenario to obtain a first sample feature. Utilizing the TCN and Mamba in the expert model corresponding to the target driving scenario, the first sample feature is subjected to dimensionality reduction extraction and mapping to obtain a sample driving feature. Utilizing the decoding layer in the expert model corresponding to the target driving scenario, the sample driving feature is subjected to feature reconstruction to obtain a sample reconstructed feature. Based on the sample driving feature and the clustering network in the expert model corresponding to the target driving scenario, the sample driving style in the target driving scenario is determined. Utilizing the reconstruction loss between the first sample feature and the sample reconstructed feature and the clustering loss corresponding to the sample driving style, the expert model corresponding to the target driving scenario is optimized. The parameters of the expert model are jointly optimized through the reconstruction loss and the clustering loss, so that the expert model can fully learn the data representing the driving style and improve the accuracy of the driving style recognition results.

[0014] One possible implementation is to use the Kullback-Leibler (KL) divergence of the clustering network as the clustering loss. KL divergence directly measures the distribution differences in the clustering process and is highly interpretable. It also limits the complexity of the expert model and prevents overfitting.

[0015] One possible implementation of the structured state-space model is the selective structured state-space model. This model is computationally more efficient and can dynamically select key information, filter redundancy and noise, and improve the expressive power of the expert model.

[0016] One possible implementation involves determining the driving style for a target driving scenario based on the driving characteristics of the target driving scenario and the clustering network in the corresponding expert model. Specifically, this can be achieved by clustering the driving characteristics of the target driving scenario and the clustering network in the corresponding expert model to obtain the style weight and number of style events corresponding to each driving style. A driving style score for the target driving scenario is then determined based on the style weight and number of style events. This score is then used to quantitatively evaluate the driving style, improving the accuracy and precision of the driving style results.

[0017] In a second aspect, the present application provides a driving style identification device, which includes an acquisition module, a processing module and an identification module.

[0018] The acquisition module is used to obtain driving data of a user in at least one driving scenario.

[0019] The processing module is used to determine an expert model corresponding to at least one driving scenario.

[0020] The recognition module is used to determine the driving style of at least one driving scenario based on the expert model corresponding to the at least one driving scenario and the driving data in the at least one driving scenario, and is also used to determine the driving style of the user based on the driving style of the at least one driving scenario.

[0021] Among them, the expert model includes a hybrid autoencoder and a clustering network. The expert model learns the driving characteristics of the input driving data through the temporal convolutional network TCN and the structured state space network Mamba, and determines the driving style based on the learned driving characteristics.

[0022] In one possible implementation, the recognition module is specifically configured to: for a target driving scenario in at least one driving scenario, extract features from driving data for the target driving scenario to obtain a first feature; and determine the driving style for the target driving scenario based on the first feature and an expert model corresponding to the target driving scenario. The first feature includes at least one of the following: a statistical feature, a frequency domain feature, and a time domain feature; the statistical dimensions of the statistical feature include, but are not limited to, mean, maximum, minimum, variance, and standard deviation.

[0023] In one possible implementation, a hybrid autoencoder includes a TCN and Mamba. The recognition module is specifically configured to: utilize the TCN in the expert model corresponding to the target driving scenario to perform dimensionality reduction extraction on the first feature to obtain a second feature for the target driving scenario; utilize Mamba in the expert model corresponding to the target driving scenario to map the second feature into a latent space to obtain driving characteristics for the target driving scenario; and determine the driving style for the target driving scenario based on the driving characteristics of the target driving scenario and the clustering network in the expert model corresponding to the target driving scenario.

[0024] In one possible implementation, the identification module is specifically configured to: determine a driving time of a user in at least one driving scenario; determine a weight corresponding to the at least one driving scenario based on a ratio of the driving time of the user in the at least one driving scenario to the user's driving time; and determine the user's driving style based on the weight corresponding to the at least one driving scenario and the driving style.

[0025] In one possible implementation, the driving data includes at least the vehicle's power data and operation data. The power data includes speed, longitudinal acceleration, lateral acceleration, motor speed and motor torque. The operation data includes accelerator pedal opening, brake pedal opening and steering angle.

[0026] In one possible implementation, the hybrid autoencoder further includes a decoding layer. The processing module is further configured to: extract features from the sample driving data in the target driving scenario to obtain a first sample feature. Utilize the TCN and Mamba in the expert model corresponding to the target driving scenario to perform dimensionality reduction extraction and mapping on the first sample feature to obtain a sample driving feature. Utilize the decoding layer in the expert model corresponding to the target driving scenario to perform feature reconstruction on the sample driving feature to obtain a sample reconstructed feature. Based on the sample driving feature and the clustering network in the expert model corresponding to the target driving scenario, determine the sample driving style in the target driving scenario. Utilize the reconstruction loss between the first sample feature and the sample reconstructed feature and the clustering loss corresponding to the sample driving style to optimize the expert model corresponding to the target driving scenario.

[0027] One possible implementation method is that the clustering loss is the KL divergence of the clustering network.

[0028] In a possible implementation, the structured state space model is a selective structured state space model.

[0029] In one possible implementation, the recognition module is specifically configured to: cluster the driving characteristics of a target driving scenario and a clustering network in an expert model corresponding to the target driving scenario to obtain a style weight and a number of style events corresponding to each driving style; and determine a driving style score for the target driving scenario based on the style weight and the number of style events.

[0030] In a third aspect, the present application provides a computing device, which includes the driving style recognition device in any embodiment of the second aspect.

[0031] In a fourth aspect, a vehicle is provided, the vehicle comprising the computing device according to the third aspect.

[0032] It should be noted that various possible implementations of any of the above aspects can be combined under the premise that the solutions are not contradictory. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of the structure of a computer system provided for an exemplary embodiment;

[0034] Figure 2 A flowchart of a driving style identification method provided by an exemplary embodiment;

[0035] Figure 3 A schematic diagram of the structure of an expert model provided for an exemplary embodiment;

[0036] Figure 4 A schematic diagram of the internal structure of an expert model provided for an exemplary embodiment;

[0037] Figure 5 A schematic structural diagram of a driving style recognition system provided by an exemplary embodiment;

[0038] Figure 6 A schematic structural diagram of a driving style recognition device provided by an exemplary embodiment;

[0039] Figure 7 A schematic structural diagram of a computing device is provided for an exemplary embodiment. DETAILED DESCRIPTION

[0040] In the embodiments of the present application, in order to clearly describe the technical solutions of the embodiments of the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that words such as "first" and "second" do not limit the quantity or execution order, and words such as "first" and "second" do not necessarily mean different. There is no order of precedence or priority between the technical features described by "first" and "second".

[0041] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner to facilitate understanding.

[0042] In the embodiments of the present application, at least one can also be described as one or more, and multiple can be two, three, four or more, which is not limited in this application.

[0043] For ease of understanding, the driving style recognition method, apparatus, computing device, and vehicle provided in this application are described in detail below with reference to the accompanying drawings.

[0044] The solution provided in this application can be applied to Figure 1 In the computer system shown, Figure 1 As shown, the computer system provided in the embodiment of the present application includes a computing device 100. The computing device 100 can be a high-performance server, which serves as the core of the computer system and is used to obtain driving data 200 of the user in at least one driving scenario, and perform driving style recognition based on the driving data 200 and the expert model corresponding to the driving data to obtain the user's driving style. The computing device 100 can directly / indirectly obtain the driving data 200 and perform driving style recognition based on the driving data 200. The computing device 100 can also accept instructions from staff and flexibly configure parameters involved in expert model reasoning and training based on the instructions. The "acquisition" of the computing device 100 includes any terms with acquisition functions such as query, discovery, extraction, etc., and this application is not limited to this.

[0045] Alternatively, the computing device 100 may be an independent physical server, a server cluster or distributed system composed of multiple physical servers, embedded hardware for real-time simulation, or a cloud server providing cloud computing services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and cloud servers for basic cloud computing services such as big data. The embodiments of the present application do not limit the implementation method and application scenarios of the computing device 100.

[0046] like Figure 2 As shown, the driving style recognition method provided by the embodiment of the present application includes:

[0047] Step S201: The computing device obtains driving data of the user in at least one driving scenario.

[0048] Driving data includes vehicle power data and operational data. Power data includes speed, longitudinal acceleration, lateral acceleration, motor speed, and motor torque. Operational data includes accelerator pedal opening, brake pedal opening, and steering angle. Driving scenarios include highways, residential roads, commercial streets, and other common user driving scenarios. This application is not limited to this. Table 1 shows the signal fields of driving data provided in the embodiments of this application.

[0049] Table 1: Signal fields of driving data

[0050] Field Name meaning unit id Unique identifier for the vehicle - terminal_time time s ESP_VehicleSpeed speed m / s ESP_LatAccel lateral acceleration <![CDATA[m / s 2 ]]> ESP_LongAccel Longitudinal acceleration <![CDATA[m / s 2 ]]> SAS_SteeringAngle Steering angle degrees PCU_AccPedl Accelerator pedal opening degrees PCU_BrkPedlSts Brake pedal opening degrees

[0051] In some embodiments, initial driving data of a user in at least one driving scenario is obtained, and the initial driving data is preprocessed to obtain driving data.

[0052] Exemplarily, pre-processing includes identifying and processing outliers in the initial driving data.

[0053] Optionally, the abnormal value includes a value that is greatly different from other initial driving data and a plurality of consecutive values ​​that have not changed.

[0054] For example, the mean μ and standard deviation σ of the initial driving data are calculated, and data values ​​exceeding μ + 3σ are identified as outliers. A boxplot is also constructed based on the quartiles of the initial driving data, and data values ​​above or below the boxplot distribution boundaries are identified as outliers. For example, a data visualization can be used to select data values ​​outside the 95th and 5th quartiles as outliers. Outliers identified by the standard deviation and boxplot can be cross-checked.

[0055] For example, outliers are processed by deletion, sampling, etc. For example, values ​​that differ significantly from other initial driving data are deleted. Multiple consecutive unchanged values ​​are downsampled to avoid errors caused by data redundancy.

[0056] Since the initial driving data may be missing during the collection and cloud data transmission process, initialization also includes deleting the data segment when it is determined that there are continuous missing data in a certain section of the initial driving data, and filling the data segment based on the forward data filling method and average value when a small amount of missing data occurs.

[0057] When the initial driving data is affected by frequent random fluctuations or obvious periodic changes, and the random fluctuations and changes are not the core of driving style identification, the initial driving data is smoothed by using Kalman filtering or other filtering methods to remove or reduce the interference of noise and additional factors.

[0058] Step S202: The computing device determines an expert model corresponding to at least one driving scenario.

[0059] Expert models are systems or algorithms based on the knowledge and experience of experts in a specific field. They are designed to simulate the decision-making capabilities of human experts and solve complex problems within a specific field. Each driving scenario has its own corresponding expert model. Each expert model infers driving data from that specific scenario. Through division of labor and collaboration, they can improve the accuracy of driving style recognition in complex driving scenarios.

[0060] For example, if Figure 3 As shown, based on the driving data, the driving scenarios are divided into driving scenario 1, driving scenario 2, and driving scenario 3. Expert model 1 processes the driving data of driving scenario 1, expert model 2 processes the driving data of driving scenario 2, and expert model 3 processes the driving data of driving scenario 3. The processing results of expert model 1, expert model 2, and expert model 3 are integrated to determine the driving style.

[0061] Exemplarily, the expert model includes a hybrid autoencoder and a clustering network.

[0062] Among them, the hybrid autoencoder includes an encoder and a decoding layer. The hybrid autoencoder achieves data compression and reconstruction by learning to map the input data to a latent space and then mapping the representation back to the original space. Specifically, the encoder is used to convert the input data into an encoded representation of a hidden layer, and the decoding layer is used to restore the encoded representation to reconstructed data close to the original input. The training goal of the autoencoder is to minimize the error between the input data and the reconstructed data. Specifically, the mean square error is used as the loss function to measure the difference between the input data and the reconstructed data. In addition, the nonlinearity is introduced by the activation function, which enables the hybrid autoencoder to capture and learn the complex patterns and structures in the data, and then effectively extract features from the driving data that are helpful for driving style recognition.

[0063] For example, the encoder's mapping function formula is as follows:

[0064] h=f(wx+b)

[0065] Where: f is the activation function of the encoder, w is the weight of the encoder, b is the bias vector, x is the input data, and h is the encoded representation of the hidden layer.

[0066] The mapping function formula of the decoding layer is as follows:

[0067] x′=g(w′h+b′)

[0068] Where g is the activation function of the decoding layer, w′ is the weight matrix of the decoding layer, b′ is the bias vector of the decoding layer, h is the encoding representation of the hidden layer, and x′ is the reconstructed data.

[0069] The formula of the loss function is as follows:

[0070] L(x,x′)=||xx′|| 2

[0071] Among them, L(x,x′) represents the loss function, x represents the input data, and x′ represents the reconstructed data.

[0072] Optionally, the hybrid autoencoder is an autoencoder constructed by a long short-term memory network (LSTM), a gated recurrent unit (GRU), etc.

[0073] In some embodiments, the encoder in the hybrid autoencoder includes TCN and Mamba.

[0074] TCN is a deep learning architecture for processing time series data. It is used to reduce the dimensionality of input data and obtain secondary features. TCN can better process time series data, capture longer-term dependencies and global features, and retain as much information as possible.

[0075] Mamba is used to map the second feature into a smaller latent space to obtain driving features, which can retain as much information as possible.

[0076] Because the encoding process of the autoencoder is unsupervised, the decoder layer in the hybrid autoencoder needs to reconstruct the encoder output to obtain reconstructed data. The reconstructed data is then compared with the input data. Based on the comparison results, the error of the hybrid autoencoder is continuously reduced to ensure that the features after dimensionality reduction can effectively represent the input data. For example, the decoding layer consists of an upsampling layer and a deconvolutional layer.

[0077] For example, the training steps for the expert model corresponding to the target driving scenario include:

[0078] Step 1: Extract features from the sample driving data under the target driving scenario to obtain the first sample features.

[0079] Step 2: Use TCN and Mamba in the expert model corresponding to the target driving scenario to perform dimensionality reduction extraction and mapping on the first sample feature to obtain the sample driving feature.

[0080] Step 3: Use the decoding layer in the expert model corresponding to the target driving scenario to reconstruct the sample driving features and obtain the sample reconstructed features.

[0081] Step 4: Based on the sample driving characteristics and the clustering network in the expert model corresponding to the target driving scenario, determine the sample driving style under the target driving scenario.

[0082] Step 5: Optimize the expert model corresponding to the target driving scenario using the reconstruction loss between the first sample feature and the sample reconstruction feature and the clustering loss corresponding to the sample driving style.

[0083] In some embodiments, the clustering network is clustered using a KL divergence-based clustering method, and the clustering loss is the KL divergence of the clustering network.

[0084] In some embodiments, Mamba is a selective structural state-space model that can dynamically adjust parameters to adapt to the context of the input sequence.

[0085] Among them, the state space model is used to describe the dynamic behavior of the system in time. Specifically, the state of the system is described by state variables, and the mathematical model of the system is established through state equations and observation equations.

[0086] For example, the state space model can map the input sequence x(t) to the output sequence y(t) through the potential state h(t), which can be expressed as:

[0087] h(t)′=Ah(t)+Bx(t)

[0088] y(t)=Ch(t)

[0089] Where h(t)' represents the rate of change or derivative of the potential state at time t, indicating the change of the state over time. A represents the state transfer matrix, A∈R N×N . B represents the input matrix, B∈R D×N . C represents the output matrix, C∈R N×D The variables N and D represent the state expansion factor and dimensionality factor, respectively. A and B can be discretized by zero-order hold and time sampling with an interval of Δ, which is specifically expressed as the following formula:

[0090]

[0091]

[0092] in, represents the discretized state transfer matrix, Represents the discretized input matrix. The discretized state space model can be expressed as:

[0093]

[0094] y t =Ch t

[0095] Among them, h t represents the potential state at discrete time t, h t-1 Represents the potential state at the last discrete time t-1, y t represents the output vector at discrete time t, x t Represents the output vector at discrete time t.

[0096] In some embodiments, the discretized state-space model is trained in parallel using convolution operations and inference is performed using a recurrent neural network.

[0097] Step S203: The computing device determines a driving style for at least one driving scenario based on an expert model corresponding to the at least one driving scenario and driving data for the at least one driving scenario, and determines a driving style of the user based on the driving style for the at least one driving scenario.

[0098] Exemplarily, the expert model learns driving characteristics of the input driving data through TCN and Mamba, and determines the driving style based on the learned driving characteristics.

[0099] In some embodiments, for a target driving scene in at least one driving scene, feature extraction is performed on driving data under the target driving scene to obtain a first feature, and the driving style of the target driving scene is determined based on the first feature and an expert model corresponding to the target driving scene.

[0100] Optionally, the first feature includes a statistical feature, a frequency domain feature, and a time domain feature. The statistical dimensions of the statistical feature include an average value, a maximum value, a minimum value, a variance, and a standard deviation.

[0101] For example, the driving data under different driving scenarios are segmented according to a preset sliding window to obtain multiple driving segments. The first feature of the driving data in each driving segment is calculated respectively. For example, the driving data under different driving scenarios are segmented according to a time window with a time length of t seconds (such as 60 seconds) and a sliding step of Δt seconds (such as 10 seconds), and the driving segments obtained by segmentation are stored as a matrix.

[0102] For example, the standard deviation is calculated as follows:

[0103]

[0104] Among them, x i represents the i-th data point in the driving segment, represents the average value of driving points in the driving clip, t represents the number of data points in the driving clip, and S represents the standard deviation.

[0105] Optionally, the statistical dimensions of the frequency domain features include kurtosis, skewness, wavelet energy, spectral entropy, average frequency, etc.

[0106] For example, the kurtosis is calculated as follows:

[0107]

[0108] Among them, x i represents the i-th data point in the driving clip, μ represents the mean of the driving points in the driving clip, σ represents the standard deviation of the driving points in the driving clip, t represents the number of data points in the driving clip, and K represents the kurtosis.

[0109] For example, the time-domain features capture the changes in users' instantaneous driving behaviors by creating a time series data stream at the user level, and use the time-varying driving fluctuations as a measure of the aggressiveness of the driving style.

[0110] Among them, time-varying driving fluctuation is used to characterize the degree to which driving data deviates from the normal state. The higher the time-varying driving fluctuation, the greater the fluctuations in the vehicle's lateral and longitudinal speed, acceleration, and deceleration during the user's driving process, and the higher the user's driving instability.

[0111] For example, the calculation formula for time-varying driving fluctuations is as follows:

[0112]

[0113] Among them, x i represents the i-th data point in the driving segment, x i-1 represents the i-1th data point in the driving segment, Represents r i The average value of r i The number of V f Indicates time-varying driving fluctuations.

[0114] Specifically, the TCN in the expert model corresponding to the target driving scenario is used to perform dimensionality reduction extraction on the first feature to obtain the second feature of the target driving scenario. The Mamba in the expert model corresponding to the target driving scenario is used to map the second feature into a latent space to obtain the driving characteristics of the target driving scenario. Based on the driving characteristics of the target driving scenario and the clustering network in the expert model corresponding to the target driving scenario, the driving style of the target driving scenario is determined.

[0115] In some embodiments, the clustering network performs clustering based on a KL divergence clustering method to determine the driving style of the target driving scene.

[0116] For example, the K-means++ algorithm is used to initialize the centroids, the data points of the driving characteristics are divided into k clusters, and a centroid is calculated for each cluster to obtain an initial centroid set consisting of k centroids. Then, the probability of each data point to the cluster center is calculated by Student's distribution. The soft distribution calculation method of the data point to each cluster center is as follows:

[0117]

[0118] Among them, siml(z i ,w j ) represents the data point z i With cluster center w j similarity, α is the smoothing parameter, q i,j represents z i Belong to w j The probability of k is the number of clusters.

[0119] In unsupervised learning, data points are distinguished by introducing target distribution. The calculation formula of target distribution is as follows:

[0120]

[0121] Among them, q i,j is the soft distribution of data points to cluster centers, f i is the probability of the soft distribution, p i,j is the target distribution, and k represents the number of clusters.

[0122] In some embodiments, the clustering loss KL divergence of the clustering network is provided, and minimizing the clustering loss can enable the low-dimensional sequence to be better distributed into k clusters by the clustering model. Specifically, the clustering method based on KL divergence is a clustering method based on probability distribution differences, which measures the asymmetric index of the difference between two probability distributions (the probability distribution (P distribution) indicating that the data point belongs to the cluster and the probability distribution of the center of the cluster (Q distribution)). The calculation formula of KL divergence from distribution P to distribution Q is as follows:

[0123]

[0124] Among them, P represents the probability distribution of a data point belonging to a cluster, and Q represents the probability distribution of the center of the cluster.

[0125] The KL divergence of the embodiment of the present application is the KL divergence between the soft distribution and the target distribution, and is calculated as follows:

[0126]

[0127] Among them, Cluloss represents KL divergence, pi,j represents the target distribution, q i,j Represents a soft distribution.

[0128] In some embodiments, a hybrid autoencoder and a clustering network are jointly optimized to improve the recognition accuracy of the expert model.

[0129] Optionally, the types of driving styles include conservative, normal, and aggressive.

[0130] In some embodiments, based on the driving characteristics of the target driving scenario and a clustering network in an expert model corresponding to the target driving scenario, a style weight and a number of style events corresponding to each driving style are obtained by clustering, and a driving style score for the target driving scenario is determined based on the style weight and the number of style events.

[0131] For example, the driving style type is determined based on the driving score, including conservative, normal, and aggressive, providing a more refined identification of driving style. For example, a driving score between 0 and 30 indicates a conservative driving style; a driving score between 30 and 70 indicates a normal driving style; and a driving score between 70 and 100 indicates an aggressive driving style.

[0132] For example, the driving score is calculated as follows:

[0133]

[0134] Among them, S represents the driving score, f represents the score mapping relationship, i represents the i-th driving scene, j represents the j-th driving style, and w j represents the style weight corresponding to the jth driving style in the i-th driving scene, count j The number of style events for the jth driving style in the i-th driving scenario is represented by . The number of style events is the number of data points in each cluster during the clustering process. The membership degree obtained by clustering can be used as the style weight.

[0135] In some embodiments, the driving time of the user in at least one driving scenario is determined, the weight corresponding to the at least one driving scenario is determined based on the ratio of the driving time of the user in at least one driving scenario to the driving time of the user, and the driving style of the user is determined based on the weight corresponding to the at least one driving scenario and the driving style.

[0136] For example, the ratio of the user's driving time in driving scenario A to the user's total driving time is 1 / 3, and the ratio of the user's driving time in driving scenario B to the user's total driving time is 2 / 3. Then the driving score = 1 / 3 * the score of driving scenario A + 2 / 3 * the score of driving scenario B.

[0137] like Figure 4 As shown, the expert model of this embodiment of the present application includes a hybrid autoencoder 400 and a clustering network 410. Hybrid autoencoder 400 outputs driving features based on a first input feature, which enables clustering network 410 to output a driving style recognition result based on the driving features. Hybrid autoencoder 400 and clustering network 410 can jointly optimize based on clustering loss and reconstruction loss to improve the accuracy of driving style recognition results.

[0138] like Figure 5 As shown, an embodiment of the present application also provides a driving style recognition system, comprising a driving data acquisition unit 501, a preprocessing unit 502, a first feature extraction unit 503, an encoding and clustering unit 504, and a driving score calculation unit 505. The data acquisition unit 501 is used to collect initial driving data from a user under different driving scenarios. The driving data includes vehicle power data and operation data. The power data includes vehicle speed, longitudinal acceleration, lateral acceleration, motor speed, and motor torque, and the operation data includes accelerator pedal opening, brake pedal opening, and steering angle. The preprocessing unit 502 is used to perform preprocessing operations such as data padding and outlier processing on the collected initial driving data. The first feature extraction unit 503 is used to extract statistical features, frequency domain features, and time domain features. The encoding and clustering unit 504 is used to determine driving style using a hybrid autoencoder based on TCN-Mamba and a clustering network based on KL divergence. The driving score calculation unit 505 is used to cluster the clustering networks under multiple driving scenarios to obtain a weighted sum of driving styles, determine a driving score, and quantify the driving style.

[0139] like Figure 6 As shown, the driving style recognition device provided in the embodiment of the present application may include an acquisition module 601, a processing module 602 and an identification module 603. Among them, the acquisition module 601 is used to perform Figure 2 In the operation of step S201 of the illustrated method, the processing module 602 is used to perform Figure 2 In the operation of step S202 of the illustrated method, the identification module 603 is used to perform Figure 2 The operation of step S203 in the illustrated method.

[0140] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of the method. In order to realize the above functions, the data transmission device or computing device includes a hardware structure and / or software module corresponding to the execution of each function. It should be easily appreciated by those skilled in the art that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in a hardware or computer software driven hardware manner depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0141] In the embodiments of the present application, the data transmission device or computing device can be divided into functional modules according to the above-mentioned data transmission method. For example, the driving style recognition device or computing device can include functional modules corresponding to the functional divisions, or two or more functions can be integrated into one processing module. The above-mentioned integrated modules can be implemented in the form of hardware or software functional modules. It should be noted that the division of modules in the embodiments of the present application is schematic and is only a logical functional division. In actual implementation, other division methods may be used.

[0142] like Figure 7 As shown, the computing device provided in the embodiment of the present application may include a processor 701, a bus 702, a communication interface 703, and a memory 704. The processor 701, the memory 704, and the communication interface 703 communicate with each other via the bus 702. It should be understood that the present application does not limit the number of processors and memories in the network device.

[0143] The bus 702 may be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, or a universal serial bus (USB). Buses may be classified as address buses, data buses, control buses, etc. For ease of presentation, Figure 7 The bus 702 may include a path for transmitting information between various components of the network device (eg, memory 704, processor 701, communication interface 703).

[0144] The processor 701 may include any one or more processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0145] The memory 704 may include a volatile memory, such as a random access memory (RAM). The processor 701 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0146] The communication interface 703 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the network device and other devices or a communication network.

[0147] The memory 704 stores executable program codes, and the processor 701 executes the executable program codes to implement the functions of the aforementioned method embodiments. That is, the memory 704 stores instructions for executing the aforementioned driving style recognition method.

[0148] On the other hand, a computer-readable storage medium is provided, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to implement the driving style recognition method provided in the above-mentioned method embodiments.

[0149] On the other hand, a computer program product is provided. The computer program product includes a computer program or instructions. When the computer program or instructions are executed by a processor, the driving style recognition method provided by the above method embodiments is implemented.

[0150] It should be noted that when the instructions in the above-mentioned computer-readable storage medium or one or more instructions in the computer program product are executed by the processor of the computing device, the various processes of the above-mentioned method embodiment are implemented and the same technical effect as the above-mentioned method can be achieved. To avoid repetition, they will not be repeated here.

[0151] The driving style recognition method, apparatus, and computing device provided in the embodiments of the present application can be applied in a vehicle. A vehicle may also be referred to as a vehicle, mobile carrier, electric vehicle (EV), hybrid electric vehicle (HEV), plug-in hybrid electric vehicle (PHEV), fuel cell vehicle (FCV), autonomous vehicle, intelligent and connected vehicle (ICV), driverless vehicle, etc.

[0152] In the embodiments of this application, the vehicle may be a sedan, a sport utility vehicle (SUV), a truck, an electric vehicle, a motorcycle, a tricycle, a special vehicle (such as an ambulance, fire truck, police car, etc.), a driverless taxi, an intelligent network-connected bus, an autonomous logistics vehicle, an electric truck, etc. Furthermore, this method is also applicable to various special-purpose vehicles, such as agricultural vehicles, mining vehicles, forestry vehicles, airport vehicles, and port vehicles. This application does not impose any specific restrictions on this.

[0153] Through the description of the above implementation methods, technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete the full classification or partial functions described above.

[0154] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods, 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.

[0155] The units described as separate components may or may not be physically separate, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0156] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or in other words, the part that contributes to the prior art or the full classification part or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute the full classification part or part of the steps of the method of each embodiment of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk or an optical disk.

[0157] The above are only specific implementation methods of this application, but the scope of protection of this application is not limited to them. Any simple modifications (such as adjustments to the analysis model, boundary conditions, and design variables), equivalent changes, and replacements (such as adjustments to constraints and their weights, and increases or decreases in evaluation indicators) made to the above specific implementation methods based on the technical essence of this application without departing from the content of this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for identifying a driving style, characterized in that: The driving style recognition method includes: Obtaining driving data of the user in at least one driving scenario; Determining an expert model corresponding to at least one of the driving scenarios, wherein the expert model includes a hybrid autoencoder and a clustering network; Determining a driving style for at least one driving scenario based on an expert model corresponding to at least one driving scenario and driving data for at least one driving scenario; wherein the expert model learns driving characteristics of the input driving data using a temporal convolutional network (TCN) and a structured state-space network (Mamba), and determines the driving style based on the learned driving characteristics; A driving style of a user is determined based on the driving style of at least one of the driving scenarios.

2. The driving style recognition method according to claim 1, characterized in that: The determining, based on the expert model corresponding to the at least one driving scenario and the driving data for the at least one driving scenario, the driving style of the at least one driving scenario includes: For at least one target driving scene in the driving scenes, feature extraction is performed on driving data in the target driving scene to obtain a first feature; wherein the first feature includes at least one of the following: a statistical feature, a frequency domain feature, and a time domain feature; and the statistical dimensions of the statistical feature include but are not limited to a mean value, a maximum value, a minimum value, a variance, and a standard deviation; A driving style of the target driving scenario is determined based on the first feature and an expert model corresponding to the target driving scenario.

3. The driving style recognition method according to claim 2, characterized in that: The hybrid autoencoder includes the TCN and the Mamba; The determining the driving style of the target driving scene based on the first feature and the expert model corresponding to the target driving scene includes: Using the TCN in the expert model corresponding to the target driving scene, performing dimensionality reduction extraction on the first feature to obtain a second feature under the target driving scene; Mapping the second feature into a latent space using Mamba in the expert model corresponding to the target driving scene to obtain driving features of the target driving scene; The driving style of the target driving scene is determined based on the driving characteristics of the target driving scene and a clustering network in an expert model corresponding to the target driving scene.

4. The driving style recognition method according to claim 1, characterized in that: The determining the user's driving style based on the driving style of at least one of the driving scenarios includes: determining a driving time of the user in at least one of the driving scenarios; determining a weight corresponding to at least one of the driving scenarios based on a ratio of the driving time of the user in at least one of the driving scenarios to the driving time of the user; The driving style of the user is determined based on a weight corresponding to at least one of the driving scenarios and the driving style.

5. The driving style recognition method according to any one of claims 1 to 4, characterized in that: The driving data includes at least power data and operation data of the vehicle, the power data includes speed, longitudinal acceleration, lateral acceleration, motor speed and motor torque, and the operation data includes accelerator pedal opening, brake pedal opening and steering angle.

6. The method for identifying driving style according to claim 3, characterized in that: The hybrid autoencoder further includes a decoding layer; the training process of the expert model corresponding to the target driving scenario includes: Extracting features from the sample driving data under the target driving scenario to obtain a first sample feature; Using the TCN and Mamba in the expert model corresponding to the target driving scenario, perform dimensionality reduction extraction and mapping on the first sample features to obtain sample driving features; Reconstructing the sample driving features using a decoding layer in an expert model corresponding to the target driving scenario to obtain sample reconstructed features; determining a sample driving style under the target driving scenario based on the sample driving characteristics and a clustering network in an expert model corresponding to the target driving scenario; The expert model corresponding to the target driving scene is optimized by utilizing the reconstruction loss between the first sample feature and the sample reconstruction feature and the clustering loss corresponding to the sample driving style.

7. The driving style recognition method according to claim 6, characterized in that: The clustering loss is the KL divergence of the clustering network.

8. The driving style recognition method according to claim 4, characterized in that: The structured state space model is a selective structured state space model.

9. The driving style recognition method according to claim 3, characterized in that: The determining the driving style of the target driving scenario based on the driving characteristics of the target driving scenario and a clustering network in an expert model corresponding to the target driving scenario includes: clustering to obtain a style weight and a number of style events corresponding to each driving style based on the driving characteristics of the target driving scene and a clustering network in an expert model corresponding to the target driving scene; A driving style score of the target driving scenario is determined based on the style weight and the number of style events.

10. A driving style recognition device, characterized in that: The driving style recognition device includes an acquisition module, a processing module and a recognition module; The acquisition module is used to acquire driving data of the user in at least one driving scenario; The processing module is configured to determine an expert model corresponding to at least one of the driving scenarios, wherein the expert model comprises a hybrid autoencoder and a clustering network; The recognition module is configured to determine a driving style for at least one driving scenario based on an expert model corresponding to at least one driving scenario and driving data for at least one driving scenario; wherein the expert model learns driving characteristics of the input driving data using a temporal convolutional network (TCN) and a structured state space network (Mamba), and determines the driving style based on the learned driving characteristics; The recognition module is further configured to determine the user's driving style based on the driving style of at least one of the driving scenarios.

11. A computing device, characterized in that The computing device applies the driving style recognition apparatus according to claim 10 .

12. A vehicle, characterized in that: The vehicle includes the computing device of claim 11.

Images