Vehicle control method, device, medium, electronic equipment and vehicle

Abstract

The present disclosure belongs to the technical field of vehicle control, and relates to a vehicle control method, device, medium, electronic equipment and vehicle. The method comprises: acquiring state data of a vehicle; using a first time series model to perform prediction processing on the state data to obtain a target adhesion coefficient, and controlling the vehicle to operate according to the target adhesion coefficient, wherein the first time series model learns the knowledge of a second time series model through knowledge distillation, and the second time series model has long time series modeling capability. The present disclosure uses the first time series model obtained through knowledge distillation to predict the target adhesion coefficient, solves the problem of poor modeling accuracy in a long period working condition in the related art, also improves the anti-interference capability and generalization capability of the first time series model when predicting the target adhesion coefficient, and improves the prediction accuracy and robustness of the first time series model when facing untrained complex working conditions.

Description

Technical Field

[0001] This disclosure relates to the field of vehicle control technology, and in particular to a vehicle control method, a vehicle control device, a computer-readable storage medium, an electronic device, and a vehicle. Background Technology

[0002] The coefficient of friction (COP) is a key parameter that directly defines the maximum friction threshold that can be generated between the tire and the road surface. Therefore, accurate estimation of the COP is of significant practical importance and is a critical problem that urgently needs to be solved. Summary of the Invention

[0003] To overcome the problems existing in the related technologies, this disclosure provides a vehicle control method, a vehicle control device, a computer-readable storage medium, an electronic device, and a vehicle.

[0004] According to a first aspect of the present disclosure, a vehicle control method is provided, comprising: Obtain vehicle status data; The target adhesion coefficient is obtained by predicting the state data using a first time series model, and the vehicle operation is controlled according to the target adhesion coefficient. The first time series model learns the knowledge of the second time series model through knowledge distillation, and the second time series model has long time series modeling capability.

[0005] In some possible implementations, the step of using a first time-series model to predict the state data to obtain the target adhesion coefficient includes: Statistical features are determined based on the state data, and the state data and statistical features are standardized to obtain standardized state data and statistical features. The feature sequence is determined based on the standardized state data and statistical characteristics, and the feature sequence is input into the trained first time series model to obtain the target adhesion coefficient.

[0006] In some possible implementations, the method further includes: Before inputting the feature sequence into the trained first time-series model to obtain the target attachment coefficient, sequence samples and corresponding coefficient samples are obtained, and the sequence samples are input into the second time-series model to obtain soft label coefficients; The sequence samples are input into the first time series model to be trained to obtain prediction coefficients, and the first model features of the first time series model to be trained and the second model features of the second time series model are obtained. The target loss is determined based on the coefficient samples, the soft label coefficients, the predicted coefficients, the first model features, and the second model features, and the target loss is used to train the first time series model to be trained.

[0007] In some possible implementations, determining the target loss based on the coefficient samples, the soft-label coefficients, the predicted coefficients, the first model features, and the second model features includes: The task loss is determined based on the predicted coefficients and the coefficient samples, and the first distillation loss is determined based on the predicted coefficients and the soft label coefficients; The second distillation loss is determined based on the first model feature and the second model feature, and the target loss is determined based on the task loss, the first distillation loss, and the second distillation loss.

[0008] In some possible implementations, the step of inputting the feature sequence into a trained first temporal model to obtain the target adhesion coefficient includes: The feature sequence is input into the trained first time series model to obtain the current adhesion coefficient, and the historical adhesion coefficient is obtained. The target adhesion coefficient is obtained by performing median filtering on the current adhesion coefficient and the historical adhesion coefficient.

[0009] In some possible implementations, the step of performing median filtering on the current adhesion coefficient and the historical adhesion coefficient to obtain the target adhesion coefficient includes: The current adhesion coefficient and the historical adhesion coefficient are sorted to obtain the sorting result; The target adhesion coefficient is determined from the current adhesion coefficient and the historical adhesion coefficient based on the sorting results.

[0010] In some possible implementations, the first time-series model includes a long short-term memory network, and the second time-series model includes a converter model.

[0011] According to a second aspect of the present disclosure, a vehicle control device is provided, comprising: The data acquisition module is configured to acquire vehicle status data; The coefficient prediction module is configured to use a first time series model to predict the state data to obtain a target adhesion coefficient, and control the vehicle operation according to the target adhesion coefficient. The first time series model learns the knowledge of the second time series model through knowledge distillation, and the second time series model has long time series modeling capability.

[0012] In some possible implementations, the coefficient prediction module is configured as follows: Statistical features are determined based on the state data, and the state data and statistical features are standardized to obtain standardized state data and statistical features. The feature sequence is determined based on the standardized state data and statistical characteristics, and the feature sequence is input into the trained first time series model to obtain the target adhesion coefficient.

[0013] In some possible implementations, the vehicle control device is further configured to: Before inputting the feature sequence into the trained first time-series model to obtain the target attachment coefficient, sequence samples and corresponding coefficient samples are obtained, and the sequence samples are input into the second time-series model to obtain soft label coefficients; The sequence samples are input into the first time series model to be trained to obtain prediction coefficients, and the first model features of the first time series model to be trained and the second model features of the second time series model are obtained. The target loss is determined based on the coefficient samples, the soft label coefficients, the predicted coefficients, the first model features, and the second model features, and the target loss is used to train the first time series model to be trained.

[0014] In some possible implementations, the vehicle control device is further configured to: The task loss is determined based on the predicted coefficients and the coefficient samples, and the first distillation loss is determined based on the predicted coefficients and the soft label coefficients; The second distillation loss is determined based on the first model feature and the second model feature, and the target loss is determined based on the task loss, the first distillation loss, and the second distillation loss.

[0015] According to a third aspect of the present disclosure, a computer-readable storage medium is provided that stores computer program instructions thereon, which, when executed by a processor, implement the steps of the vehicle control method provided in any of the first aspects of the present disclosure.

[0016] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute the executable instructions to implement the vehicle control method provided in any of the first aspects of this disclosure.

[0017] According to a fifth aspect of the present disclosure, a vehicle is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute executable instructions stored in the memory to implement the steps of the vehicle control method provided in any of the first aspects of this disclosure.

[0018] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: the first time-series model obtained by knowledge distillation predicts the target adhesion coefficient, which solves the problem of poor modeling accuracy under long-cycle working conditions in related technologies, and also improves the anti-interference ability and generalization ability of the first time-series model in predicting the target adhesion coefficient, and improves the prediction accuracy and robustness of the first time-series model when facing untrained complex working conditions.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0021] Figure 1 This is a flowchart illustrating a vehicle control method according to an exemplary embodiment.

[0022] Figure 2 This is a flowchart illustrating a method for determining a feature sequence based on state data according to an exemplary embodiment.

[0023] Figure 3 This is a flowchart illustrating a method for training a first time series model according to an exemplary embodiment.

[0024] Figure 4 This is a flowchart illustrating a method for further determining target loss according to an exemplary embodiment.

[0025] Figure 5 This is a flowchart illustrating a method for determining a target adhesion coefficient using a first time-series model, according to an exemplary embodiment.

[0026] Figure 6 This is a flowchart illustrating a method for determining the target adhesion coefficient by median filtering according to an exemplary embodiment.

[0027] Figure 7 This is a flowchart illustrating a vehicle control method in an application scenario according to an exemplary embodiment.

[0028] Figure 8 This is a schematic diagram illustrating the process of training an LSTM model in an application scenario according to an exemplary embodiment.

[0029] Figure 9 This is a block diagram illustrating a vehicle control device according to an exemplary embodiment.

[0030] Figure 10This is a block diagram illustrating a vehicle according to an exemplary embodiment.

[0031] Figure 11 This is a block diagram illustrating a vehicle control device according to an exemplary embodiment.

[0032] Figure 12 This is a block diagram illustrating a vehicle control device according to an exemplary embodiment. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0034] It should be noted that all actions involving the acquisition of signals, information, or data in this disclosure are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with authorization from the owner of the relevant device.

[0035] More and more cars are now equipped with Advanced Driver Assistance Systems (ADAS), which effectively reduce the workload of drivers. This system uses a multi-modal sensor fusion architecture, integrating core functional modules such as Adaptive Cruise Control (ACC), Anti-lock Braking System (ABS), Traction Control System (TCS), and Electronic Stability Program (ESP) to form a multi-dimensional active safety network. This greatly reduces the driver's workload and provides strong protection for vehicle driving safety. The control efficiency of the ADAS system is closely related to the accurate estimation of the road surface adhesion coefficient.

[0036] The coefficient of friction, a key parameter, directly defines the maximum friction threshold that can be generated between the tire and the road surface. For example, on loose gravel roads, the control algorithm needs to suppress the torque of the drive wheels in real time to prevent excessive slippage. It also needs to coordinate with the air suspension to increase the vehicle height and enhance the damping coefficient of the shock absorbers. By increasing the tire's contact projection area and suppressing high-frequency bouncing, it maintains effective contact between the tire and the road surface. Since the coefficient of friction varies significantly depending on the road surface type, such as the dry and wet state of asphalt, the compaction degree of snow, and the gravel cover, using a fixed threshold control strategy will lead to problems such as slip ratio overshoot in the braking system and yaw moment imbalance in the steering system. This can result in safety hazards such as vehicle sideslip, steering inaccuracy, and even multi-axle coupling instability.

[0037] Therefore, accurate estimation of the road surface adhesion coefficient is of great practical significance and is a key problem that urgently needs to be solved.

[0038] Figure 1 This is a flowchart illustrating a vehicle control method according to an exemplary embodiment, such as... Figure 1 As shown, the vehicle control method used in the terminal includes the following steps: In step S110, the vehicle's status data is acquired.

[0039] In step S120, the target adhesion coefficient is obtained by predicting the state data using the first time series model, and the vehicle operation is controlled according to the target adhesion coefficient. The first time series model learns the knowledge of the second time series model through knowledge distillation, and the second time series model has long time series modeling capability.

[0040] In the exemplary embodiments of this disclosure, the first time-series model obtained by knowledge distillation is used to predict the target adhesion coefficient, which solves the problem of poor modeling accuracy under long-cycle conditions in related technologies. It also improves the anti-interference ability and generalization ability of the first time-series model when predicting the target adhesion coefficient, and improves the prediction accuracy and robustness of the first time-series model when facing complex untrained conditions.

[0041] The following is a detailed explanation of each step in the vehicle control method.

[0042] In step S110, the vehicle's status data is acquired.

[0043] In an exemplary embodiment of this disclosure, when acquiring vehicle status data, sampling can be performed at a fixed time interval of 10ms to address the issue of differences in data acquisition frequencies among different sensors.

[0044] The acquired status data includes information such as brake pedal travel, steering wheel angle, yaw rate, wheel speed, wheel clamping force, motor torque, acceleration, ABS activation flag, VDC (Vehicle Dynamics Control) activation flag, and TCS activation flag.

[0045] In step S120, the target adhesion coefficient is obtained by predicting the state data using the first time series model, and the vehicle operation is controlled according to the target adhesion coefficient. The first time series model learns the knowledge of the second time series model through knowledge distillation, and the second time series model has long time series modeling capability.

[0046] In an exemplary embodiment of this disclosure, after obtaining the vehicle's state data, a feature sequence can be determined based on the state data.

[0047] In an optional embodiment, Figure 2 A flowchart illustrating a method for determining feature sequences based on state data is shown, such as... Figure 2 As shown, the method may include at least the following steps: In step S210, statistical features are determined based on the state data, and the state data and statistical features are standardized to obtain standardized state data and statistical features.

[0048] When constructing statistical features, sliding features can be extracted using a fixed time window, and key statistics can be calculated as statistical features. These statistical features include the speed difference between the left and right wheels, the speed difference between the front and rear wheels, and the mean and variance of the four wheel speeds. For example, the size of the fixed time window can be 200ms or 500ms, and the sliding step size can be 10ms. This exemplary embodiment does not impose specific limitations on the window size and step size.

[0049] Furthermore, the state data and statistical characteristics are standardized to unify the scale and eliminate dimensional differences.

[0050] In step S220, the feature sequence is determined based on the standardized state data and statistical characteristics, and the feature sequence is input into the trained first time series model to obtain the target adhesion coefficient.

[0051] The standardized state data and statistical features are slidably generated in temporal order to form a feature sequence including continuous time steps, which is the input required for the first time series model in the format of [number of samples, time step, number of features]. Here, the number of samples can be the total number of state data and statistical features, the time step can be 10ms, and the number of features can be all features included in the feature sequence. Table 1 shows the number of features in the feature sequence:

[0052] Table 1 Before inputting the feature sequence into the first time series model, the first time series model can be trained first.

[0053] In an optional embodiment, Figure 3 A flowchart illustrating the method for training the first time-series model is shown, such as... Figure 3 As shown, the method may include at least the following steps: In step S310, before inputting the feature sequence into the trained first time-series model to obtain the target attachment coefficient, sequence samples and corresponding coefficient samples are obtained, and the sequence samples are input into the second time-series model to obtain the soft label coefficient.

[0054] In an optional embodiment, the first time-series model includes a long short-term memory network, and the second time-series model includes a converter model.

[0055] Transformer models, with their self-attention mechanism, have a natural advantage in capturing long-range dependencies in long-term time-series data, making them a core choice for building teacher models. When the second time-series model is a Transformer model, the sequence samples are input into the Transformer model to obtain soft-label coefficients. During training, HuberLoss is used as the loss function. HuberLoss reuses the MSE (Mean Square Error) property when the error is small to ensure the fitting accuracy of normal time-series data; when the error is large, it switches to the MAE (Mean Absolute Error) property to effectively reduce the interference of outliers on model training, balancing stability and noise resistance.

[0056] In step S320, the sequence samples are input into the first time series model to be trained to obtain the prediction coefficients, and the first model features of the first time series model to be trained and the second model features of the second time series model are obtained.

[0057] When the first time series model is an LSTM (Long Short-Term Memory) model, the sequence samples are synchronously input into the LSTM model to obtain the prediction coefficients.

[0058] When sequence samples are synchronously input into the Transformer teacher model and the LSTM student model, key intermediate features of the teacher model are extracted, such as the self-attention weight matrix of the encoder layer (reflecting the dependence strength between temporal nodes) and the feature mapping vector (containing an abstract representation of long-range patterns), which serve as feature templates that the LSTM needs to emulate.

[0059] In step S330, the target loss is determined based on the coefficient samples, soft label coefficients, prediction coefficients, first model features, and second model features, so as to use the target loss to train the first time series model to be trained.

[0060] In an optional embodiment, Figure 4 A flowchart illustrating the method for further determining the target loss is shown, such as... Figure 4 As shown, the method may include at least the following steps: in step S410, the task loss is determined based on the prediction coefficients and coefficient samples, and the first distillation loss is determined based on the prediction coefficients and soft label coefficients.

[0061] The HuberLoss of the predicted coefficients and coefficient samples can be used as the task loss, and the MSE of the predicted coefficients and soft-label coefficients can be used as the first distillation loss.

[0062] In step S420, the second distillation loss is determined based on the first model features and the second model features, and the target loss is determined based on the task loss, the first distillation loss and the second distillation loss.

[0063] The cosine similarity loss of the first model features and the second model features is determined as the second distillation loss.

[0064] After determining the task loss, the first distillation loss, and the second distillation loss, the target loss can be determined by weighted summation. This target loss = α × task loss + β × first distillation loss + γ × second distillation loss, where α + β + γ = 1.

[0065] Among them, the three types of loss that constitute the target loss enable the LSTM model to perform long-term modeling kernels of the deep learning teacher model under a lightweight architecture, significantly improving the generalization ability in complex scenarios.

[0066] After training the first time-series model, the target adhesion coefficient can be determined using the trained first time-series model.

[0067] In the optional implementation, Figure 5 A flowchart illustrating the method for determining the target adhesion coefficient using a first time-series model is shown, such as... Figure 5 As shown, the method may include at least the following steps: In step S510, the feature sequence is input into the trained first time series model to obtain the current adhesion coefficient, and the historical adhesion coefficient is obtained.

[0068] Considering that the current attachment coefficient obtained by inputting the feature sequence into the trained first time series model has a spike phenomenon, historical attachment coefficients can be obtained and used to filter the current attachment coefficient.

[0069] The current adhesion coefficient can be obtained through a fixed window size. This fixed window size can be 5 time steps, or other sizes can be set according to actual conditions; this exemplary embodiment does not impose any special limitations on this. Therefore, when processing the current adhesion coefficient, historical adhesion coefficients within the window can be extracted, and this extraction process includes the current adhesion coefficient.

[0070] In step S520, the target adhesion coefficient is obtained by median filtering of the current adhesion coefficient and the historical adhesion coefficient.

[0071] In an optional embodiment, Figure 6 A flowchart illustrating a method for determining the target adhesion coefficient using median filtering is shown, as follows: Figure 6 As shown, the method may include at least the following steps: In step S610, the current adhesion coefficient and the historical adhesion coefficient are sorted to obtain a sorting result.

[0072] The current adhesion coefficient and historical adhesion coefficients can be sorted according to their numerical values ​​to obtain the sorting results.

[0073] In step S620, the target adhesion coefficient is determined from the current adhesion coefficient and the historical adhesion coefficient based on the sorting results.

[0074] After obtaining the ranking results, the median value of the current adhesion coefficient and the historical adhesion coefficient involved in the ranking results can be used as the target adhesion coefficient. This method can effectively remove glitch in the prediction results, retain the true trend of data change, and make the model prediction results more consistent with the actual changes in road adhesion coefficient.

[0075] The vehicle control method in this embodiment will be described in detail below with reference to an application scenario.

[0076] Figure 7 A flowchart illustrating a vehicle control method in an application scenario is shown, such as... Figure 7 As shown, in step S710, sensor information is acquired.

[0077] It acquires information such as brake pedal travel, steering wheel angle, yaw rate, wheel speed, wheel clamping force, motor torque, acceleration, ABS activation flag, VDC activation flag, and TCS activation flag. Data is collected at fixed intervals of 10ms to address the issue of varying data acquisition frequencies from different sensors.

[0078] In step S720, time-series features are constructed.

[0079] This feature is mainly divided into raw collected features and constructed statistical features. During construction, sliding features are first extracted using a fixed time window, and key statistics can be calculated, including the speed difference between the left and right wheels, the speed difference between the front and rear wheels, and the mean and variance of the four wheel speeds.

[0080] Furthermore, all features are standardized to unify the scale and eliminate dimensional differences.

[0081] Finally, the feature sequence containing continuous time steps is generated by sliding in chronological order, forming the input in the format of [number of samples, time step, number of features] required by the LSTM model.

[0082] In step S730, the Transformer teacher model is constructed and trained.

[0083] Transformer models, with their self-attention mechanism, have a natural advantage in capturing long-range dependencies in long-term time-series data, making them a core choice for building teacher models.

[0084] During training, HuberLoss is used as the loss function: when the error is small, it reuses the MSE feature to ensure the fitting accuracy of normal time series data; when the error is large, it switches to the MAE feature to effectively reduce the interference of outliers on model training, thus balancing stability and noise resistance.

[0085] In terms of model design, the direct learning capability of temporal patterns is enhanced through an encoder-decoder structure, providing a high-performance teacher model foundation for subsequent knowledge distillation.

[0086] In step S740, the LSTM time series model is trained.

[0087] To enable the lightweight LSTM model to inherit the long-term temporal modeling capability of the Transformer teacher model, a knowledge distillation strategy is adopted to enhance training. The core is to realize the transfer of high-order temporal cognition of the teacher model through feature distillation.

[0088] First, we focus on knowledge transfer at the feature level. Raw time-series data is synchronously input into both the Transformer teacher model and the LSTM student model to extract key intermediate features from the teacher model, such as the self-attention weight matrix and feature mapping vector of the encoder layer. Simultaneously, we retain the soft labels output by the teacher model as the target for output layer distillation.

[0089] Figure 8 The diagram illustrates the process of training an LSTM model in an application scenario, as shown below. Figure 8As shown, a three-layer hybrid loss function is designed to drive LSTM training: Total loss = α × Task loss (Huber Loss of LSTM prediction and real label) + β × Output distillation loss (MSE of LSTM prediction and teacher soft label) + γ × Feature distillation loss (cosine similarity loss between LSTM hidden state and teacher intermediate features, α+β+γ=1).

[0090] Among them, the feature distillation loss forces the LSTM to learn the teacher's hierarchical processing logic for temporal features, such as the weight allocation of long-range dependencies and the feature enhancement of key nodes. The output distillation loss enhances prediction consistency, and the task loss ensures basic accuracy. The three work together to enable the LSTM to deeply learn the teacher's long-term temporal modeling kernel under a lightweight architecture, significantly improving the generalization performance in complex scenarios.

[0091] In step S750, the prediction result is subjected to median filtering.

[0092] Considering the presence of glitches in the model's prediction results, median filtering can stably and effectively improve overall performance. Median filtering sorts the data within a certain window and selects the median value as the filter output. It corrects the current time-point data by saving the prediction results from a fixed time window prior to the current time point.

[0093] Specifically, a fixed window size is first set, for example, 5 time steps, with an odd number to facilitate taking the median value. In the initial stage, when the amount of data has not reached the window size, the original predicted values ​​are used, and the prediction results of each step are cached synchronously. When processing the current time point, the historical predicted values ​​within the window (including the current point) are extracted, sorted, and the median value is taken as the prediction result. Then the cache is updated (keeping the latest window data), and the next time point is processed in a sliding window manner.

[0094] This method can effectively remove burrs in the prediction results, retain the true trend of data change, and make the model prediction results more consistent with the actual changes in the road surface adhesion coefficient.

[0095] In step S760, the final predicted road adhesion coefficient is output.

[0096] After outputting the final predicted road adhesion coefficient, the results can be integrated into systems such as adaptive cruise control, anti-lock braking system, traction control system, and electronic stability program.

[0097] In adaptive cruise control, a precise road adhesion coefficient allows the vehicle to adjust speed and distance more accurately based on actual road conditions. When facing slippery surfaces, the system can reduce speed in advance and increase the safe distance from the vehicle in front based on an accurate adhesion coefficient, avoiding rear-end collisions caused by prolonged braking distance. When the anti-lock braking system (ABS) is working, an accurate estimation of the road adhesion coefficient allows the system to control braking force more rationally. On surfaces with different adhesion coefficients, such as icy or gravel roads, the ABS can intelligently adjust braking pressure based on the accurate coefficient provided by this solution to prevent wheel lock-up and ensure good vehicle handling during braking. For traction control systems, an accurate road adhesion coefficient helps the system distribute power rationally. When starting or accelerating on low-traction surfaces, the TCS can prevent excessive wheel slippage based on the currently estimated accurate road adhesion coefficient, achieving smooth and efficient power output. Electronic stability program (ESP) relies on a precise road adhesion coefficient to monitor the vehicle's driving status in real time. In extreme situations such as emergency avoidance, ESP applies different braking forces to each wheel precisely based on the derived coefficient, correcting the vehicle's trajectory and ensuring driving safety.

[0098] Therefore, this solution specifically addresses the problems of long-term information loss and poor generalization performance in traditional LSTM road adhesion coefficient estimation schemes. The core of this solution leverages the strong modeling capabilities and advantages of the Transformer teacher model in long-term prediction problems, empowering LSTM training through knowledge distillation. The Transformer model can deeply mine the evolution patterns of dynamic features such as wheel speed, torque, and acceleration in continuous time-series data, accurately identify the dynamic patterns of wheels under different road conditions such as wet, icy, and dry conditions, and effectively filter instantaneous noise, providing high-order temporal knowledge templates for LSTM.

[0099] This approach transfers multiple capabilities to the LSTM through distillation: First, it captures long-term, cross-time-step correlations, such as the causal relationship between the sudden drop in wheel speed during initial slippage on wet roads and subsequent torque adjustments, addressing the issue of traditional LSTMs easily forgetting early crucial information, such as the micro-slip signal during icy starts, thus enhancing the ability to capture long-term correlations. Second, it transmits real-time dynamic feature weight adjustment logic, such as amplifying the impact of wheel speed fluctuations on snowy roads and strengthening the priority of torque response on dry roads, improving estimation bias in mixed road conditions and achieving dynamic adaptation to diverse operating conditions. Third, it empowers the LSTM to identify and filter instantaneous noise, such as false wheel speed jumps over speed bumps and abnormal sensor values, comprehensively improving anti-interference and generalization capabilities. Furthermore, the optimized LSTM student model has a parameter count that is essentially the same as the traditional LSTM, requiring no additional computing power. This solution can provide highly reliable adhesion coefficient data for ADAS systems, such as ACC adaptive cruise control, ABS anti-lock braking, and TCS traction control, helping assisted driving functions more accurately predict changes in road grip and effectively improving vehicle driving safety in complex road conditions.

[0100] This approach not only effectively compensates for the shortcomings of traditional LSTM in long-sequence modeling, but also accurately captures causal relationships across time steps within long periods, such as the long-term evolution patterns of data like wheel speed and torque, by leveraging the Transformer teacher model. This long-term time-series modeling capability is then transferred to LSTM through distillation, significantly improving prediction accuracy under long-term operating conditions. Furthermore, it comprehensively optimizes feature adaptability, anti-interference ability, and generalization performance. The Transformer can dynamically adjust feature weights according to different road surface scenarios and actively filter out noise interference such as road bumps. These capabilities are inherited by LSTM through the distillation process, overcoming the problem of rigid feature weights in traditional models and reducing LSTM's sensitivity to outliers. Simultaneously, by learning underlying general temporal patterns, the algorithm's robustness in real-world complex operating conditions is further enhanced.

[0101] In the exemplary embodiments of this disclosure, the first time-series model obtained by knowledge distillation is used to predict the target adhesion coefficient, which solves the problem of poor modeling accuracy under long-cycle conditions in related technologies. It also improves the anti-interference ability and generalization ability of the first time-series model when predicting the target adhesion coefficient, and improves the prediction accuracy and robustness of the first time-series model when facing complex untrained conditions.

[0102] Furthermore, in an exemplary embodiment of this disclosure, a vehicle control device is also provided. Figure 9 This is a block diagram illustrating a vehicle control device according to an exemplary embodiment. (Refer to...) Figure 9 The vehicle control device 900 includes a data acquisition module 910 and a coefficient prediction module 920.

[0103] Data acquisition module 910 is configured to acquire vehicle status data; The coefficient prediction module 920 is configured to use a first time series model to predict the state data to obtain a target adhesion coefficient, and control the vehicle operation according to the target adhesion coefficient. The first time series model learns the knowledge of the second time series model through knowledge distillation, and the second time series model has long time series modeling capability.

[0104] In some embodiments of this disclosure, the coefficient prediction module 920 is configured to: Statistical features are determined based on the state data, and the state data and statistical features are standardized to obtain standardized state data and statistical features. The feature sequence is determined based on the standardized state data and statistical characteristics, and the feature sequence is input into the trained first time series model to obtain the target adhesion coefficient.

[0105] In some embodiments of this disclosure, the vehicle control device 900 is further configured to: Before inputting the feature sequence into the trained first time-series model to obtain the target attachment coefficient, sequence samples and corresponding coefficient samples are obtained, and the sequence samples are input into the second time-series model to obtain soft label coefficients; The sequence samples are input into the first time series model to be trained to obtain prediction coefficients, and the first model features of the first time series model to be trained and the second model features of the second time series model are obtained. The target loss is determined based on the coefficient samples, the soft label coefficients, the predicted coefficients, the first model features, and the second model features, and the target loss is used to train the first time series model to be trained.

[0106] In some embodiments of this disclosure, the vehicle control device 900 is further configured to: The task loss is determined based on the predicted coefficients and the coefficient samples, and the first distillation loss is determined based on the predicted coefficients and the soft label coefficients; The second distillation loss is determined based on the first model feature and the second model feature, and the target loss is determined based on the task loss, the first distillation loss, and the second distillation loss.

[0107] In some embodiments of this disclosure, the coefficient prediction module 920 is configured to: The feature sequence is input into the trained first time series model to obtain the current adhesion coefficient, and the historical adhesion coefficient is obtained. The target adhesion coefficient is obtained by performing median filtering on the current adhesion coefficient and the historical adhesion coefficient.

[0108] In some embodiments of this disclosure, the coefficient prediction module 920 is configured to: The current adhesion coefficient and the historical adhesion coefficient are sorted to obtain the sorting result; The target adhesion coefficient is determined from the current adhesion coefficient and the historical adhesion coefficient based on the sorting results.

[0109] In some embodiments of this disclosure, the first time-series model includes a long short-term memory network, and the second time-series model includes a converter model.

[0110] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0111] This disclosure also provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the steps of the vehicle control method provided in this disclosure.

[0112] Figure 10 This is a block diagram illustrating a vehicle 1000 according to an exemplary embodiment. For example, vehicle 1000 may be a hybrid vehicle, a non-hybrid vehicle, an electric vehicle, a fuel cell vehicle, or other types of vehicle. Vehicle 1000 may have driver assistance functions.

[0113] Reference Figure 10 The vehicle 1000 may include various subsystems, such as an infotainment system 1010, a perception system 1020, a decision control system 1030, a drive system 1040, and a computing platform 1050. The vehicle 1000 may also include more or fewer subsystems, and each subsystem may include multiple components. Furthermore, each subsystem and component of the vehicle 1000 can be interconnected via wired or wireless means.

[0114] In some embodiments, the infotainment system 1010 may include a communication system, an entertainment system, and a navigation system, etc.

[0115] The perception system 1020 may include several types of sensors for sensing information about the environment surrounding the vehicle 1000. For example, the perception system 1020 may include a global positioning system (which may be a GPS system, a BeiDou system, or another positioning system), an inertial measurement unit (IMU), a lidar, a millimeter-wave radar, an ultrasonic radar, and a camera device.

[0116] The decision control system 1030 may include a computing system, a vehicle controller, a steering system, a throttle, and a braking system.

[0117] The drive system 1040 may include components that provide powered motion to the vehicle 1000. In one embodiment, the drive system 1040 may include an engine, an energy source, a transmission system, and wheels. The engine may be one or a combination of internal combustion engines, electric motors, and compressed air engines. The engine is capable of converting energy provided by the energy source into mechanical energy.

[0118] Some or all of the functions of the vehicle 1000 are controlled by a computing platform 1050. The computing platform 1050 may include at least one processor 1051 and a memory 1052, the processor 1051 being able to execute instructions 1053 stored in the memory 1052.

[0119] The processor 1051 can be any conventional processor, such as a commercially available CPU. The processor may also include graphics processing units (GPUs), field-programmable gate arrays (FPGAs), systems on chips (SoCs), application-specific integrated circuits (ASICs), or combinations thereof.

[0120] The memory 1052 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0121] In addition to instruction 1053, memory 1052 can also store data, such as road maps, route information, vehicle position, direction, speed, and other data. The data stored in memory 1052 can be used by computing platform 1050.

[0122] In this embodiment of the disclosure, the processor 1051 may execute instructions 1053 to complete all or part of the steps of the vehicle control method described above.

[0123] Figure 11 This is a block diagram illustrating a vehicle control device 1100 according to an exemplary embodiment. For example, device 1100 may be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0124] Reference Figure 11The device 1100 may include one or more of the following components: a first processing component 1102, a first memory 1104, a first power supply component 1106, a multimedia component 1108, an audio component 1110, a first input / output interface 1112, a sensor component 1114, and a communication component 1116.

[0125] The first processing component 1102 typically controls the overall operation of the device 1100, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. The first processing component 1102 may include one or more first processors 1120 to execute instructions to complete all or part of the steps of the methods described above. Furthermore, the first processing component 1102 may include one or more modules to facilitate interaction between the first processing component 1102 and other components. For example, the first processing component 1102 may include a multimedia module to facilitate interaction between the multimedia component 1108 and the first processing component 1102.

[0126] The first memory 1104 is configured to store various types of data to support the operation of the device 1100. Examples of such data include instructions for any application or method operating on the device 1100, contact data, phonebook data, messages, pictures, videos, etc. The first memory 1104 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0127] The first power supply component 1106 provides power to various components of the device 1100. The first power supply component 1106 may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device 1100.

[0128] Multimedia component 1108 includes a screen that provides an output interface between the device 1100 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 1108 includes a front-facing camera and / or a rear-facing camera. When the device 1100 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0129] Audio component 1110 is configured to output and / or input audio signals. For example, audio component 1110 includes a microphone (MIC) configured to receive external audio signals when device 1100 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in first memory 1104 or transmitted via communication component 1116. In some embodiments, audio component 1110 also includes a speaker for outputting audio signals.

[0130] The first input / output interface 1112 provides an interface between the first processing component 1102 and the peripheral interface module, which may be a keyboard, click wheel, buttons, etc. These buttons may include, but are not limited to, a home button, volume buttons, a start button, and a lock button.

[0131] Sensor assembly 1114 includes one or more sensors for providing status assessments of various aspects of device 1100. For example, sensor assembly 1114 may detect the on / off state of device 1100, the relative positioning of components such as the display and keypad of device 1100, changes in the position of device 1100 or a component of device 1100, the presence or absence of user contact with device 1100, the orientation or acceleration / deceleration of device 1100, and temperature changes of device 1100. Sensor assembly 1114 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 1114 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 1114 may also include an accelerometer, a gyroscope, a magnetometer, a pressure sensor, or a temperature sensor.

[0132] Communication component 1116 is configured to facilitate wired or wireless communication between device 1100 and other devices. Device 1100 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 1116 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 1116 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0133] In an exemplary embodiment, the apparatus 1100 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0134] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a first memory 1104 including instructions, which can be executed by a first processor 1120 of the device 1100 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0135] In another exemplary embodiment, a computer program product is also provided, the computer program product comprising a computer program executable by a programmable device, the computer program having a code portion for performing the vehicle control method described above when executed by the programmable device.

[0136] Figure 12 This is a block diagram illustrating a vehicle control device 1200 according to an exemplary embodiment. For example, the device 1200 may be provided as a server. (Refer to...) Figure 12 The apparatus 1200 includes a second processing component 1222, which further includes one or more processors, and memory resources represented by a second memory 1232 for storing instructions, such as application programs, that can be executed by the second processing component 1222. The application programs stored in the second memory 1232 may include one or more modules, each corresponding to a set of instructions. Furthermore, the second processing component 1222 is configured to execute instructions to perform the methods described above.

[0137] The device 1200 may also include a second power supply component 1226 configured to perform power management of the device 1200, a wired or wireless network interface 1250 configured to connect the device 1200 to a network, and a second input / output interface 1258. The device 1200 can operate on an operating system stored in a second memory 1232.

[0138] Furthermore, the term “exemplary” is used herein to mean serving as an example, instance, or illustration. Any aspect or design described herein as “exemplary” is not necessarily to be construed as advantageous compared to other aspects or designs. Rather, the use of the term “exemplary” is intended to present the concept in a concrete manner. As used herein, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from the context, “X applies A or B” is intended to mean any of the natural inclusive arrangements. That is, “X applies A or B” satisfies any of the foregoing instances if X applies A; X applies B; or both X applies A and B. Additionally, unless otherwise specified or clear from the context to refer to the singular form, the articles “a” and “an” as used in this application and the appended claims are generally understood to mean “one or more.”

[0139] Similarly, although this disclosure has been shown and described with respect to one or more implementations, equivalent variations and modifications will occur to those skilled in the art upon reading and understanding this specification and the accompanying drawings. This disclosure includes all such modifications and variations and is limited only by the scope of the claims. In particular, with respect to the various functions performed by the components described above (e.g., elements, resources, etc.), unless otherwise indicated, the terminology used to describe such components is intended to correspond to any component (functionally equivalent) that performs the specific function of the described component, even if structurally not equivalent to the disclosed structure. Furthermore, although specific features of this disclosure may have been disclosed with respect to only one of several implementations, such features may be combined with one or more other features of other implementations, as may be desired and advantageous to any given or particular application. Moreover, with regard to the terms “comprising,” “owning,” “having,” “having,” or variations thereof as used in the detailed description or claims, such terms are intended to be inclusive in a manner similar to the term “including.”

[0140] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

[0141] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A vehicle control method, characterized in that, include: Obtain vehicle status data; The target adhesion coefficient is obtained by predicting the state data using a first time series model, and the vehicle operation is controlled according to the target adhesion coefficient. The first time series model learns the knowledge of the second time series model through knowledge distillation, and the second time series model has long time series modeling capability.

2. The vehicle control method according to claim 1, characterized in that, The step of using a first time-series model to predict the state data to obtain the target adhesion coefficient includes: Statistical features are determined based on the state data, and the state data and statistical features are standardized to obtain standardized state data and statistical features. The feature sequence is determined based on the standardized state data and statistical characteristics, and the feature sequence is input into the trained first time series model to obtain the target adhesion coefficient.

3. The vehicle control method according to claim 2, characterized in that, The method further includes: Before inputting the feature sequence into the trained first time-series model to obtain the target attachment coefficient, sequence samples and corresponding coefficient samples are obtained, and the sequence samples are input into the second time-series model to obtain soft label coefficients; The sequence samples are input into the first time series model to be trained to obtain prediction coefficients, and the first model features of the first time series model to be trained and the second model features of the second time series model are obtained. The target loss is determined based on the coefficient samples, the soft label coefficients, the predicted coefficients, the first model features, and the second model features, and the target loss is used to train the first time series model to be trained.

4. The vehicle control method according to claim 3, characterized in that, The step of determining the target loss based on the coefficient samples, the soft-label coefficients, the predicted coefficients, the first model features, and the second model features includes: The task loss is determined based on the predicted coefficients and the coefficient samples, and the first distillation loss is determined based on the predicted coefficients and the soft label coefficients; The second distillation loss is determined based on the first model feature and the second model feature, and the target loss is determined based on the task loss, the first distillation loss, and the second distillation loss.

5. The vehicle control method according to claim 2, characterized in that, The step of inputting the feature sequence into the trained first temporal model to obtain the target adhesion coefficient includes: The feature sequence is input into the trained first time series model to obtain the current adhesion coefficient, and the historical adhesion coefficient is obtained. The target adhesion coefficient is obtained by performing median filtering on the current adhesion coefficient and the historical adhesion coefficient.

6. The vehicle control method according to claim 5, characterized in that, The step of obtaining the target adhesion coefficient by performing median filtering on the current adhesion coefficient and the historical adhesion coefficient includes: The current adhesion coefficient and the historical adhesion coefficient are sorted to obtain the sorting result; The target adhesion coefficient is determined from the current adhesion coefficient and the historical adhesion coefficient based on the sorting results.

7. The vehicle control method according to claim 1, characterized in that, The first time-series model includes a long short-term memory network, and the second time-series model includes a converter model.

8. A vehicle control device, characterized in that, include: The data acquisition module is configured to acquire vehicle status data; The coefficient prediction module is configured to use a first time series model to predict the state data to obtain a target adhesion coefficient, and control the vehicle operation according to the target adhesion coefficient. The first time series model learns the knowledge of the second time series model through knowledge distillation, and the second time series model has long time series modeling capability.

9. The vehicle control device according to claim 8, characterized in that, The coefficient prediction module is configured as follows: Statistical features are determined based on the state data, and the state data and statistical features are standardized to obtain standardized state data and statistical features. The feature sequence is determined based on the standardized state data and statistical characteristics, and the feature sequence is input into the trained first time series model to obtain the target adhesion coefficient.

10. The vehicle control device according to claim 9, characterized in that, The vehicle control device is also configured to: Before inputting the feature sequence into the trained first time-series model to obtain the target attachment coefficient, sequence samples and corresponding coefficient samples are obtained, and the sequence samples are input into the second time-series model to obtain soft label coefficients; The sequence samples are input into the first time series model to be trained to obtain prediction coefficients, and the first model features of the first time series model to be trained and the second model features of the second time series model are obtained. The target loss is determined based on the coefficient samples, the soft label coefficients, the predicted coefficients, the first model features, and the second model features, and the target loss is used to train the first time series model to be trained.

11. The vehicle control device according to claim 10, characterized in that, The vehicle control device is also configured to: The task loss is determined based on the predicted coefficients and the coefficient samples, and the first distillation loss is determined based on the predicted coefficients and the soft label coefficients; The second distillation loss is determined based on the first model feature and the second model feature, and the target loss is determined based on the task loss, the first distillation loss, and the second distillation loss.

12. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When executed by a processor, the program instructions implement the steps of the method described in any one of claims 1 to 7.

13. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the steps of the method according to any one of claims 1 to 7.

14. A vehicle, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to execute executable instructions stored in the memory to implement the steps of the vehicle control method according to any one of claims 1 to 7.

Images