Vehicle control method, electronic device, and vehicle

Abstract

The present disclosure relates to the technical field of vehicle intelligent driving, and provides a vehicle control method, an electronic device and a vehicle. A vehicle torque change is detected, vehicle operation data and environment data are obtained, and target deviation characteristics are determined according to the vehicle operation data and the environment data. A vehicle initial torque is obtained, and a vehicle candidate torque is determined according to the target deviation characteristics and the vehicle initial torque. Vehicle running data is obtained, a vehicle target torque is determined according to the vehicle running data, the target deviation characteristics and the vehicle candidate torque based on a deviation adjustment model obtained through pre-training, and the vehicle is controlled to travel according to the vehicle target torque. In the present disclosure, when the vehicle target torque is determined, the vehicle running data, the deviation degree between the driver operation behavior and the vehicle expected response and the ideal torque value are comprehensively considered, the problem that the driver throttle misstep intention recognition is inaccurate and the torque output is inaccurate due to the environment inadaptation is solved, and the safety and stability of driving are improved.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent driving technology for vehicles, and in particular to a vehicle control method, electronic device, and vehicle. Background Technology

[0002] With the rapid development of vehicle technology, vehicles have become an important means of transportation in people's daily lives. In related technologies, driving behavior modeling is mostly based on traditional regression prediction or simple classification methods, which makes it difficult to accurately identify the conflict between the driver's subjective intentions and the current environmental state. This results in a lack of timely intervention when the driver makes a mistake, threatening driving safety. Summary of the Invention

[0003] In view of this, the purpose of this disclosure is to propose a vehicle control method, electronic device and vehicle to solve the problem that the current lack of understanding of the deep interaction between driver intentions and environmental needs leads to the inability to intervene in a timely manner when the driver makes a mistake, thus threatening driving safety.

[0004] To achieve the above objectives, a first aspect of this disclosure provides a vehicle control method, the method comprising:

[0005] The vehicle torque change is detected, vehicle control data and environmental data are acquired, and the target deviation characteristics are determined based on the vehicle control data and the environmental data. Obtain the initial torque of the vehicle, and determine the candidate torque of the vehicle based on the target deviation characteristics and the initial torque of the vehicle; The vehicle operation data is acquired, and the target torque of the vehicle is determined based on the pre-trained deviation adjustment model according to the vehicle operation data, the target deviation characteristics, and the candidate torque of the vehicle. The vehicle driving is controlled according to the target torque.

[0006] Based on the same inventive concept, a second aspect of this disclosure provides a vehicle control device, comprising: The data acquisition module is configured to detect changes in vehicle torque, acquire vehicle control data and environmental data, and determine target deviation characteristics based on the vehicle control data and the environmental data. The torque determination module is configured to acquire the initial torque of the vehicle and determine the candidate torque of the vehicle based on the target deviation characteristics and the initial torque of the vehicle. The vehicle control module is configured to acquire vehicle operating data, determine the vehicle target torque based on the pre-trained deviation adjustment model according to the vehicle operating data, the target deviation characteristics and the vehicle candidate torque, and control the vehicle driving according to the vehicle target torque.

[0007] Based on the same inventive concept, a third aspect of this disclosure proposes an electronic device including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the vehicle control method as described above when executing the computer program.

[0008] Based on the same inventive concept, a fourth aspect of this disclosure provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to perform the vehicle control method as described above.

[0009] Based on the same inventive concept, the fifth aspect of this disclosure provides a vehicle including the vehicle control device described in the second aspect, the electronic device described in the third aspect, or the storage medium described in the fourth aspect.

[0010] As can be seen from the above, this disclosure proposes a vehicle control method, electronic device, and vehicle. The method detects changes in vehicle torque, acquires vehicle control data and environmental data, and determines a target deviation feature based on the vehicle control data and the environmental data. The target deviation feature represents the degree of mismatch between the driver's control behavior and the vehicle's expected response. The method acquires the vehicle's initial torque and determines a candidate torque based on the target deviation feature and the initial torque. The candidate torque is the ideal torque. The method acquires vehicle operating data and determines a target torque based on a pre-trained deviation adjustment model, according to the vehicle operating data, the target deviation feature, and the candidate torque. The method then controls the vehicle's movement based on the target torque. In determining the target torque, the method comprehensively considers vehicle operating data, the degree of deviation between the driver's control behavior and the vehicle's expected response, and the ideal torque value. This solves the problems of inaccurate recognition of the driver's accelerator pedal mis-pressing intention and inaccurate torque output due to environmental maladaptation. Subsequently, the method controls the vehicle based on the target torque, improving driving safety and stability. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in this disclosure or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a vehicle control method according to an embodiment of the present disclosure; Figure 2 This is a structural block diagram of a vehicle control device according to an embodiment of the present disclosure; Figure 3This is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0014] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0015] The following are definitions of terms used in this disclosure: Equiformer: Equiformer is a graph neural network. The core innovation of this model is that it replaces the original module in the standard Transformer with equivariant operations and introduces a brand-new mechanism called "equivariant graph attention", which effectively encodes symmetries such as rotation and translation during information transmission.

[0016] Score-Based Diffusion Model: This model is a generative model that progressively adds noise to the data (forward diffusion process) and learns to reverse this process. Its core is to train a neural network to estimate the log density gradient of the data distribution. By learning these scores on multiple noise scales, the model can start from pure noise, follow the estimated score field for Langevin dynamics sampling, and thus progressively denoise and restore a new high-fidelity sample consistent with the training data distribution.

[0017] Neuromodulation Control Head: The neuromodulation output layer is an intelligent control mechanism inspired by biological neuroscience. By simulating the dynamic regulation of neurotransmitters, it enables the model to actively generate control signals based on context and internal state to dynamically adjust the connection weights, activation function characteristics, or information flow of the neural network itself. This allows for adaptive switching to different tasks or environments, solving the problem that traditional static models cannot flexibly cope with complex changes.

[0018] In related technologies, driving behavior modeling methods are mostly based on traditional regression prediction models or simple classification algorithms. These methods usually rely on pre-defined rules or statistical features to fit the relationship between driver operation and vehicle response.

[0019] However, these models lack a deep understanding of the semantics of driving scenarios, making it difficult to construct a structured relationship between the driver's subjective intentions and the current environmental state, and failing to accurately identify potential conflicts between the two. Especially in scenarios involving driver error, such as when the driver should lightly apply the brakes or maintain a constant speed but mistakenly presses the accelerator due to misjudgment or nervousness, existing systems often trigger acceleration responses based solely on the instantaneous pedal opening, failing to promptly identify the contradiction between this operation and safety requirements. This delays or misses the opportunity to intervene, leading to unexpected vehicle acceleration and increasing driving risks.

[0020] Specifically, at the expression level, the relevant technologies fail to effectively establish a structured connection between driving intentions and environmental semantics; that is, the system lacks the ability to jointly understand what the driver wants and what the current road conditions require. Secondly, most current control strategies still employ linear reduction mechanisms, such as directly reducing power output based on a single threshold, lacking flexible adjustment methods based on dynamic scenarios. This results in abrupt interventions, affecting driving smoothness and acceptability. Finally, there is a lack of effective coordination mechanisms between functional modules. The model cannot dynamically integrate behavioral deviations, changes in operating conditions, and personalized driving styles based on real-time data. Consequently, when facing complex and ever-changing real-world driving environments, the system struggles to adaptively adjust its control strategies.

[0021] In summary, related technologies generally suffer from significant shortcomings in driving misjudgment detection and buffer control, including insufficient accuracy, poor adaptability, and weak flexibility. This not only limits the performance of advanced driver assistance systems (ADAS) in critical safety scenarios but also weakens the trust foundation and practical value of human-machine co-driving systems to some extent. Therefore, there is an urgent need to develop novel driving behavior modeling methods with semantic understanding, flexible control, and collaborative decision-making capabilities to improve the safety and adaptability of systems in complex interactive scenarios.

[0022] Therefore, this embodiment proposes a vehicle control method, such as... Figure 1 As shown, the method includes: Step 101: Detect vehicle torque change, acquire vehicle control data and environmental data, and determine target deviation characteristics based on the vehicle control data and environmental data.

[0023] In practice, a change in vehicle torque is detected, indicating that the user is actively controlling the vehicle while driving, such as by pressing the accelerator or brake pedal. Vehicle control data and environmental data are acquired, and a target deviation feature is determined based on the vehicle control data and the environmental data. The target deviation feature represents the degree of mismatch between the driver's control behavior and the vehicle's expected response.

[0024] In this embodiment, the vehicle control data represents the user's subjective input data, including at least one of the following: accelerator pedal opening, accelerator pedal opening change rate, accelerator pedal application force, accelerator pedal pressing duration, brake pedal opening, brake pedal opening change rate, brake pedal application force, brake pedal pressing duration, steering wheel angle, steering wheel rotation speed, etc.

[0025] In this embodiment, the environmental data represents environmental information of the vehicle's environment, including at least one of the following: road type, traffic flow on the road, ambient temperature, ambient visibility, road adhesion coefficient, etc.

[0026] In this embodiment, when determining the target deviation features based on vehicle control data and environmental data, a pre-trained deviation recognition model can be obtained. Then, the vehicle control data and environmental data are input into the deviation recognition model, and the target deviation features are obtained through processing by the deviation recognition model.

[0027] Specifically, the training process of the deviation recognition model includes: Step a, obtain the first training dataset and the initial deviation recognition model, wherein the first training dataset includes historical vehicle handling data, historical environmental data and historical deviation features.

[0028] Step b: Input the training data in the first training dataset into the initial deviation recognition model for training, determine that the first preset training termination condition is met, and obtain the deviation recognition model.

[0029] In specific implementation, a first training dataset and an initial deviation recognition model are obtained, wherein the first training dataset includes historical vehicle handling data, historical environmental data, and historical deviation features. The training data from the first training dataset is input into the initial deviation recognition model for training. Once a first preset training termination condition is met, the deviation recognition model is obtained.

[0030] The first preset training termination condition includes at least one of the following: determining that all data in the first training dataset has been input into the initial deviation identification model for training, determining that the loss function of the initial deviation identification model has converged to a first convergence threshold, or determining that the initial deviation identification model has been iteratively trained to a first preset number of iterations.

[0031] For example, the first preset training termination condition is to determine that all data in the first training dataset has been input into the initial bias recognition model for training: The first training dataset contains fifty sets of data, each set including historical vehicle handling data, historical environmental data, and historical deviation features. The first preset training termination condition is that all data in the first training dataset has been input into the initial deviation recognition model for training. That is, when all fifty sets of data have been input into the initial deviation recognition model, there is no training data in the first training dataset that has not yet been input into the initial deviation recognition model. At this point, the initial deviation recognition model training is considered complete, and the deviation recognition model is obtained.

[0032] In another example, the first preset training termination condition is to determine that the loss function of the initial bias identification model converges to a first convergence threshold: The training data from the first training dataset is input into the initial deviation identification model for training, and the training results are output. A loss function is determined based on the training results and the actual deviation characteristics. The loss function may include at least one of the following: mean squared error loss function, cross-entropy loss function, logarithmic loss function, exponential loss function, squared loss function, or absolute value loss function, etc. When the loss function converges to a first convergence threshold, a first preset training termination condition is satisfied, and the deviation identification model is obtained.

[0033] In another example, the first preset training termination condition is to determine the initial deviation identification model iteratively train to the first preset number of iterations.

[0034] The training data in the first training dataset is input into the initial deviation recognition model for iterative training. The number of iterations is recorded. When the number of iterations is equal to the first preset number of iterations, the first preset training termination condition is met, and the deviation recognition model is obtained.

[0035] In this embodiment, to better extract the dynamic structural change characteristics of the vehicle driven by driver behavior, a vehicle dynamic symmetry analysis framework is introduced. Based on data such as the difference in speed between the left and right wheels, the difference in compression travel between the front and rear suspensions, and the ratio of lateral acceleration to steering wheel angle, the structural symmetry residual of the vehicle is calculated, thus obtaining the symmetry characteristics of the vehicle's dynamic response. Furthermore, these symmetry characteristics, along with vehicle handling data and environmental data, are input into a pre-trained deviation recognition model. The deviation recognition model processes the data to obtain the target deviation characteristics. It is understood that historical symmetry characteristics can be input into the model for training during deviation recognition model training.

[0036] Specifically, under ideal conditions of smooth driving, a vehicle's structural response should maintain left-right geometric symmetry. However, when the driver performs maneuvers such as changing lanes, cornering, or traversing bumpy roads, the vehicle's force state changes, disrupting the original symmetrical balance. By collecting real-time data from sensors deployed at key vehicle nodes, a comprehensive symmetry characteristic of the vehicle's dynamic response is constructed.

[0037] Specifically, this symmetry feature integrates the left and right wheel speed difference, which characterizes the difference in drive wheel slippage; the front and rear suspension compression travel difference, which reflects changes in vehicle pitch and roll attitude; and the ratio of lateral acceleration to steering wheel angle, which characterizes steering response sensitivity. By calculating the residual magnitude of these multidimensional data relative to the ideal symmetry state, the abstract structural dynamic changes caused by driver behavior can be visualized as a quantifiable and traceable sequence of structural symmetry residuals, thus providing crucial data support for subsequent vehicle dynamics analysis and driver intent recognition.

[0038] In this embodiment, the deviation recognition model is a neural network structure model. Preferably, the deviation recognition model is a Transformer model structure based on symmetry perception. Specifically, the model is based on the Equiformer model to model the control deviation in subjective driving behavior. The Equiformer is a Transformer model optimized for physical structural symmetry, which has the ability to identify deviation features in symmetrical structures.

[0039] Specifically, the encoder part of the Equiformer model consists of a multi-layered symmetric attention module. Each layer extracts the offset direction, offset magnitude, and coupling relationship with the current vehicle state in the control behavior by constructing a cross-attention map with left-right / front-back / time symmetric dimensions. The decoder part converts the features into quantitative control deviation scores through an embedded deviation mapping network, and outputs a set of time-series deviation distributions, i.e., target deviation features, which quantifies the degree of structural mismatch between the driver's behavior and the vehicle's expected response at each moment.

[0040] For example, if the driver presses the accelerator too hard during the start-up process, but the vehicle is in a low-traction environment, such as a slippery road surface, the model identifies behavioral samples in the symmetry dimension where the longitudinal acceleration-throttle change rate deviates from the symmetry range and marks them as high-risk mis-pressing features.

[0041] In this embodiment, a cross-channel adaptive residual connection mechanism is incorporated into the deviation recognition model to improve its adaptability to different driving styles while preserving long-term dependency structures. During training, a symmetric loss function and a weighted strategy for mis-pressing samples are used to focus the model on driving behavior regions prone to mis-pressing. By introducing traditional behavior sequence modeling into a symmetric feature space through the Equiformer model, the micro-temporal and structural anomalies of excessive or skewed pressing by the driver are effectively captured, thereby achieving prior recognition of misjudged actions.

[0042] Step 102: Obtain the initial torque of the vehicle, and determine the candidate torque of the vehicle based on the target deviation characteristics and the initial torque of the vehicle.

[0043] In practice, the initial torque of the vehicle is obtained, where the initial torque is the original torque output under the actual pedaling action of the driver. Based on the target deviation characteristics and the initial torque, a candidate torque for the vehicle is determined, where the candidate torque is the ideal torque.

[0044] In this embodiment, after acquiring vehicle control data, environmental data, symmetry features, vehicle initial torque, and vehicle operation data, timestamp alignment and sampling frequency unification are performed. That is, synchronous processing and anomaly removal are performed in the local edge computing unit to construct a high-quality, high-resolution multi-source dataset that reflects both the randomness of the driver's subjective control and the objective regularity of the vehicle's physical response, providing accurate and rich feature support for subsequent model training.

[0045] Step 103: Obtain vehicle operation data, determine the vehicle target torque based on the pre-trained deviation adjustment model according to the vehicle operation data, the target deviation characteristics and the vehicle candidate torque, and control the vehicle driving according to the vehicle target torque.

[0046] In specific implementation, vehicle operating data is acquired, including at least one of the following: vehicle speed, vehicle acceleration, engine speed, current gear, torque output curve and its gradient change, vehicle pitch angle, vehicle roll angle, vehicle yaw rate of change, and ESP and TCS intervention status. ESP is the Electronic Stability Program, and TCS is the Traction Control System.

[0047] A pre-trained deviation adjustment model is obtained. Based on the deviation adjustment model, the vehicle target torque is determined according to the vehicle operation data, the target deviation characteristics, and the vehicle candidate torque. The vehicle driving is controlled according to the vehicle target torque.

[0048] In this embodiment, after determining the vehicle's target torque, a redundant path verification is performed, which is compared in parallel with the original driver's intention to ensure that the system can quickly revert to its previous state in abnormal conditions, thus guaranteeing the safety and reliability of the main control path. Furthermore, based on the vehicle's target torque, the execution modules, including the throttle opening controller, fuel injection system management unit, transmission shift logic, and traction control system, are controlled.

[0049] The above scheme detects changes in vehicle torque, acquires vehicle control data and environmental data, and determines a target deviation feature based on the vehicle control data and environmental data. The target deviation feature represents the degree of mismatch between the driver's control behavior and the vehicle's expected response. The initial vehicle torque is acquired, and a candidate torque is determined based on the target deviation feature and the initial vehicle torque. The candidate torque is the ideal torque. Vehicle operating data is acquired, and a target torque is determined based on a pre-trained deviation adjustment model, the vehicle operating data, the target deviation feature, and the candidate torque. The vehicle is then controlled according to the target torque. In determining the target torque, the scheme comprehensively considers vehicle operating data, the degree of deviation between the driver's control behavior and the vehicle's expected response, and the ideal torque value. This addresses the problems of inaccurate recognition of the driver's accelerator pedal mis-pressing intention and inaccurate torque output due to environmental maladaptation. Consequently, subsequent vehicle control based on the target torque improves driving safety and stability.

[0050] In some embodiments, when determining the target torque of a vehicle, vehicle operating data and target deviation characteristics can first be input into a deviation adjustment model to obtain the target risk level. Then, the target torque of the vehicle is determined based on the target risk level and candidate vehicle torques. Specifically, step 103, which uses a pre-trained deviation adjustment model to determine the target torque based on the vehicle operating data, the target deviation characteristics, and the candidate vehicle torques, includes: Step 1031: Determine the target fusion vector based on the vehicle operation data and the target deviation characteristics; Step 1032: Input the target fusion vector into the pre-trained bias adjustment model, and process it through the bias adjustment model to obtain the target risk level; Step 1033: Determine the target torque of the vehicle based on the target risk level and the candidate torque of the vehicle.

[0051] In specific implementation, a target fusion vector is determined based on the vehicle operation data and the target deviation characteristics. The target fusion vector is input into a pre-trained deviation adjustment model, and after processing by the deviation adjustment model, a target risk level is obtained. The target risk level represents the degree of risk caused by the intention deviation in the current vehicle operation scenario, which indicates the degree of driving safety risk.

[0052] In this embodiment, the target risk level is a first level, a second level, or a third level, and is ordered according to the intensity of the risk level, from high risk to low risk as first level, second level, and third level. Specifically, the first level indicates a high risk level due to intention deviation, requiring output based on the vehicle's determined ideal torque. The second level indicates a risk due to intention deviation, but a low risk, allowing for flexible adjustment of the user-inputted initial vehicle torque. The third level indicates that the intention deviation does not threaten driving safety.

[0053] Specifically, the training process of the deviation adjustment model includes: Step a, obtain the second training dataset and the initial bias adjustment model, wherein the second training dataset includes historical fusion vectors and historical risk levels.

[0054] Step b: Input the training data from the second training dataset into the initial bias adjustment model for training, determine that the second preset training termination condition is met, and obtain the bias adjustment model.

[0055] In practice, a second training dataset and an initial bias adjustment model are obtained, wherein the second training dataset includes historical fusion vectors and historical risk levels. Training data from the second training dataset is input into the initial bias adjustment model for training. Once a second preset training termination condition is met, the bias adjustment model is obtained.

[0056] The second preset training termination condition includes at least one of the following: determining that all data in the second training dataset has been input into the initial bias adjustment model for training, determining that the loss function of the initial bias adjustment model has converged to the second convergence threshold, or determining that the initial bias adjustment model has been iterated for training to the second preset number of iterations.

[0057] For example, the second preset training termination condition is to determine that all data in the second training dataset has been input into the initial bias adjustment model for training: The second training dataset contains fifty sets of data, each set including a historical fusion vector and a historical risk level. The second preset training termination condition is that all data in the second training dataset has been input into the initial bias adjustment model for training. That is, when all fifty sets of data have been input into the initial bias adjustment model, there is no training data in the second training dataset that has not yet been input into the initial bias adjustment model. At this point, the initial bias adjustment model training is considered complete, and the bias adjustment model is obtained.

[0058] In another example, the second preset training termination condition is to determine that the loss function of the initial bias-adjusting model converges to a second convergence threshold: The training data from the second training dataset is input into the initial bias adjustment model for training, and the training results are output. A loss function is determined based on the training results and the historical risk level. The loss function may include at least one of the following: mean squared error loss function, cross-entropy loss function, logarithmic loss function, exponential loss function, squared loss function, or absolute value loss function, etc. When the loss function converges to a second convergence threshold, it is determined that the second preset training termination condition is met, and the bias adjustment model is obtained.

[0059] In another example, the second preset training termination condition is to determine the initial bias adjustment model iteratively trained up to the second preset number of iterations.

[0060] The training data in the second training dataset is input into the initial bias adjustment model for iterative training. The number of iterations is recorded. When the number of iterations is equal to the second preset number of iterations, the second preset training termination condition is met, and the bias adjustment model is obtained.

[0061] After determining the target risk level, the target torque of the vehicle is determined based on the target risk level and the candidate torques of the vehicle.

[0062] In this embodiment, the deviation adjustment model is a neural network structure model, preferably a model employing a neural modulator structure. The internal structure of the deviation adjustment model includes a multi-layer modulation factor extraction network. First, the input features are normalized and channel-encoded, whereby the input features include vehicle operation data and the target deviation features. Then, an attention mechanism and a state gating unit dynamically weight each input channel, outputting a set of fused weights and a modulation vector. This vector is further input into the control head structure, which consists of a set of high-dimensional nonlinear mapping layers. This structure performs control trajectory reconstruction and disturbance suppression function generation under multiple input states, ultimately obtaining the target risk level. By automatically determining whether the current behavior is in a normal intention control state or a suspected accidental collision state using neural dynamic modulation, the target risk level is determined accordingly.

[0063] Specifically, step 1031, which involves determining the target fusion vector based on the vehicle operation data and the target deviation characteristics, includes: Step 10311: Map and align the vehicle operation data and the target deviation features to obtain the operation data embedding vector and the deviation vector; Step 10312: Obtain the target fusion weights, input the target fusion weights, the running data embedding vector, and the deviation vector into a pre-trained vector fusion model, process the vector fusion model, and output the target fusion vector.

[0064] In specific implementation, the vehicle operation data and the target deviation features are mapped and aligned to obtain the operation data embedding vector, the deviation vector, and the temperature embedding vector.

[0065] Specifically, a linear transformation is performed on the vehicle operation information to map it to a preset semantic space, resulting in an operation data embedding vector. Similarly, a linear transformation is performed on the target deviation feature to map it to the preset semantic space, resulting in a deviation vector.

[0066] The aligned running data embedding vector and bias vector are further fused through a multi-head attention mechanism, enabling the model to identify deep relationships and causal connections between them. Specifically, the target fusion weights, the running data embedding vectors, and the bias vectors are input into a pre-trained vector fusion model, which outputs the target fusion vector.

[0067] In this embodiment, the target fusion weight represents the weight value corresponding to the running data embedding vector and the deviation vector when fusing the running data embedding vector and the deviation vector to obtain the target fusion vector. In this embodiment, the target fusion weight can be preset or dynamically adjusted according to the accuracy of the running data embedding vector and the deviation vector, so as to ensure that the obtained target fusion vector is more consistent with the vehicle's environment and vehicle running data.

[0068] The above scheme fuses the embedded vector and the deviation vector of the running data to obtain the target fusion vector. This allows the subsequent deviation adjustment model to automatically identify the correlation between the two modes when analyzing and processing the target fusion vector, thereby improving the accuracy of determining the target risk level.

[0069] In some embodiments, determining the vehicle target torque based on the target risk level and the vehicle candidate torque in step 1033 specifically includes: Step 10331: In response to the target risk level being Level 1, determine the vehicle target torque as the vehicle candidate torque; or, Step 10332: In response to the target risk level being level two, determine a target adjustment coefficient based on the target deviation characteristics, and determine the vehicle target torque based on the target adjustment coefficient and the vehicle initial torque; or, Step 10333: In response to the target risk level being level three, determine the vehicle target torque as the vehicle initial torque.

[0070] In practice, after determining the target risk level, if the target risk level is the first level, it means that the risk level is high due to intention deviation. The output needs to be based on the ideal torque determined by the vehicle, that is, the target torque of the vehicle is determined as the candidate torque of the vehicle.

[0071] If the target risk level is level two, it indicates that there is a risk due to intentional deviation, but the risk is relatively small. Therefore, the initial vehicle torque input by the user can be flexibly adjusted. A target adjustment coefficient is determined based on the target deviation characteristics, and the target vehicle torque is determined based on the target adjustment coefficient and the initial vehicle torque. Smooth following is ensured by slowing down the throttle response rate and reducing the maximum torque amplitude.

[0072] If the target risk level is level three, it means that the intention deviation will not threaten driving safety. Therefore, the initial torque of the vehicle corresponding to the user's control behavior can be directly used as the target torque of the vehicle.

[0073] For example, in scenarios such as low-speed following in urban areas and dense urban traffic, if the target risk level is level three, meaning a mis-press of the accelerator is detected but the intention deviation is small and does not threaten driving safety, the output curve is closer to the original intention, and the target torque of the vehicle is determined as the initial torque of the vehicle. If the target risk level is level two, such as detecting a mis-press of the accelerator and the distance to the vehicle in front being too close, the initial torque of the vehicle is flexibly adjusted. If the target risk level is level three, meaning the deviation is extremely large, the candidate torque of the vehicle is directly used to replace the driver's input, improving safety.

[0074] The above scheme determines the target torque of the vehicle based on the target risk level, thereby achieving smooth control of the vehicle and an efficient balance between the stability of the accidental braking control and the driver's subjective experience.

[0075] In some embodiments, step 10332, which involves determining the target adjustment coefficient based on the target deviation characteristics and determining the vehicle target torque based on the target adjustment coefficient and the vehicle initial torque, specifically includes: Step A: Input the target deviation features into the pre-trained deviation level determination model, and process them through the deviation level determination model to obtain the target intention deviation level; Step B: Determine the target adjustment coefficient based on the target intention deviation level; Step C: Multiply the target adjustment coefficient and the initial torque of the vehicle to obtain the target torque of the vehicle.

[0076] In practice, a pre-trained deviation level determination model is obtained, the target deviation features are input into the deviation level determination model, and the model processes the data to obtain the target intent deviation level. The intent deviation level represents the degree of intent deviation.

[0077] Based on the target intent deviation level, a target adjustment coefficient is determined. Specifically, a database is searched based on the target intent deviation level to determine the target adjustment coefficient corresponding to that level. The database stores the correspondence between target intent deviation levels and target adjustment coefficients.

[0078] The target adjustment coefficient and the initial torque of the vehicle are multiplied together to obtain the target torque of the vehicle.

[0079] Specifically, step B, which involves determining the target adjustment coefficient based on the target intention deviation level, includes: Step B1: Obtain vehicle model data and driver identification; determine a first correction coefficient based on the vehicle model data; and determine a second correction coefficient based on the driver identification. Step B2: Determine the initial adjustment coefficient based on the target intention deviation level, and correct the initial adjustment coefficient based on the first correction coefficient and the second correction coefficient to obtain the target adjustment coefficient.

[0080] In practice, vehicle model data and driver identifiers are acquired. The vehicle model data represents the vehicle's model information, and the driver identifier is a unique identifier distinguishing different drivers. A first correction coefficient is determined based on the vehicle model data, and a second correction coefficient is determined based on the driver identifier.

[0081] Specifically, a first correction coefficient corresponding to the vehicle model data is determined by searching a database based on the vehicle model data, wherein the database stores the correspondence between vehicle model data and correction coefficients. A second correction coefficient corresponding to the driver identifier is determined by searching a database based on the driver identifier, wherein the database stores the correspondence between driver identifiers and correction coefficients.

[0082] An initial adjustment coefficient is determined based on the target intent deviation level, specifically by searching a database to find the initial adjustment coefficient corresponding to the target intent deviation level. Then, the initial adjustment coefficient is corrected based on the first correction coefficient and the second correction coefficient to obtain the target adjustment coefficient.

[0083] Specifically, step B2, which involves correcting the initial adjustment coefficient based on the first and second correction coefficients to obtain the target adjustment coefficient, includes: Step B21: Compare the first correction coefficient and the second correction coefficient, and select the maximum value of the first correction coefficient and the second correction coefficient as the target correction coefficient; Step B22: Multiply the target correction coefficient and the initial adjustment coefficient to obtain the target adjustment coefficient.

[0084] In practice, after determining the first correction coefficient based on vehicle model data and the second correction coefficient based on driver identification, the first and second correction coefficients are compared. The maximum value of the first and second correction coefficients is selected as the target correction coefficient. That is, if the first correction coefficient is greater than or equal to the second correction coefficient, the target correction coefficient is determined to be the first correction coefficient. If the first correction coefficient is less than the second correction coefficient, the target correction coefficient is determined to be the second correction coefficient.

[0085] The target adjustment coefficient is obtained by multiplying the target correction coefficient and the initial adjustment coefficient.

[0086] The above scheme determines the target adjustment coefficient based on driver identification and vehicle model data, and then multiplies it with the initial torque of the vehicle to obtain the target torque. That is, when the target risk level is level two, the adjustment of the initial torque of the vehicle takes into account the driver's habits and vehicle model parameters, which improves the accuracy of the target adjustment coefficient, enhances the adjustment intensity, and ensures a consistent driving experience and safety feedback under different driving styles.

[0087] In some embodiments, the step 102 of determining the candidate vehicle torque based on the target deviation characteristics and the initial vehicle torque specifically includes: Step 1021: Obtain the pre-trained torque determination model; Step 1022: Input the target deviation feature and the initial torque of the vehicle into the torque determination model, and process them through the torque determination model to obtain the candidate torque of the vehicle.

[0088] In practice, a pre-trained torque determination model is obtained, and the target deviation features and the initial torque of the vehicle are input into the torque determination model. The torque determination model then processes the torque to obtain the candidate torque of the vehicle.

[0089] In this embodiment, the torque determination model is a model employing a neural network structure. Preferably, the torque determination model is a score-based diffusion model, used to model the mapping relationship between the ideal torque distribution and the output response under mis-stepping actions. The score-based diffusion model is based on a generative modeling framework and simulates the reconstruction process from a noisy torque sequence to a high-quality torque control trajectory through a stepwise inverse process. The core idea is to guide mis-stepping behavior to the sample space of a reasonable control curve.

[0090] Specifically, the inputs to the torque determination model include target deviation features and the initial vehicle torque. The target deviation features serve as a guiding condition encoder, and the initial vehicle torque forms a diffusion trajectory through Gaussian noise perturbation. The main structure of the torque determination model includes multiple score-based residual prediction modules, each learning to estimate the optimal residual direction between the current noise state and the target torque state at different time steps.

[0091] In this embodiment, during training, a fractional function is learned to recover the ideal state from an initial state with arbitrary disturbances by minimizing the distance between the predicted residual and the actual residual at each step. Simultaneously, the torque determination model incorporates multi-condition embeddings during training, such as slope information, current vehicle weight, drive mode, and temperature, as condition vectors to enhance the model's adaptive ability to soften torque curves under complex environments. Furthermore, to achieve more refined torque generation control, the torque determination model simultaneously learns torque distributions at multiple scales, from instantaneous output with microsecond-level response to dynamic output with second-level smooth control, ensuring that the model can generate reasonable softening curves at different control frequencies. The final model output is a set of adjustable ideal torque curve candidate distributions, including upper and lower limits, the main curve, and a buffer band, thus obtaining the vehicle's candidate torque, which is provided to the control module for fusion decision-making.

[0092] The above approach introduces the torque control problem into the diffusion modeling space, and learns to generate an ideal control trajectory through a fractional guidance method. Compared with traditional supervised regression methods, it exhibits higher robustness and generation quality in mis-pressing identification and control response. The high-dimensional generation capability of the torque determination model allows for the nonlinear mapping from mis-pressing behavior to the desired control curve while preserving driver intention characteristics, significantly improving the system's buffering and intervention capabilities for sudden mis-pressing events.

[0093] In this embodiment, the deviation identification model, deviation adjustment model, and torque determination model can be jointly trained. The specific process includes: In the joint training, the deviation identification model module first processes the input data and outputs a deviation distribution indicating whether the current driving behavior involves mis-pressing the pedal. This distribution is then injected into the torque determination model and the deviation adjustment model, providing important prior knowledge for subsequent generation and fusion decisions.

[0094] Specifically, during its training process, the torque determination model, in addition to maintaining its core reconstruction loss—that is, minimizing the residual between the noisy torque state at each step and the final target ideal torque through iterative denoising—innovatively introduces the degree of deviation from the deviation identification model as a dynamic adjustment factor for the torque softening amplitude. This means that the greater the probability of misstepping detected by the deviation identification model and the higher the degree of deviation, the more automatically this factor will adjust the number of denoising steps and the final output torque amplitude of the torque determination model. This achieves the expected generation effect that the greater the deviation, the stronger the softening intervention, thus ensuring a precise match between the control response and the risk level.

[0095] Meanwhile, the deviation adjustment model, as the final decision fusion layer, constructs an adaptive fusion loss function. It not only optimizes the distance between the final control output and the target control curve (i.e., tracking accuracy) but also integrates multiple mutually constraining objectives. It introduces an intent retention loss to ensure that, in risk-free or low-risk scenarios, the system's gentle intervention does not weaken the driver's normal control intent. A gentleness penalty factor quantifies and suppresses drastic fluctuations in torque output, ensuring smoothness. Simultaneously, a stability evaluation index is incorporated to ensure the overall stability of the vehicle's dynamic response from a higher dimension. Thus, the deviation adjustment model can learn and maintain a dynamic optimal balance among the three conflicting dimensions of control accuracy, driver intent retention, and output smoothness, making the final torque command both safe and natural.

[0096] To improve the model's performance in critical hazardous scenarios, a sample weight adjustment mechanism is embedded throughout the training process to proactively increase the contribution of accidental stepping samples to the loss function. Simultaneously, a hard sample mining strategy is employed to prioritize training samples that fall on the boundary between accidental stepping and normal behavior, and are prone to model errors, thereby strengthening the model's ability to discriminate complex and ambiguous driving behaviors. At the optimization level, an end-to-end joint scheduling approach is adopted, and a gradient separation mechanism is used to prevent any single module (especially models with a large number of parameters) from dominating the training direction of the entire model, ensuring that all modules receive sufficient and balanced optimization. Furthermore, a dynamic inter-model synergy index is used to evaluate the consistency and effectiveness of the outputs of each sub-module in real time, providing a quantitative basis for monitoring and adjusting the training process. Throughout the entire training process, the weights of all modules remain updatable, thus achieving full-process, global, and interconnected optimization from the original input to the final control output.

[0097] In this embodiment, a conditional space perturbation sampling strategy is introduced for the torque determination model. Gaussian perturbation is applied to the current environment embedding vector to dynamically generate multiple sets of candidate torque outputs. By comparing the deviation between these outputs and the actual outputs, the residual correction path based on the score is optimized, thereby improving the generation robustness and control smoothness of the model under boundary conditions.

[0098] In this embodiment, an individualized error correction model is constructed by continuously tracking the driver's pedal correction behavior after the softening strategy is triggered, in order to determine the user's acceptance level of the current softening strategy. If the system detects that the driver repeatedly attempts to exceed the system limit by increasing the pedal opening, it will automatically adjust the softening intensity and optimize the weight distribution of the control output, making the model more in line with the individual's driving habits and expectations.

[0099] In this embodiment, all model parameter tuning behaviors and results are reported to the backend server through a log recording system, supporting regular cloud-based review, batch updates, and cross-vehicle model comparisons. A safety monitoring module is also embedded in the closed-loop optimization process. If the system experiences a significant error peak or triggers safety mechanisms such as the ESP system during a continuous control cycle, it will enter protection mode and forcibly revert to the default torque strategy to ensure safe vehicle operation.

[0100] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0101] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0102] Based on the same inventive concept, corresponding to any of the above-described embodiments, this disclosure also provides a vehicle control device.

[0103] refer to Figure 2 , Figure 2 The vehicle control device, as described in this embodiment, includes: The data acquisition module 201 is configured to detect changes in vehicle torque, acquire vehicle control data and environmental data, and determine target deviation characteristics based on the vehicle control data and the environmental data. The torque determination module 202 is configured to acquire the initial torque of the vehicle and determine the candidate torque of the vehicle based on the target deviation characteristics and the initial torque of the vehicle. The vehicle control module 203 is configured to acquire vehicle operating data, determine the vehicle target torque based on the pre-trained deviation adjustment model according to the vehicle operating data, the target deviation characteristics and the vehicle candidate torque, and control the vehicle driving according to the vehicle target torque.

[0104] In some embodiments, the vehicle control module 203 is specifically configured as follows: Based on the vehicle operation data and the target deviation characteristics, a target fusion vector is determined; The target fusion vector is input into a pre-trained bias adjustment model, and the target risk level is obtained by processing the bias adjustment model. The target torque of the vehicle is determined based on the target risk level and the candidate torque of the vehicle.

[0105] In some embodiments, the vehicle control module 203 is specifically configured as follows: The vehicle operation data and the target deviation features are mapped and aligned to obtain the operation data embedding vector and the deviation vector. Obtain the target fusion weights, input the target fusion weights, the running data embedding vector, and the deviation vector into a pre-trained vector fusion model, process the vector fusion model, and output the target fusion vector.

[0106] In some embodiments, the vehicle control module 203 is specifically configured as follows: In response to the target risk level being Level 1, the target torque of the vehicle is determined as the candidate torque of the vehicle; or... In response to the target risk level being level two, a target adjustment coefficient is determined based on the target deviation characteristics, and a vehicle target torque is determined based on the target adjustment coefficient and the vehicle's initial torque; or... In response to the target risk level being Level 3, the target torque of the vehicle is determined to be the initial torque of the vehicle.

[0107] In some embodiments, the vehicle control module 203 is specifically configured as follows: The target deviation features are input into a pre-trained deviation level determination model, and the target intention deviation level is obtained by processing the deviation level determination model. Based on the level of deviation from the stated target intent, determine the target adjustment coefficient; The target torque of the vehicle is obtained by multiplying the target adjustment coefficient and the initial torque of the vehicle.

[0108] In some embodiments, the vehicle control module 203 is specifically configured as follows: Acquire vehicle model data and driver identification; determine a first correction coefficient based on the vehicle model data; and determine a second correction coefficient based on the driver identification. An initial adjustment coefficient is determined based on the target intention deviation level, and the initial adjustment coefficient is corrected based on the first correction coefficient and the second correction coefficient to obtain the target adjustment coefficient.

[0109] In some embodiments, the vehicle control module 203 is specifically configured as follows: The first correction coefficient and the second correction coefficient are compared, and the maximum value of the first correction coefficient and the second correction coefficient is selected as the target correction coefficient; The target adjustment coefficient is obtained by multiplying the target correction coefficient and the initial adjustment coefficient.

[0110] In some embodiments, the torque determination module 202 is specifically configured as follows: Obtain the pre-trained torque determination model; The target deviation characteristics and the initial torque of the vehicle are input into the torque determination model, and the candidate torque of the vehicle is obtained through processing by the torque determination model.

[0111] For ease of description, the above apparatus is described in terms of its functions, divided into various modules. Of course, in implementing this disclosure, the functions of each module can be implemented in one or more software and / or hardware.

[0112] The apparatus of the above embodiments is used to implement the corresponding vehicle control method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0113] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the vehicle control method described in any of the above embodiments.

[0114] Figure 3 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0115] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0116] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0117] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0118] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0119] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0120] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0121] The electronic devices described above are used to implement the corresponding vehicle control methods in any of the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0122] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to execute the vehicle control method as described in any of the above embodiments.

[0123] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0124] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the vehicle control method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0125] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a vehicle, including the vehicle control device, the electronic device, and the computer-readable storage medium in the above embodiments, wherein the vehicle device implements the vehicle control method described in any of the above embodiments.

[0126] The vehicles described in the above embodiments are used to implement the vehicle control method described in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0127] It is understood that before using the technical solutions of the various embodiments in this disclosure, users will be informed of the type, scope of use, and usage scenarios of the personal information involved in an appropriate manner, and user authorization will be obtained.

[0128] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media performing the operations of this disclosed technical solution.

[0129] As an optional but not limited implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0130] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0131] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this disclosure as described above, which are not provided in detail for the sake of brevity.

[0132] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this disclosure, the provided drawings may or may not show well-known power / ground connections to integrated circuit (IC) chips and other components. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this disclosure, and this also takes into account the fact that the details of implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this disclosure will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that the embodiments of this disclosure can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0133] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0134] This disclosure is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A vehicle control method characterized by, include: The vehicle torque change is detected, vehicle control data and environmental data are acquired, and the target deviation characteristics are determined based on the vehicle control data and the environmental data. Obtain the initial torque of the vehicle, and determine the candidate torque of the vehicle based on the target deviation characteristics and the initial torque of the vehicle. The vehicle operation data is acquired, and the target torque of the vehicle is determined based on the pre-trained deviation adjustment model according to the vehicle operation data, the target deviation characteristics, and the candidate torque of the vehicle. The vehicle driving is controlled according to the target torque.

2. The method according to claim 1, characterized in that, The deviation adjustment model based on pre-training determines the vehicle target torque according to the vehicle operating data, the target deviation characteristics, and the vehicle candidate torque, including: Based on the vehicle operation data and the target deviation characteristics, a target fusion vector is determined; The target fusion vector is input into a pre-trained bias adjustment model, and the target risk level is obtained by processing the bias adjustment model. The target torque of the vehicle is determined based on the target risk level and the candidate torque of the vehicle.

3. The method according to claim 2, characterized in that, The step of determining the target fusion vector based on the vehicle operation data and the target deviation characteristics includes: The vehicle operation data and the target deviation features are mapped and aligned to obtain the operation data embedding vector and the deviation vector. Obtain the target fusion weights, input the target fusion weights, the running data embedding vector, and the deviation vector into a pre-trained vector fusion model, process the vector fusion model, and output the target fusion vector.

4. The method according to claim 2, characterized in that, Determining the target torque of the vehicle based on the target risk level and the candidate vehicle torques includes: In response to the target risk level being Level 1, the target torque of the vehicle is determined as the candidate torque of the vehicle; or... In response to the target risk level being level two, a target adjustment coefficient is determined based on the target deviation characteristics, and a vehicle target torque is determined based on the target adjustment coefficient and the vehicle's initial torque; or... In response to the target risk level being level three, the target torque of the vehicle is determined to be the initial torque of the vehicle. Based on the risk level intensity, from high risk to low risk, they are ranked as Level 1, Level 2, and Level 3.

5. The method according to claim 4, characterized in that, The step of determining the target adjustment coefficient based on the target deviation characteristics, and determining the vehicle target torque based on the target adjustment coefficient and the vehicle initial torque, includes: The target deviation features are input into a pre-trained deviation level determination model, and the target intention deviation level is obtained by processing the deviation level determination model. Based on the level of deviation from the stated target intent, determine the target adjustment coefficient; The target torque of the vehicle is obtained by multiplying the target adjustment coefficient and the initial torque of the vehicle.

6. The method according to claim 5, characterized in that, The step of determining the target adjustment coefficient based on the target intention deviation level includes: Obtain vehicle model data and driver identification; determine a first correction coefficient based on the vehicle model data; and determine a second correction coefficient based on the driver identification. An initial adjustment coefficient is determined based on the target intention deviation level, and the initial adjustment coefficient is corrected based on the first correction coefficient and the second correction coefficient to obtain the target adjustment coefficient.

7. The method according to claim 6, characterized in that, The step of correcting the initial adjustment coefficient based on the first correction coefficient and the second correction coefficient to obtain the target adjustment coefficient includes: The first correction coefficient and the second correction coefficient are compared, and the maximum value of the first correction coefficient and the second correction coefficient is selected as the target correction coefficient; The target adjustment coefficient is obtained by multiplying the target correction coefficient and the initial adjustment coefficient.

8. The method according to claim 1, characterized in that, The step of determining the candidate torque of the vehicle based on the target deviation characteristics and the initial torque of the vehicle includes: Obtain the pre-trained torque determination model; The target deviation characteristics and the initial torque of the vehicle are input into the torque determination model, and the candidate torque of the vehicle is obtained through processing by the torque determination model.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 8.

10. A vehicle, characterized in that, The vehicle includes the electronic equipment as described in claim 9.

Images