Vehicle driving style self-adaption method based on online learning

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses a vehicle driving style self-adaption method based on online learning, and the method comprises the steps: obtaining the vehicle driving data of a plurality of continuous moments including the current moment in the vehicle driving process; inputting the vehicle driving data into a driving style detection model obtained by online training to obtain a driving style of the driver at the next moment; the vehicle is correspondingly controlled based on the driving style; the online training process of the driving style detection model comprises the steps of predicting a vehicle driving data predicted value at the next moment based on the vehicle driving data, generating an online feedback strategy based on an error between the vehicle driving data and the vehicle driving data predicted value, and performing online feedback training on the driving style detection model based on the online feedback strategy. According to the method, the driving style of the driver can be dynamically predicted, the vehicle is adaptively adjusted based on the predicted driving style, and the purpose of'thousands of vehicles and thousands of people 'is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a vehicle driving style adaptation method based on online learning. Background Art

[0002] Advanced Driver Assistance System is a set of built-in coordination technologies that help drivers improve vehicle driving safety. It helps drivers perceive possible dangers through perception, decision-making and execution. It effectively combines driving and braking, which to a certain extent reduces the driver's driving pressure and greatly reduces the risk of traffic accidents caused by improper driver operation.

[0003] In the related art, the driver's driving style is usually divided into aggressive, ordinary and cautious, etc., and the overall focus is still on "people adapting to the car". Specifically, the related art pre-trains the driving style prediction model by using offline learning, and predicts the driver's driving style through the driving style prediction model, and fixes the predicted driving style as the driver's driving style. For example, if a driver's driving style is predicted to be aggressive, then the driver's driving style is fixed to the aggressive type, and the vehicle is controlled according to the vehicle control method corresponding to the aggressive type. In fact, the driver usually does not always maintain an aggressive driving style, which affects the driver's driving experience. Summary of the invention

[0004] The purpose of the present invention is to dynamically predict the driver's driving style, unlike the related art where the driving style is immutable. Moreover, the present invention can adaptively adjust the vehicle based on the predicted driving style, achieving the goal of "one thousand vehicles, one thousand faces and one thousand people, one thousand faces". At the same time, the driving style detection model can be trained with online feedback, dynamically adjust the model parameters of the driving style detection model, and optimize the accuracy and response speed of the overall system.

[0005] In a first aspect, an embodiment of the present invention provides a vehicle driving style adaptation method based on online learning, the method comprising: During the driving process of the vehicle, obtaining the vehicle driving data at a plurality of consecutive moments including the current moment; Inputting the vehicle driving data into a driving style detection model obtained through online training to obtain the driver's driving style at the next moment; controlling the vehicle accordingly based on the driving style; Among them, the online training process of the driving style detection model includes: predicting a predicted value of vehicle driving data at the next moment based on the vehicle driving data, and generating an online feedback strategy based on an error between the vehicle driving data and the predicted value of the vehicle driving data, and performing online feedback training on the driving style detection model based on the online feedback strategy.

[0006] Optionally, the process of predicting the driver's driving style at the next moment by the driving style detection model includes: Generate vehicle driving time series data based on the vehicle driving data, wherein the vehicle driving time series data is used to characterize time series characteristics and context association information of the vehicle driving data; Extracting key time series data from the vehicle driving time series data and the attention weight of each key time series data based on the attention mechanism; Converting the key time series data into target time series data in a time series format; Based on the target time series data and the attention weights corresponding to each target time series data, the driver's driving style at the next moment is predicted.

[0007] Optionally, the driving style detection model includes a time series encoder, a latent space attention mechanism module including a self-attention mechanism and a cross-attention mechanism, and a time series decoder; The generating of vehicle driving time series data based on the vehicle driving data comprises: Encoding the vehicle driving data into vehicle driving time series data by the time series encoder; Accordingly, the key time series data in the vehicle driving time series data and the attention weight of each key time series data are extracted based on the attention mechanism, including: The vehicle driving time series data is processed by the latent space attention mechanism module to obtain key time series data in the vehicle driving time series data and the attention weight of each key time series data; Correspondingly, the converting of the key time series data into target time series data in a time series format includes: converting the key time series data into target time series data in a time series format by using the time series decoder.

[0008] Optionally, the driving style detection model further includes a pedal and steering prediction module and an online gradient network; The method predicts a predicted value of vehicle driving data at the next moment based on the vehicle driving data, generates an online feedback strategy based on an error between the vehicle driving data and the predicted value of the vehicle driving data, and performs online feedback training on a driving style detection model based on the online feedback strategy to obtain a driving style detection model after online training, including: Based on the target time series data and the attention weights corresponding to each target time series data, predicting the pedal data prediction value and the steering data prediction value at the next moment through the pedal and steering prediction module; Calculating the pedal data and steering data included in the vehicle driving data by the online gradient network, and the errors between the pedal data prediction value and the steering data prediction value, determining an online feedback strategy based on the errors, and feeding back the online feedback strategy to the latent space attention mechanism module; The latent space attention mechanism module is trained online based on the online feedback strategy.

[0009] Optionally, the obtaining of vehicle driving data at a plurality of consecutive moments including the current moment includes: Acquire initial vehicle driving data at a plurality of consecutive moments including the current moment, wherein the initial vehicle driving data includes pedal data, steering data, driving speed, and yaw speed of the vehicle; The initial vehicle driving data is preprocessed to obtain preprocessed vehicle driving data, wherein the preprocessing includes data fusion and data denoising.

[0010] Optionally, the driving style includes multiple types, and the step of inputting the vehicle driving data into a driving style detection model obtained through online training to obtain the driver's driving style at the next moment includes: Inputting the vehicle driving data into a driving style detection model obtained through online training, and outputting multiple types of driving styles and the probability corresponding to each type of driving style through the driving style detection model; Based on the multiple types of driving styles and the probability corresponding to each type of driving style, the driving style of the driver at the next moment is calculated.

[0011] Optionally, the controlling the vehicle accordingly based on the driving style includes: receiving a torque request input by a driver via an accelerator pedal; Obtain torque variation curves corresponding to various types of driving styles; Based on the probabilities corresponding to the various types of driving styles, weighted summation is performed on the torque change curves corresponding to the various types of driving styles to obtain the torque change curve after the weighted summation; The output torque is obtained through the weighted summed torque variation curve, and the vehicle is controlled through the output torque.

[0012] In a second aspect, an embodiment of the present invention provides a vehicle driving style adaptation device based on online learning, the device comprising: The vehicle driving data acquisition module is used to acquire the vehicle driving data of a plurality of consecutive moments including the current moment during the vehicle driving process; A driving style detection module, used for inputting the vehicle driving data into a driving style detection model obtained through online training to obtain the driver's driving style at the next moment; A vehicle control module, configured to control the vehicle accordingly based on the driving style; Among them, the online training process of the driving style detection model includes: predicting a predicted value of vehicle driving data at the next moment based on the vehicle driving data, and generating an online feedback strategy based on an error between the vehicle driving data and the predicted value of the vehicle driving data, and performing online feedback training on the driving style detection model based on the online feedback strategy.

[0013] In a third aspect, an embodiment of the present invention provides an electronic device, including: at least one processor; a memory for storing the at least one processor-executable instruction; The at least one processor is configured to execute the instructions to implement the method described in the first aspect.

[0014] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method described in the first aspect.

[0015] The technical solution provided by the embodiment of the present invention can input the vehicle driving data into the driving style detection model obtained through online training after obtaining the vehicle driving data, obtain the driver's driving style at the next moment, and control the vehicle accordingly based on the driving style, so that the predicted driving style can meet the driver's current driving state, and as the input continuous data stream changes, the predicted driving style will also change, unlike the related art, where the driving style is immutable. Moreover, the present invention can adaptively adjust the vehicle based on the predicted driving style, achieving the purpose of "thousands of vehicles and thousands of people", ensuring that the vehicle's response matches the driver's intention and the driving conditions at the time, and improving the driver's driving experience.

[0016] In addition, the present invention determines the online feedback strategy through the error between the real vehicle driving data and the predicted vehicle driving data, performs online feedback training on the driving style detection model, dynamically adjusts the model parameters of the driving style detection model, and optimizes the accuracy and response speed of the overall system. That is, the output driving style is more accurate, and the vehicle is controlled accordingly based on the driver's driving style, which improves the safety of the vehicle, further ensures that the vehicle's response matches the driver's intention and the driving conditions at the time, and further improves the driver's driving experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of the overall technical solution architecture provided by an embodiment of the present invention; Figure 2 A schematic diagram of a specific implementation method provided for an embodiment of the present invention; Figure 3 A flowchart of a vehicle driving style adaptation method based on online learning provided by an embodiment of the present invention; Figure 4 for Figure 3 A flowchart of an implementation method of S330; Figure 5 A schematic diagram of the structure of a vehicle driving style adaptation device based on online learning provided by an embodiment of the present invention; Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be described in detail below through examples.

[0019] Advanced Driver Assistance System is a set of built-in coordination technologies that help drivers improve vehicle driving safety. It helps drivers perceive possible dangers through perception, decision-making and execution. It effectively combines driving and braking, which to a certain extent reduces the driver's driving pressure and greatly reduces the risk of traffic accidents caused by improper driver operation.

[0020] In the related art, the driving style of drivers is usually divided into aggressive, ordinary and cautious, etc., and the overall focus is still on "people adapting to cars". However, "cars have different faces" and "people have different faces". The purpose of the embodiments of the present invention is to achieve "cars adapting to people" to adapt to various driving styles. Different drivers have different driving styles, and there are many factors that affect the driver's driving behavior. Therefore, driver adaptation technology needs to be further studied to achieve a smooth transition between different vehicle modes to adapt to different driving styles.

[0021] At present, the end-to-end large model application method is often used to improve driver adaptation technology. Through large-scale data training, control instructions can be generated directly from sensor data, avoiding delays and error accumulation in the intermediate links, and improving the overall performance of the system. However, the problem with this method is that it is highly complex to implement and requires splitting the model and coordinating the calculation results, making the model more "black box", reducing the interpretability of the network, and making it difficult to determine the contribution of each component in the model to the final goal, resulting in reduced model flexibility.

[0022] The present invention aims to adaptively adjust the vehicle based on the driver's driving style. Specifically, a driving style detection model is built by introducing the driver's driving pedal, brake pedal, steering and other data during driving. The driving style detection model can also be called a driving style detection model; the driving style of the driver at the next moment is predicted by the driving style detection model, and the vehicle is controlled to switch smoothly in real time between the sports mode, economic mode and comfort mode according to the driving style. The present invention performs online learning through a local feedback network, trains and optimizes the detection model in real time, and updates the parameters of vehicle driving, steering and braking.

[0023] like Figure 1 The figure is a schematic diagram of the overall technical solution architecture provided by an embodiment of the present invention.

[0024] from Figure 1 It can be seen that the embodiment of the present invention uses the vehicle's pedal driving data, pedal braking data, steering data, vehicle speed, vehicle yaw speed and other characteristic information (which may be referred to as vehicle driving data) as input. It should be emphasized that the input vehicle driving data is vehicle driving data of multiple consecutive moments including the current moment, that is, the input vehicle driving data is a continuous data stream.

[0025] Artificial intelligence is used to detect the driver's driving behavior and output a driving mode based on the driving behavior. The driving mode can be called a driving style, which can include cautious and aggressive types, etc. Of course, in actual applications, other types of driving styles can also be included, which are not listed here one by one.

[0026] The dashed box represents the operation performed by the driving style detection model. The driving style prediction model predicts the driving behavior through the above vehicle driving data, and obtains the driving mode based on the driving behavior. The driving style detection model outputs various driving modes and the corresponding probability of each driving mode. Finally, the vehicle control system can control the vehicle accordingly based on the driving mode. The vehicle control system may include the vehicle controller, electric power steering system, electronic hydraulic braking system, intelligent driving system, advanced driver assistance system, etc.

[0027] The technical solution provided by the embodiment of the present invention, when predicting the driver's driving style at the next moment, first obtains the vehicle driving data of multiple consecutive moments including the current moment, that is, obtains a continuous video stream, so that the predicted driving style can meet the driver's current driving state, and as the input continuous data stream changes, the predicted driving style will also change, unlike the related art, where the driving style is immutable. In addition, the present invention can adaptively adjust the vehicle based on the predicted driving style, achieving the purpose of "a thousand faces for a thousand cars and a thousand faces for a thousand people", ensuring that the vehicle's response matches the driver's intention and the driving conditions at the time, and improving the driver's driving experience.

[0028] like Figure 2 The figure is a schematic diagram of a specific implementation method provided by an embodiment of the present invention.

[0029] from Figure 2 It can be seen that the pedal drive data, pedal braking data and steering data of multiple consecutive moments including the current moment are first obtained, and the pedal drive data and pedal braking data of the vehicle are fused through the pedal fusion module to obtain the fused pedal data, and the steering data of the vehicle is filtered through a filter to obtain the filtered steering data.

[0030] The fused pedal data and filtered steering data are input into the time series encoder, which generates vehicle driving time series data. The function of the time series encoder is to capture the changing patterns of pedal data and steering data at multiple consecutive moments over time, and convert the pedal data and steering data at multiple consecutive moments into data that can be received by the latent space attention mechanism.

[0031] The format of vehicle driving time series data is not a simple sequence of numbers, but a high-dimensional feature sequence, for example, it can be a feature vector, embedding vector, attention weight, etc. These representations can capture the time series features and contextual information in the vehicle driving data. The feature vector, embedding vector and attention weight are explained below.

[0032] 1. Feature vector. The time series encoder converts the input continuous data stream into a set of feature vectors that can represent the key features of the original data.

[0033] 2. Embedding vector: Embedding vector is a method to map discrete data into continuous vector space, which can capture the semantics and relationships of discrete data.

[0034] 3. Attention weight: Indicates the attention paid to different time steps when the model processes the sequence.

[0035] After obtaining the vehicle driving time series data through the time series encoder, the vehicle driving time series data can be input into the latent space attention mechanism module including the self-attention mechanism and the cross-attention mechanism. The latent space attention mechanism module will screen out the key time series data in the vehicle driving time series data, as well as the attention weights corresponding to the key moment data at each moment. This allows the model to establish connections between different parts of the time series data sequence, thereby better understanding the context of the time series data.

[0036] The time series decoder converts the key time series data processed by the latent space attention mechanism module back to the time series format in order to predict the driver's driving method at the next moment, that is, to predict the driver's driving style at the next moment.

[0037] from Figure 2 It can be seen that the pedal prediction module outputs the pedal prediction value and steering prediction value of the next moment based on the data output by the latent space attention mechanism module, where the pedal prediction value includes the pedal driving data prediction value, the pedal braking data prediction value and the steering data prediction value, that is, predicting whether the driver will accelerate or decelerate at the next moment, and whether he will turn.

[0038] The core function of the feedback mechanism of the embodiment of the present invention is to optimize the overall performance by predicting the future driver's operating behavior (including the above-mentioned pedal driving data prediction value, pedal braking data prediction value and steering data prediction value) and passing these prediction results to the system's online gradient network.

[0039] The input data of the online gradient network includes the initial pedal driving data, pedal braking data and steering data (data after data fusion and filtering); the input data of the online gradient network also includes the pedal driving data prediction value, pedal braking data prediction value and steering data prediction value output by the pedal prediction module.

[0040] The online gradient network calculates the error between the initial pedal driving data and the pedal driving data prediction value, the error between the initial pedal braking data and the pedal braking data prediction value, and the error between the initial steering data and the steering data prediction value, and determines the online feedback strategy based on these three errors. The online feedback strategy is then passed to the latent space attention mechanism module to dynamically adjust the parameters of the latent space attention mechanism module (such as attention weights) to optimize the accuracy and response speed of the overall system.

[0041] The embodiment of the present invention uses an online gradient network to calculate the error between the initial input real pedal driving data and the predicted pedal driving data, the error between the real pedal braking data and the predicted pedal braking data, and the error between the real steering data and the predicted steering data, and can determine the online feedback strategy, and perform online feedback training on the latent space attention mechanism module, dynamically adjust the parameters of the latent space attention mechanism module (such as attention weight), and optimize the accuracy and response speed of the overall system. That is, the output driving style is more accurate, and the vehicle is controlled accordingly based on the driver's driving style, which improves the safety of the vehicle, ensures that the vehicle's response matches the driver's intention and the driving conditions at the time, and improves the driver's driving experience.

[0042] The following is a detailed description of a vehicle driving style adaptation method based on online learning provided by an embodiment of the present invention. Figure 3 As shown, a vehicle driving style adaptation method based on online learning provided by an embodiment of the present invention may include the following steps, namely: S310, during the vehicle driving process, obtaining vehicle driving data at a plurality of consecutive moments including the current moment.

[0043] Specifically, in order to adaptively predict the driver's driving style at the next moment according to the user's current driving state, vehicle driving data of multiple consecutive moments including the current moment can be obtained during the vehicle driving process. The vehicle driving data may include pedal driving data, pedal braking data, steering data, vehicle speed, and vehicle yaw speed, etc., which are not specifically limited in the embodiments of the present invention.

[0044] As an implementation method of the embodiment of the present invention, obtaining vehicle driving data at multiple consecutive moments including the current moment may include the following steps: Initial vehicle driving data at a plurality of consecutive moments including the current moment is acquired, wherein the initial vehicle driving data includes pedal data, steering data, driving speed, and yaw speed of the vehicle.

[0045] The initial vehicle driving data is preprocessed to obtain preprocessed vehicle driving data, wherein the preprocessing includes data fusion and data denoising.

[0046] In this implementation, after the initial vehicle driving data is acquired, pre-processing operations such as data fusion and noise removal may be performed on the initial vehicle driving data. Thus, after the vehicle driving data is input into the driving style detection model, it helps the driving style detection model to input a driving style with a higher accuracy.

[0047] S320: Input the vehicle driving data into the driving style detection model obtained through online training to obtain the driver's driving style at the next moment.

[0048] S330, controlling the vehicle accordingly based on the driving style.

[0049] Among them, the online training process of the driving style detection model includes: predicting the vehicle driving data prediction value at the next moment based on the vehicle driving data, and generating an online feedback strategy based on the error between the vehicle driving data and the vehicle driving data prediction value, and performing online feedback training on the driving style detection model based on the online feedback strategy.

[0050] Specifically, after obtaining the vehicle driving data, the vehicle driving data can be input into the driving style detection model obtained through online training to obtain the driver's driving style at the next moment, and the vehicle can be controlled accordingly based on the driving style. In this way, the predicted driving style can meet the driver's current driving state, and as the input continuous data stream changes, the predicted driving style will also change, unlike the related art, where the driving style is immutable. Moreover, the present invention can adaptively adjust the vehicle based on the predicted driving style, achieving the purpose of "one thousand faces for one thousand vehicles and one thousand faces for one thousand people", ensuring that the vehicle's response matches the driver's intention and the driving conditions at the time, and improving the driver's driving experience.

[0051] In addition, the present invention determines the online feedback strategy through the error between the real vehicle driving data and the predicted vehicle driving data, performs online feedback training on the driving style detection model, dynamically adjusts the model parameters of the driving style detection model, and optimizes the accuracy and response speed of the overall system. That is, the output driving style is more accurate, and the vehicle is controlled accordingly based on the driver's driving style, which improves the safety of the vehicle, further ensures that the vehicle's response matches the driver's intention and the driving conditions at the time, and further improves the driver's driving experience.

[0052] exist Figure 3 Based on the illustrated embodiment, in one implementation, the process of the driving style detection model predicting the driver's driving style at the next moment may include the following four steps, namely: Step 1: Generate vehicle driving time series data based on vehicle driving data.

[0053] Among them, the vehicle driving time series data is used to characterize the time series characteristics of the vehicle driving data and contextual association information.

[0054] Step 2: Extract key time series data from the vehicle driving time series data and the attention weight of each key time series data based on the attention mechanism.

[0055] Step 3: Convert the key time series data into target time series data in time series format.

[0056] Step 4: predict the driver's driving style at the next moment based on the target time series data and the attention weight corresponding to each target time series data.

[0057] In practical applications, the driving style detection model may include a time series encoder, a latent space attention mechanism module including a self-attention mechanism and a cross-attention mechanism, and a time series decoder.

[0058] At this time, step 1, generating vehicle driving time series data based on vehicle driving data, may include the following steps: Step 11, encoding the vehicle driving data into vehicle driving time series data through a time series encoder; Accordingly, step 2, extracting key time series data from the vehicle driving time series data based on the attention mechanism, and the attention weight of each key time series data, may include the following steps: Step 21, processing the vehicle driving time series data through the latent space attention mechanism module to obtain key time series data in the vehicle driving time series data and the attention weight of each key time series data; Accordingly, step 3, converting the key time series data into target time series data in a time series format, may include the following steps: Step 31, converting the key time series data into target time series data in a time series format through a time series decoder.

[0059] Since steps 1 to 4, as well as step 11, step 21 and step 31 have been described in detail in the above embodiments, they will not be described in detail in the embodiment of the present invention.

[0060] Based on the above embodiment, the driving style detection model further includes a pedal and steering prediction module and an online gradient network.

[0061] At this time, predicting the predicted value of the vehicle driving data at the next moment based on the vehicle driving data, and generating an online feedback strategy based on the error between the vehicle driving data and the predicted value of the vehicle driving data, and performing online feedback training on the driving style detection model based on the online feedback strategy to obtain the driving style detection model after online training may include the following steps: Based on the target time series data and the attention weights corresponding to each target time series data, the pedal data prediction value and the steering data prediction value at the next moment are predicted through the pedal and steering prediction module; The pedal data and steering data included in the vehicle driving data are calculated through an online gradient network, and the errors between the pedal data prediction value and the steering data prediction value are calculated. An online feedback strategy is determined based on the error, and the online feedback strategy is fed back to the latent space attention mechanism module. The latent space attention mechanism module is trained online based on the online feedback strategy.

[0062] The above steps are the online feedback training process of the driving style detection model. Figure 2 This has been explained in the illustrated embodiment and will not be described in detail here.

[0063] Based on the above embodiment, in one implementation, the driving style may include multiple types, and inputting the vehicle driving data into the driving style detection model obtained by online training to obtain the driver's driving style at the next moment may include the following steps: The vehicle driving data is input into the driving style detection model obtained through online training, and the driving style detection model outputs multiple types of driving styles and the corresponding probability of each type of driving style.

[0064] Based on multiple types of driving styles and the probability corresponding to each type of driving style, the driver's driving style at the next moment is calculated.

[0065] Specifically, the driving style detection model may output multiple driving styles, for example, aggressive, economical, and cautious, etc., and the probability corresponding to each driving style may be output, for example, 20% for aggressive, 50% for economical, 30% for cautious, etc. Then, a weighted sum of multiple types of driving styles and corresponding probabilities is performed to obtain the driver's driving style at the next moment.

[0066] As an implementation method of the embodiment of the present invention, S330, the vehicle is controlled accordingly based on the driving style, such as Figure 4 As shown, the following steps may be included: S331, receiving a torque request input by the driver through the accelerator pedal.

[0067] S332, obtaining torque variation curves corresponding to various types of driving styles.

[0068] S333: Based on the probabilities corresponding to the various types of driving styles, weighted summation is performed on the torque variation curves corresponding to the various types of driving styles to obtain the torque variation curve after the weighted summation.

[0069] S334, obtaining the output torque through the weighted summation of the torque variation curve, and controlling the vehicle through the output torque.

[0070] Specifically, after the vehicle control system receives the torque request input by the driver by stepping on the accelerator, it can obtain the torque change curves corresponding to various types of driving styles. It can be understood that different types of driving styles have different torque change curves. The curvature of the torque change curve of the aggressive driving style is larger, and the curvature of the torque change curve of the cautious driving style is smaller.

[0071] Since the driving style detection model outputs multiple types of driving styles and the corresponding probabilities of each type of driving style, in order to obtain the torque change curve of the driving style at the next moment, the torque change curves corresponding to the various types of driving styles can be weighted and summed based on the corresponding probabilities of the various types of driving styles to obtain the weighted summed torque change curve, which is the torque change curve of the driving style at the next moment. Finally, the output torque is obtained through the weighted summed torque change curve, and the vehicle is controlled through the output torque.

[0072] In the embodiment of the present invention, the predicted driving style at the next moment is not fixed, but changes with the input continuous vehicle driving data stream. In addition, the present invention can adaptively adjust the vehicle based on the predicted driving style, achieving the purpose of "one car, one person, one vehicle", ensuring that the vehicle's response matches the driver's intention and the driving conditions at the time, and improving the driver's driving experience.

[0073] In a second aspect, an embodiment of the present invention provides a vehicle driving style adaptive device 50 based on online learning, such as Figure 5 As shown, the device comprises: The vehicle driving data acquisition module 510 is used to acquire the vehicle driving data of a plurality of consecutive moments including the current moment during the vehicle driving process; A driving style detection module 520, configured to input the vehicle driving data into a driving style detection model obtained through online training to obtain the driver's driving style at the next moment; A vehicle control module 530, configured to control the vehicle accordingly based on the driving style; Among them, the online training process of the driving style detection model includes: predicting a predicted value of vehicle driving data at the next moment based on the vehicle driving data, and generating an online feedback strategy based on an error between the vehicle driving data and the predicted value of the vehicle driving data, and performing online feedback training on the driving style detection model based on the online feedback strategy.

[0074] In a third aspect, an embodiment of the present invention provides an electronic device 600, such as Figure 6 As shown, including: at least one processor 601; a memory 602 for storing the at least one processor executable instruction; The at least one processor is configured to execute the instructions to implement the method described in the first aspect.

[0075] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method described in the first aspect.

[0076] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and intent of the present invention.

Claims

1. A vehicle driving style adaptation method based on online learning, characterized in that: The method comprises: During the driving process of the vehicle, obtaining the vehicle driving data at a plurality of consecutive moments including the current moment; Inputting the vehicle driving data into a driving style detection model obtained through online training to obtain the driver's driving style at the next moment; controlling the vehicle accordingly based on the driving style; The online training process of the driving style detection model includes: predicting a predicted value of vehicle driving data at the next moment based on the vehicle driving data, generating an online feedback strategy based on an error between the vehicle driving data and the predicted value of the vehicle driving data, and performing online feedback training on the driving style detection model based on the online feedback strategy; The process of the driving style detection model predicting the driver's driving style at the next moment includes: Generate vehicle driving time series data based on the vehicle driving data, wherein the vehicle driving time series data is used to characterize time series characteristics and context association information of the vehicle driving data; Extracting key time series data from the vehicle driving time series data and the attention weight of each key time series data based on the attention mechanism; Converting the key time series data into target time series data in a time series format; Based on the target time series data and the attention weights corresponding to each target time series data, the driver's driving style at the next moment is predicted.

2. The method according to claim 1, characterized in that The driving style detection model includes a time series encoder, a latent space attention mechanism module including a self-attention mechanism and a cross-attention mechanism, and a time series decoder; The generating of vehicle driving time series data based on the vehicle driving data comprises: Encoding the vehicle driving data into vehicle driving time series data by the time series encoder; Accordingly, the key time series data in the vehicle driving time series data and the attention weight of each key time series data are extracted based on the attention mechanism, including: The vehicle driving time series data is processed by the latent space attention mechanism module to obtain key time series data in the vehicle driving time series data and the attention weight of each key time series data; Correspondingly, the converting of the key time series data into target time series data in a time series format includes: converting the key time series data into target time series data in a time series format by using the time series decoder.

3. The method according to claim 2, characterized in that The driving style detection model also includes a pedal and steering prediction module and an online gradient network; The method predicts a predicted value of vehicle driving data at the next moment based on the vehicle driving data, generates an online feedback strategy based on an error between the vehicle driving data and the predicted value of the vehicle driving data, and performs online feedback training on a driving style detection model based on the online feedback strategy to obtain a driving style detection model after online training, including: Based on the target time series data and the attention weights corresponding to each target time series data, predicting the pedal data prediction value and the steering data prediction value at the next moment through the pedal and steering prediction module; Calculating the pedal data and steering data included in the vehicle driving data by the online gradient network, and the errors between the pedal data prediction value and the steering data prediction value, determining an online feedback strategy based on the errors, and feeding back the online feedback strategy to the latent space attention mechanism module; The latent space attention mechanism module is trained online based on the online feedback strategy.

4. The method according to any one of claims 1 to 3, characterized in that: The step of obtaining vehicle driving data at a plurality of consecutive moments including the current moment includes: Acquire initial vehicle driving data at a plurality of consecutive moments including the current moment, wherein the initial vehicle driving data includes pedal data, steering data, driving speed, and yaw speed of the vehicle; The initial vehicle driving data is preprocessed to obtain preprocessed vehicle driving data, wherein the preprocessing includes data fusion and data denoising.

5. The method according to any one of claims 1 to 3, characterized in that: The driving style includes multiple types, and the vehicle driving data is input into the driving style detection model obtained by online training to obtain the driver's driving style at the next moment, including: Inputting the vehicle driving data into a driving style detection model obtained through online training, and outputting multiple types of driving styles and the probability corresponding to each type of driving style through the driving style detection model; Based on the multiple types of driving styles and the probability corresponding to each type of driving style, the driving style of the driver at the next moment is calculated.

6. The method according to claim 5, characterized in that The controlling the vehicle accordingly based on the driving style includes: receiving a torque request input by a driver via an accelerator pedal; Obtain torque variation curves corresponding to various types of driving styles; Based on the probabilities corresponding to the various types of driving styles, weighted summation is performed on the torque change curves corresponding to the various types of driving styles to obtain the torque change curve after the weighted summation; The output torque is obtained through the weighted summed torque variation curve, and the vehicle is controlled through the output torque.

7. A vehicle driving style adaptive device based on online learning, characterized in that: The device comprises: The vehicle driving data acquisition module is used to acquire the vehicle driving data of a plurality of consecutive moments including the current moment during the vehicle driving process; A driving style detection module, used for inputting the vehicle driving data into a driving style detection model obtained through online training to obtain the driver's driving style at the next moment; A vehicle control module, configured to control the vehicle accordingly based on the driving style; The online training process of the driving style detection model includes: predicting a predicted value of vehicle driving data at the next moment based on the vehicle driving data, generating an online feedback strategy based on an error between the vehicle driving data and the predicted value of the vehicle driving data, and performing online feedback training on the driving style detection model based on the online feedback strategy; The process of the driving style detection model predicting the driver's driving style at the next moment includes: Generate vehicle driving time series data based on the vehicle driving data, wherein the vehicle driving time series data is used to characterize time series characteristics and context association information of the vehicle driving data; Extracting key time series data from the vehicle driving time series data and the attention weight of each key time series data based on the attention mechanism; Converting the key time series data into target time series data in a time series format; Based on the target time series data and the attention weights corresponding to each target time series data, the driver's driving style at the next moment is predicted.

8. An electronic device, characterized in that: include: at least one processor; a memory for storing the at least one processor-executable instruction; The at least one processor is configured to execute the instructions to implement the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method according to any one of claims 1 to 6.

Images