Vehicle driving style identification method and device, electronic equipment and storage medium

CN118953374APending Publication Date: 2024-11-15CHINA FAW CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411203386.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

但是,这些现有的驾驶风格识别模型在分类准确性方面仍存在不足,同时由于不同地区的驾驶习惯和交通环境的差异,识别精度可能受到一定影响,进而导致相关技术中车辆的驾驶风格识别精度较低

Benefits of technology

[0014] According to another aspect of the embodiment of the present invention, a computer-readable storage medium is also provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the storage medium is located is controlled to execute the present invention. Methods in various embodiments of the invention.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118953374A_ABST
    Figure CN118953374A_ABST
Patent Text Reader

Abstract

The invention discloses a vehicle driving style identification method and device, electronic equipment and a storage medium. The method comprises the steps that in the vehicle driving process, in response to a received driving style recognition task of a vehicle, multiple state features of the vehicle and the current weight of the multiple state features are obtained, and the current weight is used for representing the weight occupied by the multiple state features in the driving style recognition task; identifying the plurality of state features by using a first neural network to obtain a weight matrix of the plurality of state features, the weight matrix being used for representing weights of the plurality of state features in the driving style identification task through different preset values; adjusting the current weights of the plurality of state features based on the weight matrix to obtain target weights of the plurality of state features; and performing driving style identification on the plurality of state features based on the target weight to obtain a target driving style. According to the invention, the technical problem of low vehicle driving style identification precision in the prior art is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of vehicles, and in particular to a method, device, electronic equipment and storage medium for identifying a vehicle's driving style. Background Art

[0002] Driving style learning and recognition is an important branch of the field of autonomous driving technology. It has a significant impact on improving vehicle fuel economy, driving safety, and vehicle reliability. In order to achieve efficient driving style recognition, relevant technicians have proposed many recognition models, such as logistic regression models, tree-based models, and deep learning-based models. However, these existing driving style recognition models still have shortcomings in classification accuracy. At the same time, due to differences in driving habits and traffic environments in different regions, the recognition accuracy may be affected to a certain extent, which in turn leads to low vehicle driving style recognition accuracy in related technologies.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] Embodiments of the present invention provide a method, device, electronic device and storage medium for identifying a driving style of a vehicle, so as to at least solve the technical problem of low accuracy in identifying a driving style of a vehicle in the related art.

[0005] According to one aspect of an embodiment of the present invention, a method for identifying a driving style of a vehicle is provided, comprising: during vehicle driving, in response to receiving a driving style identification task for the vehicle, obtaining multiple state features of the vehicle and current weights of the multiple state features, wherein the current weights are used to represent the weights of the multiple state features in the driving style identification task; identifying the multiple state features using a first neural network to obtain a weight matrix of the multiple state features, wherein the weight matrix is ​​used to represent the weights of the multiple state features in the driving style identification task through different preset values; adjusting the current weights of the multiple state features based on the weight matrix to obtain target weights of the multiple state features; and identifying the driving style of the multiple state features based on the target weights to obtain a target driving style.

[0006] Furthermore, the first neural network includes a fully connected layer and an activation layer, the fully connected layer includes: a first fully connected layer and a second fully connected layer, the activation layer includes: a first activation layer and a second activation layer, and the first neural network is used to identify multiple state features to obtain a weight matrix of the multiple state features, including: using the first fully connected layer to map the multiple state features to a first dimension to obtain a first vector of the first dimension, wherein the first dimension is smaller than the second dimension of the multiple state features; using the first activation layer to activate the first vector to obtain a second vector, wherein the second vector is used to represent the weight of the first vector in a nonlinear manner; using the second fully connected layer to map the second vector to the second dimension to obtain a third vector; using the second activation layer to activate the third vector to obtain a weight matrix.

[0007] Further, the third vector is activated by the second activation layer to obtain a weight matrix, including: using the second activation layer to assign multiple output values ​​corresponding to the third vector based on the first preset value and the second preset value to obtain the weight matrix.

[0008] Furthermore, a second activation layer is used to assign multiple output values ​​corresponding to the third vector based on the first preset value and the second preset value to obtain a weight matrix, including: assigning a first output value among the multiple output values ​​based on the first preset value to obtain a first assignment result, wherein the first output value is an output value of the multiple output values ​​that is less than or equal to the first preset value; assigning a second output value among the multiple output values ​​based on the second preset value to obtain a second assignment result, wherein the second output value is an output value of the multiple output values ​​that is greater than the first preset value; and determining the weight matrix based on the first assignment result and the second assignment result.

[0009] Furthermore, the first fully connected layer is used to map the multiple state features to the first dimension to obtain a first vector of the first dimension, including: compressing the multiple state features to obtain multiple compressed state features of the third dimension, wherein the third dimension is smaller than the second dimension and the third dimension is larger than the first dimension; and the first fully connected layer is used to map the multiple compressed state features to the first dimension to obtain the first vector.

[0010] Furthermore, driving style recognition is performed on multiple state features based on target weights to obtain a target driving style, including: enhancing multiple state features based on the target weights to obtain multiple enhanced state features; using a second neural network to perform driving style recognition on the multiple enhanced state features to obtain the target driving style, wherein an input layer of the second neural network is connected to an output layer of the first neural network, and the output layer of the first neural network is used to output the target weights of the multiple state features.

[0011] Furthermore, the vehicle driving style recognition method also includes: obtaining multiple sample state features and sample driving styles corresponding to the multiple sample state features; using a first initial neural network to perform driving style recognition on the multiple sample state features to obtain a predicted driving style; constructing a target loss function based on the sample driving style and the predicted driving style; and adjusting the model parameters of the first initial neural network based on the target loss function to obtain a first neural network.

[0012] According to another aspect of an embodiment of the present invention, a driving style recognition device for a vehicle is also provided, including: an acquisition module, used to acquire multiple state characteristics of the vehicle and current weights of the multiple state characteristics in response to receiving a driving style recognition task for the vehicle during driving of the vehicle, wherein the current weights are used to represent the weights of the multiple state characteristics in the driving style recognition task; a first recognition module, used to identify the multiple state characteristics using a first neural network to obtain a weight matrix of the multiple state characteristics, wherein the weight matrix is ​​used to represent the weights of the multiple state characteristics in the driving style recognition task through different preset values; an adjustment module, used to adjust the current weights of the multiple state characteristics based on the weight matrix to obtain target weights of the multiple state characteristics; and a second recognition module, used to perform driving style recognition on the multiple state characteristics based on the target weights to obtain a target driving style.

[0013] According to another aspect of an embodiment of the present invention, there is further provided an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods of implementing the various embodiments of the present invention when running.

[0014] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, which includes a stored executable program, wherein when the executable program is running, the device where the storage medium is located is controlled to execute to implement the methods in various embodiments of the present invention.

[0015] In an embodiment of the present invention, first, after receiving a driving style recognition task for a vehicle, multiple state features of the vehicle and current weights of the multiple state features are obtained; then, the multiple state features are recognized using a first neural network to obtain a weight matrix of the multiple state features; secondly, the current matrix of the state features is adjusted using the obtained weight matrix to obtain a target matrix of the multiple state features; finally, the driving style of the multiple state features is recognized based on the target weights to obtain a target driving style of the vehicle. It is easy to notice that the present application obtains the current weights of multiple state features, and uses the first neural network to obtain the weight matrix of multiple state features, and then uses the weight matrix to adjust the current weights to increase the weights of important state features and reduce the weights of unimportant state features, thereby increasing the weights of important state features in driving style recognition, thereby increasing the correlation between state features and driving style, thereby improving the accuracy of driving style recognition, thereby solving the technical problem of low accuracy of vehicle driving style recognition in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0017] Figure 1 is a flow chart of a method for identifying a vehicle driving style according to an embodiment of the present invention;

[0018] Figure 2 is a network structure diagram of a vehicle driving style recognition according to an embodiment of the present invention;

[0019] Figure 3 is a schematic diagram of a first neural network in a vehicle driving style recognition network according to an embodiment of the present invention;

[0020] Figure 4 A schematic diagram of a driving style recognition device for a vehicle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] 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.

[0022] 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.

[0023] According to an embodiment of the present invention, a method for identifying a driving style of a vehicle is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0024] Figure 1 is a flow chart of a method for identifying a vehicle driving style according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0025] Step S102, during the vehicle driving process, in response to receiving a driving style recognition task of the vehicle, obtaining multiple state characteristics of the vehicle and current weights of the multiple state characteristics, wherein the current weights are used to represent the weights of the multiple state characteristics in the driving style recognition task.

[0026] The above-mentioned driving style may refer to the driver's behavior and habits when driving a vehicle. The types of driving style may include but are not limited to aggressive driving, conservative driving, economical driving, nervous driving and fatigue driving. During the driving process, the appropriate driving style can be selected according to road conditions, traffic conditions and personal abilities. Understanding the driver's driving style helps to better evaluate the driver's driving habits and select an appropriate driving plan, thereby improving road safety and driving experience.

[0027] The above-mentioned driving style recognition task may refer to a task of acquiring and analyzing multiple state characteristics of the driver during driving, and then identifying the vehicle's driving style. Among them, the specific recognition method may include driving style recognition through machine learning methods, big data analysis methods, and deep learning methods, etc. The role of the driving style recognition task may include but is not limited to improving the driver's driving experience and promoting the intelligent development of the transportation system.

[0028] The above-mentioned driving style recognition tasks may include but are not limited to the following types:

[0029] The driving style recognition task can be used to identify driving style recognition tasks on different types of roads. For different types of roads, the expressiveness of state characteristics in the driving style recognition task is different. For example, in a downhill section, the vehicle braking characteristics are more expressive, so the driving style recognition task at this time should assign a higher weight to the vehicle braking characteristics.

[0030] The driving style recognition task can be a driving style recognition task for identifying different time periods. For roads in different time periods, the expressiveness of state features in the driving style recognition task is different. For example, for the same section of road, during the working hours, the expressiveness of acceleration features and speed features is stronger, so the driving style recognition task at this time should assign higher weights to acceleration features and speed features, while during the off-duty time period, the gear switching features have stronger expressiveness, so the driving style recognition task at this time should assign higher weights to the gear switching features.

[0031] The driving style recognition task can be used to recognize driving styles in different weather conditions. The expressiveness of state characteristics is different for different weather conditions. For example, on rainy days, the vehicle braking characteristics are more expressive. Therefore, the driving style recognition task at this time should assign a higher weight to the vehicle braking characteristics.

[0032] The driving style recognition task may be a task for identifying the driving styles of different drivers in the above situations. According to the different driving habits of different drivers, the state features with a high correlation with their driving styles may be determined, and then the driving style recognition task in the above situations may be performed.

[0033] The above driving style recognition task types are only examples. The specific driving style recognition tasks need to be determined according to actual conditions and are not limited here.

[0034] The above-mentioned state characteristics may refer to characteristics used to judge or identify the vehicle driving style, and may include but are not limited to brake pedal opening signal, accelerator pedal opening signal, vehicle longitudinal acceleration signal, vehicle speed signal, steering wheel steering angle signal and braking frequency signal, etc. Each state characteristic has its specific function. The selection of specific state characteristics can be determined according to the vehicle type and selection purpose, and is not limited here.

[0035] The above-mentioned current weight may refer to the relative importance or influence of any state feature among multiple state features of the vehicle in the driving style recognition task. The current weight of multiple state features may be an indicator for quantifying the importance of the state features. The role of the current weight may include but is not limited to optimizing decisions, balancing feature influences, etc. The current weight may be obtained by including but not limited to setting by professional experience, obtaining by big data statistical analysis, user-defined setting, and calculating by existing models, etc. The specific acquisition method may be determined according to actual conditions and is not limited here.

[0036] In an optional embodiment, some driving styles of the driver when currently driving the vehicle can be reflected through multiple state features, and by obtaining the current weights of different state features, the state features with a relatively high correlation with the driving style can be determined, so as to enhance the expressiveness of these state features in the process of driving style confirmation, and by obtaining the current weights of different state features, the state features with a relatively low correlation with the driving style can also be determined, so as to reduce the expressiveness of these state features in the process of driving style confirmation, thereby achieving the purpose of enhancing important state features and reducing unimportant state features in the process of driving style confirmation.

[0037] In another optional embodiment, in order to identify the driving style of a vehicle, the state characteristics of the vehicle and the current weights of the state characteristics are obtained. The specific steps are as follows: first, the driving style recognition task needs to be defined, and the recognition objectives and requirements need to be clarified; second, multiple state characteristics of the vehicle need to be obtained, which can reflect the driving state of the vehicle and the driver's operating habits; then, the current weight of each state characteristic needs to be obtained, that is, the contribution of multiple state parameters in the vehicle style recognition task in the current state is obtained, and this is used as the current weight. The multiple state parameters obtained in the above process and their current weights are used as the basis for measuring the correlation between the multiple state characteristics and the driving style recognition task. The actual correlation of the multiple state characteristics to the driving style recognition task will be determined based on this in the subsequent process, and the weights will be adjusted based on the actual correlation.

[0038] Step S104, using the first neural network to identify multiple state features to obtain a weight matrix of the multiple state features, wherein the weight matrix is ​​used to represent the weights of the multiple state features in the driving style recognition task through different preset values.

[0039] The type of the first neural network may include, but is not limited to, a convolutional neural network, a recurrent neural network, a long short-term memory network, and a deep belief network. The specific neural network may be selected according to the network structure and the network function, which is not limited here. The first neural network may improve the efficiency of identifying multiple state features and improve the accuracy of obtaining weight matrices corresponding to multiple state features. The first neural network may determine the weights of multiple state features by expressing the multiple state features as vectors.

[0040] The above-mentioned weight matrix may refer to a matrix that describes the relationship between multiple state features and the weights of the state features through different preset values ​​in the driving style recognition task. The above-mentioned weight matrix may clearly enhance or reduce the current weights of multiple state features, control the degree of adjustment of the current weights, and have a constraining effect.

[0041] The above different preset values ​​may be preset values ​​of a combination of 0 and 1, but are not limited thereto. The above preset values ​​may also be flexibly set according to actual needs, so as to more conveniently construct the above weight matrix.

[0042] In an optional embodiment, the first neural network generates a weight matrix by identifying multiple state features, and the weight matrix includes the weights of different state features in the driving style recognition task. These weights can help the system more accurately evaluate the correlation between multiple state features and driving style, and then enhance the weights of state features with high correlation with driving style, thereby improving the accuracy of driving style recognition. At the same time, by reasonably setting the preset values ​​in the weight matrix, the system can automatically adjust the weights of state features according to the driving habits and behavioral characteristics of different drivers to achieve personalized driving style recognition.

[0043] In another optional embodiment, multiple state features may be preprocessed to obtain multiple state features after preprocessing, and the first neural network may be used to identify the multiple state features after preprocessing to obtain a weight matrix of the multiple state features after preprocessing. Preprocessing may refer to removing noise, data screening, data conversion, and data dimension reduction of multiple state features. The preprocessing methods may include but are not limited to oversampling, undersampling, standardization, normalization, and principal component analysis. The selection and implementation of the preprocessing method should be customized according to the specific problem, data type, and model used, and are not limited here. By preprocessing multiple state features, redundant data is removed, data quality is improved, and the recognition efficiency of the first neural network for multiple state features is improved, thereby reducing the duration of the driving style recognition task. In an optional embodiment, the first neural network can be pre-trained using an existing data set. During the training process, the first neural network will continuously adjust its weights and bias values ​​to reduce prediction errors until a suitable parameter setting is found and applied to the first neural network. After the training is completed, the trained first neural network is used to identify multiple state features, thereby improving the recognition effect of the first neural network on multiple state features, and then a more accurate weight matrix describing the correlation between multiple state features and driving style can be obtained. The above training process can not only enable the first neural network to master the inherent laws of the data, but also improve the generalization ability of the first neural network, thereby improving the performance of the first neural network in the driving style recognition task.

[0044] In an optional embodiment, to obtain a weight matrix of multiple state features, first, the multiple state features need to be preprocessed, including but not limited to noise removal, data standardization, etc.; then, a first neural network needs to be constructed to learn the relationship between the state features and the driving style; then, the first neural network is trained using the preprocessed multiple state features. During the training process, the network will continuously adjust the weights and biases to minimize the prediction error. The purpose of the training is to enable the network to accurately identify various state features; finally, the weight matrix of the features is extracted through the trained first neural network. Each element in the weight matrix represents the relative importance of the corresponding state feature in the driving style recognition task. This step is a direct reflection of the feature importance assessment.

[0045] Step S106, adjusting the current weights of the multiple state features based on the weight matrix to obtain target weights of the multiple state features.

[0046] The above-mentioned target weight may refer to the relative importance of each state feature in the optimization problem of multiple state features. The target weight is usually between 0 and 1, and the sum of all target weights is 1. The larger the weight, the higher the importance of the state feature in the optimization process. The target weight helps decision makers find a balance between multiple state features and improve the quality and efficiency of problem solving.

[0047] In an optional embodiment, current weights of multiple state features are adjusted based on a weight matrix, that is, weights corresponding to state features with a higher correlation with the driving style recognition task are increased according to the weight matrix, and weights corresponding to state features with a lower correlation with the driving style recognition task are decreased, so as to obtain the adjusted weight of each state feature, that is, to obtain target weights of multiple state features, which can more accurately reflect the importance of each state feature in the vehicle driving style recognition task, and state features with a higher correlation with the driving style recognition task obtain higher target weights after adjustment, thereby improving the correlation between the overall state features and the vehicle driving style recognition task, and thus improving the accuracy of vehicle driving style recognition.

[0048] Step S108: performing driving style recognition on multiple state features based on the target weight to obtain a target driving style.

[0049] The above-mentioned target driving style may refer to the driving style of the driver identified by driving style recognition, which reflects the driver's actual driving style, and may include but is not limited to aggressive driving, conservative driving, economical driving, nervous driving and fatigue driving, etc. The specific target driving style needs to be determined based on the results of driving style recognition, and is not limited here.

[0050] The above-mentioned target driving style may refer to the target driving style of the vehicle on different types of roads. For different types of roads, the target driving style of the vehicle is different. For example, on a downhill section, the target driving style will be conservative. The target driving style may refer to the target driving style of the vehicle in different time periods. For different time periods, the target driving style of the vehicle is different. For example, for the same section of road, during the working hours, the target driving style will be aggressive, while during the off-get off work hours, the target driving style will be conservative. The target driving style may refer to the target driving style of the vehicle in different weather conditions. For different weather conditions, the target driving style of the vehicle is different. For example, on rainy days, the target driving style will be nervous or conservative. The target driving style may refer to the target driving style of different drivers. In the above case, the target driving styles of different drivers are also different, and the target driving style needs to be further determined based on the state characteristics such as the driver's driving habits.

[0051] The above target driving style types are only examples. The specific target driving style needs to be determined according to actual conditions and is not limited here.

[0052] In an optional embodiment, on the basis of obtaining the target weights of multiple state features, the multiple state features are weightedly combined with their corresponding target weights, and then the multiple state features combined with the target weights are used for driving style recognition to obtain the target driving style. In the above process, the multiple state features are weightedly combined with the target weights, that is, the corresponding target weights are matched according to the expressiveness of the multiple state features in driving style recognition, the state features with strong expressiveness are matched with higher target weights, and the state features with weak expressiveness are matched with lower target weights, thereby increasing the weight of the state features with strong expressiveness in the multiple state features, and at the same time increasing the expressiveness of the multiple state features in driving style recognition, thereby effectively improving the accuracy of driving style recognition, and better evaluating the driver's driving behavior, giving targeted guidance or improvement suggestions, and thus optimizing the driver's driving behavior.

[0053] In an embodiment of the present invention, first, after receiving a driving style recognition task for a vehicle, multiple state features of the vehicle and current weights of the multiple state features are obtained; then, the multiple state features are recognized using a first neural network to obtain a weight matrix of the multiple state features; secondly, the current matrix of the state features is adjusted using the obtained weight matrix to obtain a target matrix of the multiple state features; finally, the driving style of the multiple state features is recognized based on the target weights to obtain a target driving style of the vehicle. It is easy to notice that the present application obtains the current weights of multiple state features, and uses the first neural network to obtain the weight matrix of multiple state features, and then uses the weight matrix to adjust the current weights to increase the weights of important state features and reduce the weights of unimportant state features, thereby increasing the weights of important state features in driving style recognition, thereby increasing the correlation between state features and driving style, thereby improving the accuracy of driving style recognition, thereby solving the technical problem of low accuracy of vehicle driving style recognition in related technologies.

[0054] Optionally, the first neural network includes a fully connected layer and an activation layer, the fully connected layer includes: a first fully connected layer and a second fully connected layer, the activation layer includes: a first activation layer and a second activation layer, and the first neural network is used to identify multiple state features to obtain a weight matrix of the multiple state features, including: using the first fully connected layer to map the multiple state features to a first dimension to obtain a first vector of the first dimension, wherein the first dimension is smaller than the second dimension of the multiple state features; using the first activation layer to activate the first vector to obtain a second vector, wherein the second vector is used to represent the weight of the first vector in a nonlinear manner; using the second fully connected layer to map the second vector to the second dimension to obtain a third vector; using the second activation layer to activate the third vector to obtain a weight matrix.

[0055] The above-mentioned fully connected layer may refer to a linear layer in a neural network, which can convert input data into a higher-level feature representation. Its functions include but are not limited to feature conversion, information integration, classification, and regression.

[0056] The above-mentioned first fully connected layer may refer to the first fully connected layer in the neural network, and the above-mentioned second fully connected layer may refer to the second fully connected layer in the neural network. The functions of the first fully connected layer and the second fully connected layer may include but are not limited to feature extraction, feature conversion and feature classification, etc. The specific functions can be determined according to the needs of the neural network and are not limited here. In addition, the number of fully connected layers may include but are not limited to one layer, two layers and three layers, etc. The specific number of fully connected layers can be determined according to the needs and structure of the neural network and is not limited here.

[0057] The first dimension mentioned above may refer to a lower dimension to which the first fully connected layer maps the state features, which is smaller than the second dimension of the original state features, and helps to reduce computational complexity.

[0058] The above-mentioned first vector may refer to the output of the first fully connected layer, which is the representation of the state feature in the first dimension.

[0059] The above-mentioned activation layer may refer to the key part responsible for introducing nonlinearity in the neural network. The activation layer is usually composed of activation functions, which can convert the input signal into a nonlinear output, so that the neural network can fit complex function mappings. The types of activation functions may include but are not limited to ReLU activation function, Sigmoid activation function, etc., wherein the ReLU activation function is a rectified linear unit activation function, and the Sigmoid activation function is an S-curve activation function. The type of activation function in the activation layer can be selected according to the role of the activation function, which is not limited here.

[0060] The above-mentioned first activation layer may refer to the first activation layer in the neural network, and the above-mentioned second activation layer may refer to the second activation layer in the neural network. The functions of the first activation layer and the second activation layer may include but are not limited to activating features, extracting features, and converting linear inputs into nonlinear outputs. In addition, the number of activation layers may include but is not limited to one layer, two layers, and three layers. The specific number of activation layers can be determined according to the needs and structure of the neural network, and is not limited here.

[0061] The second vector may refer to the output of the first activation layer after the first activation layer performs activation processing on the first vector, and is used to represent the weight of the first vector and is represented in a nonlinear manner.

[0062] The above-mentioned second dimension may refer to the dimension of the original state characteristics, which is usually higher than the first dimension.

[0063] The third vector may refer to the output of the second fully connected layer, which is the representation of the second vector in the second dimension.

[0064] In an optional embodiment, the first neural network includes two fully connected layers and two activation layers, namely, a first fully connected layer, a second fully connected layer and a first activation layer, a second activation layer. The first neural network is used to identify multiple state features, and the process of obtaining a weight matrix of multiple state features is as follows: first, the multiple state features are input into the first neural network; then, the first fully connected layer in the first neural network maps the multiple state features to the first dimension to obtain a first vector in the first dimension, and then the first activation layer activates the first vector to obtain a second vector; then, the second fully connected layer maps the second vector to the second dimension to obtain a third vector, and then the second activation layer activates the third vector to obtain a weight matrix, and the weight matrix calculation formula is as follows:

[0065] A=F ex (Z) = σ2(W2σ1(W1Z));

[0066] Where A represents the weight matrix, and A∈R f , Z represents multiple state features, F ex (Z) represents the function for activating Z, σ1 represents the activation function in the first activation layer, σ2 represents the activation function in the second activation layer, represents the learning parameters in the first fully connected layer, represents the learning parameters in the second fully connected layer, r represents the reduction rate, and f represents the dimension.

[0067] In the above process, the computational complexity of the first neural network is reduced by reducing the dimensions of multiple feature states, and the correlation between the multiple state features and the driving style recognition task is learned by obtaining the weight matrix corresponding to the multiple state features. This facilitates the subsequent adjustment of the current weights of the multiple state features according to the weight matrix, thereby enhancing the weights of important state features in the driving style recognition task, and thus improving the driving style recognition accuracy.

[0068] Optionally, the third vector is activated using the second activation layer to obtain a weight matrix, including: using the second activation layer to assign multiple output values ​​corresponding to the third vector based on the first preset value and the second preset value to obtain the weight matrix.

[0069] The above-mentioned first preset value and second preset value may refer to pre-set parameters for controlling the generation process of the weight matrix. Here, the first preset value may refer to preset value 0, and the second preset value may refer to preset value 1. The specific value may be adjusted according to actual conditions and is not limited here.

[0070] In an optional embodiment, the third vector is activated by the second activation layer to obtain a weight matrix, including the following steps: first, a first preset value and a second preset value need to be set, and used to control the generation process of the weight matrix; then the third vector is activated by the second activation layer based on the first preset value and the second preset value to obtain the weight matrix. The above process can further enhance the expressive power and flexibility of the network, so that the neural network can better learn and simulate the relationship between different features. In addition, according to the characteristics of the state features, appropriate activation functions, preset values ​​and other parameters can be selected to enable the neural network to achieve better performance.

[0071] Optionally, a second activation layer is used to assign multiple output values ​​corresponding to the third vector based on a first preset value and a second preset value to obtain a weight matrix, including: assigning a first output value among the multiple output values ​​based on the first preset value to obtain a first assignment result, wherein the first output value is an output value of the multiple output values ​​less than or equal to the first preset value; assigning a second output value among the multiple output values ​​based on the second preset value to obtain a second assignment result, wherein the second output value is an output value of the multiple output values ​​greater than the first preset value; and determining the weight matrix based on the first assignment result and the second assignment result.

[0072] The above-mentioned output value may refer to multiple output values ​​of the third vector, which may be divided into a first output value and a second output value, wherein the first output value is an output value in which multiple output values ​​are less than or equal to a first preset value, and the second output value is an output value in which multiple output values ​​are greater than the first preset value. By classifying multiple output values, it is convenient to subsequently assign values ​​based on different output values.

[0073] The first assignment result may refer to an assignment result obtained after the first preset value is assigned to the first output value.

[0074] The second assignment result may refer to an assignment result obtained after the second preset value is assigned to the second output value.

[0075] In an optional embodiment, first, after setting the first preset value and the second preset value, use them as control conditions for generating a weight matrix; then use the second activation layer to assign multiple output values ​​of the third vector based on the above control conditions, wherein the multiple output values ​​include the first output value and the second output value, and perform different assignment operations on the first output value and the second output value, including assigning the first output value based on the first preset value to obtain a first assignment result, and assigning the second output value based on the second preset value to obtain a second assignment result; finally, determine the weight matrix based on the first assignment result and the second assignment result. The above process dynamically adjusts the weights so that the first neural network can better capture the patterns and trends in the data, and then obtains the weight matrix, which can further adjust the impact of the input features on the model output, thereby improving the prediction accuracy of the neural network.

[0076] Optionally, using a first fully connected layer to map multiple state features to a first dimension to obtain a first vector of the first dimension, including: compressing the multiple state features to obtain multiple compressed state features of a third dimension, wherein the third dimension is smaller than the second dimension and the third dimension is larger than the first dimension; using the first fully connected layer to map the multiple compressed state features to the first dimension to obtain a first vector.

[0077] The above-mentioned compression processing may refer to reducing the feature dimensions of the input multiple state features through compression processing. The compression processing can reduce the feature dimensions while retaining key information, thereby reducing the amount of calculation and the number of parameters. The types of compression processing may include but are not limited to global average pooling operations, dimensionality reduction, feature selection, and compressed autoencoders. The specific compression processing operations can be determined according to the data type and training purpose, and are not limited here. The features after compression processing are compressed state features, and the compressed state feature dimensions at this time are smaller than the initial state feature dimensions.

[0078] The third dimension may refer to the dimension of the compressed state feature after compressing multiple state features. This dimension is smaller than the second dimension of the original state feature and larger than the first dimension, which helps to reduce the computational complexity and facilitates the subsequent mapping of the first fully connected layer.

[0079] In an optional embodiment, a global average pooling operation may be used to compress multiple state features to obtain multiple compressed state features of the third dimension. The calculation formula of the global average pooling operation is as follows:

[0080]

[0081] Among them, z i represents the compression state feature of the i-th sample in the third dimension, represents the tth feature of the ith sample, e i represents the original state characteristics of the i-th sample, F sq (e i ) indicates e i The function for performing global average pooling operation, k represents the total number of features, It means to divide the result of summing the features by the total number of features to find the average.

[0082] The compressed state features after compression processing not only retain the key feature information, but also reduce the feature dimension, which is convenient for calculation in subsequent operations, thereby reducing the network calculation amount and running time, and improving the driving style recognition efficiency of the network.

[0083] Optionally, performing driving style recognition on multiple state features based on target weights to obtain a target driving style includes: enhancing the multiple state features based on the target weights to obtain a plurality of enhanced state features; performing driving style recognition on the multiple enhanced state features using a second neural network to obtain a target driving style, wherein an input layer of the second neural network is connected to an output layer of the first neural network, and the output layer of the first neural network is used to output the target weights of the multiple state features.

[0084] The enhanced state feature mentioned above may refer to multiple enhanced state features obtained by enhancing the weight distribution of multiple state features based on the target weight.

[0085] The above-mentioned second neural network may refer to a neural network used for driving style recognition tasks. The functions of the second neural network may include but are not limited to feature classification, feature recognition, etc. The type of the second neural network may include but is not limited to convolutional neural networks, recurrent neural networks, long short-term memory networks, and deep belief networks, etc. The selection of a specific neural network can be selected based on the network structure and network function, and is not limited here.

[0086] In an optional embodiment, after processing the multiple state features through the first neural network, the target weights of the multiple state features are output, and then the weights of the multiple state features are adjusted based on the target weights, including enhancing the multiple state features based on the target weights to obtain multiple enhanced state features, and the calculation formula of the above multiple enhanced state features is as follows:

[0087] v=F ReWeight (A, E) = [a1·e1,…,a f ·e f ]=[v1,…v f ];

[0088] Among them, v represents the enhanced state feature, A represents the target weight, E represents multiple state features, and F ReWeight(A, E) represents the function of adjusting the weights of multiple state features based on the target weight, a1,…a f represents the target weights from 1 to f dimensions, e1,…e f Represents 1 to f-dimensional state features, v1,…v f Represents 1 to f-dimensional enhanced state features.

[0089] Then, the second neural network is used to identify the driving style of multiple enhanced state features to obtain the target driving style. In the above process, multiple state features are enhanced by using target weights to obtain enhanced state features, which can enable the second neural network to more accurately identify the driving style, reduce misjudgment, and improve the accuracy of driving style recognition.

[0090] In an optional embodiment, Figure 2 is a network structure diagram of a vehicle driving style recognition according to an embodiment of the present invention, comprising Figure 2 It can be seen that the vehicle driving style recognition network includes two parts. The first part can be a shallow neural network, that is, the first neural network mentioned above, and the second part can be a deep neural network, that is, the second neural network mentioned above. The first neural network includes a first fully connected layer, a first activation layer, a second fully connected layer and a second activation layer, and the second neural network includes multiple fully connected layers. Multiple state features can be input into the first neural network, and target weights are assigned to multiple state features through the first neural network to increase the weight of important state features in the driving style recognition task, thereby improving the recognition rate of the vehicle driving style; the target weight and multiple state features are input into the second neural network, and the second neural network is used to perform driving style recognition on multiple state features based on the target weight, and finally output the vehicle's driving style. Through the above process, the weight of important state features in the driving style recognition process can be increased, the weight of unimportant state features can be reduced, the weight of important state features in driving style recognition can be increased, and the correlation between state features and driving style can be increased, thereby improving the accuracy of driving style recognition.

[0091] The weight calculation formula of each fully connected layer in the above vehicle driving style recognition network is as follows:

[0092] a (l) =σ(W l a (l-1) +b (l) );

[0093] Among them, l is the model depth, σ is the activation function, and W (l) is the weight of the lth layer, b (l) The bias of the lth layer, a (l)is the output of the lth layer. The above vehicle driving style recognition network consists of multiple fully connected layers, which can further capture the interaction between state features.

[0094] In an optional embodiment, Figure 3 is a schematic diagram of a first neural network in a vehicle driving style recognition network according to an embodiment of the present invention. It can be seen that Figure 3 It mainly includes multiple state features e1, e2, e3, ..., e f , multiple enhanced state features v1, v2, v3, …, v f And three function operations, namely F sq (), F ex () and F ReWeight (). First, multiple state features e1, e2, e3, ..., e f By F sq () function performs a global average pooling operation to obtain multiple compressed state features of size 1×f, which are then ex () function performs activation processing to obtain a weight matrix of size 1×f, and adjusts the current weight based on the weight matrix to obtain the target weight. Finally, F ReWeight The () function adjusts the weights of multiple state features based on the target weight to obtain multiple enhanced state features. Through the above process, the weights of important state features in the driving recognition task are increased, and the weights of unimportant state features in the driving recognition task are reduced, thus achieving the rational distribution of weights of multiple state features in the driving style recognition task.

[0095] Optionally, the vehicle driving style recognition method also includes: obtaining multiple sample state features and sample driving styles corresponding to the multiple sample state features; using a first initial neural network to perform driving style recognition on the multiple sample state features to obtain a predicted driving style; constructing a target loss function based on the sample driving style and the predicted driving style; and adjusting model parameters of the first initial neural network based on the target loss function to obtain a first neural network.

[0096] The above-mentioned sample state characteristics may refer to state characteristics of sample data, wherein the source of sample data may include but is not limited to big data aggregation, test collection and public databases, etc. The sample state characteristics may refer to characteristics used to judge or identify vehicle driving style, which may include but are not limited to brake pedal opening signal, accelerator pedal opening signal, vehicle longitudinal acceleration signal, vehicle speed signal, steering wheel steering angle signal and braking frequency signal, etc. Each state characteristic has its specific function, and the selection of specific state characteristics can be determined according to the vehicle type and selection purpose, and is not limited here.

[0097] The above-mentioned sample driving style may refer to the driving style of the driver judged from the sample data, that is, the driver's behavior and habits when driving the vehicle judged from the sample data. The types of sample driving styles may include but are not limited to aggressive driving, conservative driving, economical driving, nervous driving and fatigue driving, etc.

[0098] The first initial neural network may refer to a neural network model for identifying driving style, which is in an initial state before the training phase and has not been trained or adjusted with any data.

[0099] The predicted driving style may refer to a driving style prediction result output by the first initial neural network after training based on input sample state features.

[0100] The above-mentioned target loss function may refer to a function that measures the difference between the model prediction result and the actual sample driving style. The smaller the value of the loss function, the closer the model prediction result is to the actual driving style. The types of target loss functions may include but are not limited to cross entropy loss function, logarithmic loss function and center loss function. The specific target loss function can be determined comprehensively based on factors such as data type and network structure, and is not limited here. Selecting an appropriate loss function can help the model better learn data features and improve prediction accuracy.

[0101] In an optional embodiment, in order to enable the first neural network to select a suitable target loss function, a first initial neural network may be constructed to obtain the target loss function: first, a certain number of driving samples are collected, including different driving styles, driving speeds, accelerations, steering angles and other information of the drivers, and then the collected driving samples are analyzed to extract the state characteristics of each sample, and the sample driving styles corresponding to the state characteristics of each sample are obtained; then, a first initial neural network is constructed to identify the driving styles of the state characteristics of multiple samples to obtain a predicted driving style; then, a target loss function is constructed according to the driving styles of the actual driving samples and the driving styles predicted by the neural network to evaluate the prediction performance of the model. The calculation formula of the target loss function may be:

[0102]

[0103] Among them, loss represents the target loss function, i represents the i-th sample, N represents the total number of samples, and y i is the true sample label of the i-th sample, For i The corresponding predicted sample label, log represents the natural logarithm function.

[0104] Finally, based on the target loss function, the model parameters of the first initial neural network are adjusted and optimized to minimize the value of the loss function, and applied to the first initial neural network to obtain the first neural network. The above process trains a large amount of driving data by constructing the first initial neural network, and continuously adjusts the parameters to obtain the target loss function, which is then used in the first neural network to ensure that the first neural network has good performance, so that the subsequent first neural network can more accurately obtain the target weights of multiple state features.

[0105] According to another aspect of an embodiment of the present invention, a vehicle driving style recognition device is also provided. The device can execute the vehicle driving style recognition method of the above embodiment. The specific implementation method and preferred application scenario are the same as those of the above embodiment and will not be repeated here.

[0106] Figure 4 is a schematic diagram of a driving style recognition device for a vehicle according to an embodiment of the present invention, such as Figure 4 As shown, the device includes the following: an acquisition module 402 , a first identification module 404 , an adjustment module 406 , and a second identification module 408 .

[0107] an acquisition module, configured to acquire, in response to receiving a driving style recognition task of the vehicle during the driving process of the vehicle, a plurality of state features of the vehicle and current weights of the plurality of state features, wherein the current weights are used to represent the weights of the plurality of state features in the driving style recognition task;

[0108] A first recognition module, used to recognize multiple state features using a first neural network to obtain a weight matrix of the multiple state features, wherein the weight matrix is ​​used to represent the weights of the multiple state features in the driving style recognition task through different preset values;

[0109] An adjustment module, used for adjusting current weights of multiple state features based on a weight matrix to obtain target weights of the multiple state features;

[0110] The second recognition module is used to perform driving style recognition on multiple state features based on the target weight to obtain a target driving style.

[0111] Optionally, the first recognition module also includes: a first neural network, the first neural network includes a fully connected layer and an activation layer, the fully connected layer includes: a first fully connected layer and a second fully connected layer, the activation layer includes: a first activation layer and a second activation layer, and the first neural network is used to identify multiple state features to obtain a weight matrix of multiple state features, including: a first mapping unit, using the first fully connected layer to map the multiple state features to a first dimension to obtain a first vector of the first dimension, wherein the first dimension is smaller than the second dimension of the multiple state features; a first processing unit, using the first activation layer to activate the first vector to obtain a second vector, wherein the second vector is used to represent the weight of the first vector in a nonlinear manner; a second mapping unit, using the second fully connected layer to map the second vector to the second dimension to obtain a third vector; a second processing unit, using the second activation layer to activate the third vector to obtain a weight matrix.

[0112] Optionally, the second processing unit includes: an assignment subunit, which uses the second activation layer to assign multiple output values ​​corresponding to the third vector based on the first preset value and the second preset value to obtain a weight matrix.

[0113] Optionally, the assignment subunit includes: assigning a first output value among multiple output values ​​based on a first preset value to obtain a first assignment result, wherein the first output value is an output value of the multiple output values ​​less than or equal to the first preset value; assigning a second output value among the multiple output values ​​based on a second preset value to obtain a second assignment result, wherein the second output value is an output value of the multiple output values ​​greater than the second preset value; determining a weight matrix based on the first assignment result and the second assignment result.

[0114] Optionally, the first mapping unit includes: compressing multiple state features to obtain multiple compressed state features of a third dimension, wherein the third dimension is smaller than the second dimension and the third dimension is larger than the first dimension; using a first fully connected layer to map the multiple compressed state features to the first dimension to obtain a first vector.

[0115] Optionally, the second recognition module includes: a first enhancement unit, which enhances multiple state features based on target weights to obtain multiple enhanced state features; a first recognition unit, which uses a second neural network to perform driving style recognition on the multiple enhanced state features to obtain a target driving style, wherein the input layer of the second neural network is connected to the output layer of the first neural network, and the output layer of the first neural network is used to output the target weights of the multiple state features.

[0116] Optionally, the above-mentioned device also includes: a second acquisition module, which acquires multiple sample state features and sample driving styles corresponding to the multiple sample state features; a third identification module, which uses the first initial neural network to identify the driving style of the multiple sample state features to obtain a predicted driving style; a construction module, which constructs a target loss function based on the sample driving style and the predicted driving style; and a second adjustment module, which adjusts the model parameters of the first initial neural network based on the target loss function to obtain the first neural network.

[0117] According to another aspect of an embodiment of the present invention, an electronic device is also provided, comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by one or more processors, the one or more processors execute the above-mentioned vehicle driving style recognition method.

[0118] The storage device in the above steps can be a kind of sequential logic circuit, a memory component used to store data and instructions, etc., mainly used to store programs and data; the processor can be a functional unit that interprets and executes instructions, and it has a unique set of operating commands, which can be called the instruction set of the processor, such as storage, loading, etc. are all operations; the storage device stores computer programs, which can be a set of instructions that can be recognized and executed by a computer, running on an electronic computer, and is an information tool that meets certain needs of people.

[0119] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is further provided, the computer-readable storage medium including a stored program, wherein when the program is run, the processor of the device where the program is located is controlled to execute the above-mentioned vehicle driving style recognition method.

[0120] The computer storage medium in the above steps can be a medium used to store certain discontinuous physical quantities in a computer memory, and the computer storage medium mainly includes semiconductors, magnetic cores, magnetic drums, magnetic tapes, laser disks, etc. The stored program included in the computer-readable storage medium can be a set of instructions that can be recognized and executed by a computer, running on an electronic computer, and is an information tool that meets certain needs of people.

[0121] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0122] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, 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 units or modules, which can be electrical or other forms.

[0123] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0124] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0125] 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 computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0126] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for identifying a vehicle's driving style, characterized in that: include: During the driving process of the vehicle, in response to receiving a driving style recognition task of the vehicle, obtaining a plurality of state features of the vehicle and current weights of the plurality of state features, wherein the current weights are used to represent the weights of the plurality of state features in the driving style recognition task; Using a first neural network to identify the multiple state features, obtaining a weight matrix of the multiple state features, wherein the weight matrix is ​​used to represent the weights of the multiple state features in the driving style recognition task through different preset values; Adjusting current weights of the plurality of state features based on the weight matrix to obtain target weights of the plurality of state features; Driving style recognition is performed on the multiple state features based on the target weight to obtain a target driving style.

2. The method for identifying a vehicle driving style according to claim 1, characterized in that: The first neural network includes a fully connected layer and an activation layer, the fully connected layer includes: a first fully connected layer and a second fully connected layer, the activation layer includes: a first activation layer and a second activation layer, and the first neural network is used to identify the multiple state features to obtain a weight matrix of the multiple state features, including: Mapping the multiple state features to a first dimension using a first fully connected layer to obtain a first vector of the first dimension, wherein the first dimension is smaller than a second dimension of the multiple state features; Activating the first vector using the first activation layer to obtain a second vector, wherein the second vector is used to represent the weight of the first vector in a nonlinear manner; Mapping the second vector to the second dimension using a second fully connected layer to obtain a third vector; The third vector is activated by using the second activation layer to obtain the weight matrix.

3. The method for identifying the driving style of a vehicle according to claim 2, characterized in that: Activating the third vector using the second activation layer to obtain the weight matrix includes: The second activation layer is used to assign multiple output values ​​corresponding to the third vector based on the first preset value and the second preset value to obtain the weight matrix.

4. The method for identifying the driving style of a vehicle according to claim 3, characterized in that: Assigning a plurality of output values ​​corresponding to the third vector based on the first preset value and the second preset value using the second activation layer to obtain the weight matrix includes: Assigning a first output value among the multiple output values ​​based on the first preset value to obtain a first assignment result, wherein the first output value is an output value of the multiple output values ​​that is less than or equal to the first preset value; Assigning a second output value among the multiple output values ​​based on the second preset value to obtain a second assignment result, wherein the second output value is an output value of the multiple output values ​​that is greater than the first preset value; The weight matrix is ​​determined based on the first assignment result and the second assignment result.

5. The method for identifying the driving style of a vehicle according to claim 2, characterized in that: Mapping the plurality of state features to a first dimension using a first fully connected layer to obtain a first vector of the first dimension includes: Performing compression processing on the multiple state features to obtain multiple compressed state features of a third dimension, wherein the third dimension is smaller than the second dimension and larger than the first dimension; The first fully connected layer is used to map the multiple compressed state features to the first dimension to obtain the first vector.

6. The method for identifying a vehicle driving style according to claim 1, characterized in that: Performing driving style recognition on the multiple state features based on the target weight to obtain a target driving style includes: Enhance the multiple state features based on the target weights to obtain multiple enhanced state features; A second neural network is used to perform driving style recognition on the multiple enhanced state features to obtain a target driving style, wherein an input layer of the second neural network is connected to an output layer of the first neural network, and the output layer of the first neural network is used to output target weights of the multiple state features.

7. The method for identifying a vehicle driving style according to claim 1, characterized in that: The method further comprises: Acquire a plurality of sample state features and sample driving styles corresponding to the plurality of sample state features; Using a first initial neural network to perform driving style recognition on the plurality of sample state features to obtain a predicted driving style; constructing a target loss function based on the sample driving style and the predicted driving style; The model parameters of the first initial neural network are adjusted based on the target loss function to obtain the first neural network.

8. A driving style recognition device for a vehicle, characterized in that: include: an acquisition module, configured to acquire, in response to receiving a driving style recognition task of the vehicle during driving of the vehicle, a plurality of state features of the vehicle and current weights of the plurality of state features, wherein the current weights are used to represent weights of the plurality of state features in the driving style recognition task; a first recognition module, configured to recognize the plurality of state features by using a first neural network, and obtain a weight matrix of the plurality of state features, wherein the weight matrix is ​​used to represent the weights of the plurality of state features in the driving style recognition task through different preset values; An adjustment module, configured to adjust current weights of the plurality of state features based on the weight matrix to obtain target weights of the plurality of state features; The second recognition module is used to perform driving style recognition on the multiple state features based on the target weight to obtain a target driving style.

9. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 7 when running.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 7.