Vehicle control method and vehicle

Abstract

The invention relates to the technical field of vehicle control, and provides a vehicle control method and a vehicle. The method comprises the following steps: acquiring state data of a vehicle and environment data around the vehicle; determining a predicted torque peak value according to the state data by using a pre-trained torque prediction model; determining terrain resistance parameters according to the environmental data by using a pre-trained terrain recognition model; and determining a load risk level of a vehicle transmission structure according to the predicted torque peak value and the terrain resistance parameter by using a pre-trained risk prediction model, and controlling the vehicle according to the load risk level. Therefore, the load risk level of the vehicle transmission structure can be accurately predicted according to the vehicle state and the terrain condition which change in real time, whether the vehicle transmission structure has the damage risk or not can be accurately judged according to the load risk level, and then accurate control over the vehicle can be achieved according to the load risk level.

Description

Technical Field

[0001] This disclosure relates to the field of vehicle control technology, and in particular to a vehicle control method and a vehicle. Background Technology

[0002] During vehicle operation, transmission components such as the differential and axle are prone to damage. However, predicting the risk of damage to the transmission components by analyzing data collected from a single sensor can be inaccurate.

[0003] Therefore, accurately predicting the risk of damage to the vehicle's transmission structure has become an urgent technical problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this disclosure is to propose a vehicle control method and a vehicle to solve the problem in the prior art that the risk of damage to the vehicle transmission structure cannot be accurately predicted.

[0005] To achieve the above objectives, a first aspect of this disclosure provides a vehicle control method, the method comprising:

[0006] Acquire vehicle status data and environmental data surrounding the vehicle; Using a pre-trained torque prediction model, the predicted peak torque is determined based on the state data; Using a pre-trained terrain recognition model, terrain resistance parameters are determined based on the environmental data; Using a pre-trained risk prediction model, the load risk level of the vehicle's transmission structure is determined based on the predicted peak torque and the terrain resistance parameters, and the vehicle is controlled according to the load risk level.

[0007] Based on the same inventive concept, a second 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 method described above when executing the computer program.

[0008] Based on the same inventive concept, a third aspect of this disclosure proposes a vehicle that includes the electronic equipment described in the second aspect.

[0009] As described above, the vehicle control method and vehicle disclosed herein acquire vehicle state data and surrounding environmental data. Using a pre-trained torque prediction model, the predicted peak torque is determined based on the state data, accurately predicting upcoming large torque demands. Using a pre-trained terrain recognition model, terrain resistance parameters are determined based on environmental data, transforming complex terrain features into specific terrain resistance parameters and quantifying the resistance impact of different terrains on vehicle operation. Using a pre-trained risk prediction model, the load risk level of the vehicle's transmission structure is determined based on the predicted peak torque and terrain resistance parameters, and the vehicle is controlled according to this load risk level. Thus, based on real-time changes in vehicle state and terrain conditions, the load risk level of the vehicle's transmission structure can be accurately predicted. Based on the load risk level, it is possible to accurately determine whether there is a risk of damage to the vehicle's transmission structure, thereby enabling precise vehicle control and timely implementation of protective measures to ensure the safety of the vehicle's transmission structure and guarantee vehicle driving safety. Attached Figure Description

[0010] 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.

[0011] Figure 1 This is a flowchart of a vehicle control method according to an embodiment of the present disclosure; Figure 2 A flowchart illustrating differential and axle overload prediction in a vehicle control method according to an embodiment of this disclosure; Figure 3 This is a schematic diagram of the structure of a vehicle control device according to an embodiment of the present disclosure; Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0012] 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.

[0013] 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.

[0014] Based on the background description, current off-road vehicles operating in complex unpaved road conditions are susceptible to early damage to critical transmission structures such as differentials and axles due to the combined effects of sudden torque impacts and high-intensity terrain resistance. Related risk prediction methods generally rely on single sensor signals or static threshold judgments, lacking the ability to model the combined effects of dynamic torque evolution trends and terrain environment, thus failing to accurately identify extreme risk conditions. Particularly in complex terrain scenarios such as high slopes, loose surfaces, and gravel-strewn terrain, traditional algorithms struggle to capture the amplification effect of unstructured terrain on structural components, leading to delayed warnings or high false alarm rates. Furthermore, the scarcity of extreme scenario data in relevant datasets results in insufficient coverage of operating conditions during model training, further reducing the reliability and generalization ability of predictions.

[0015] As mentioned above, how to accurately predict whether there is a risk of damage to the vehicle's transmission structure has become an important research question.

[0016] Based on the above description, such as Figure 1 As shown, the vehicle control method proposed in this embodiment includes: Step 101: Obtain vehicle status data and environmental data around the vehicle.

[0017] In specific implementation, the vehicle in this application embodiment can be an off-road vehicle. The vehicle's differential and axle, among other transmission structures, are prone to damage during off-road driving. The main cause of this damage is the localized stress concentration resulting from the superposition of sudden torque peaks and terrain resistance. Based on these reasons, the vehicle's state data refers to the operating state data of the power system under complex unpaved road conditions, and the environmental data surrounding the vehicle refers to the environmental data that exerts resistance on the vehicle under complex unpaved road conditions. Furthermore, dedicated strain gauges or high-frequency torque sensors are installed in key drive links to collect real-time load parameters of the half-shafts and differentials, among other transmission structures.

[0018] The vehicle's status data includes at least one of the following: engine output torque, wheel torque, vehicle speed, gear, throttle opening, steering angle, longitudinal acceleration, lateral acceleration, and wheel slip ratio. Specifically, the engine output torque and wheel torque are collected using torque sensors; vehicle speed and gear are collected using a Controller Area Network (CAN) bus; the throttle opening is collected using a throttle position sensor; the steering angle is collected using an Inertial Measurement Unit (IMU); the longitudinal and lateral accelerations are collected using an acceleration sensor; and the wheel slip ratio is collected using a wheel speed sensor. The use of the IMU allows for accurate identification of high-frequency signals indicating drastic changes in vehicle attitude, which helps in identifying sudden torque spikes.

[0019] The environmental data surrounding the vehicle includes environmental parameters and environmental images. Specifically, environmental parameters around the vehicle are collected using lidar; environmental images around the vehicle are collected using an onboard camera. In addition, terrain recognition devices can be installed on the underside and front of the vehicle to obtain the unstructured terrain surface conditions, such as slope, potholes, and gravel distribution density.

[0020] After acquiring vehicle status data and surrounding environmental data, the process includes preprocessing the status and environmental data to obtain target data. This preprocessing includes missing value removal, time synchronization, normalization, and outlier handling. Specifically, all status and environmental data must be collected synchronously with a unified timestamp, and the sampling frequency is set above 100Hz to ensure the capture of instantaneous load parameter changes. The collected status and environmental data undergo missing value removal, time synchronization, unit normalization, and outlier handling to obtain the target data.

[0021] Among them, the status data and environmental data cover the vehicle's status data and environmental data under various driving conditions such as normal operating conditions, minor impacts, and severe overloads.

[0022] Step 102: Using a pre-trained torque prediction model, determine the predicted peak torque based on the state data.

[0023] In practice, the torque prediction model is a pre-trained model used to predict the peak torque of a vehicle. Specifically, the state data is converted into torque time-series data of a preset duration; the torque time-series data of the preset duration is input into the pre-trained torque prediction model, which uses the pre-trained torque prediction model to determine the torque growth characteristics and torque decay characteristics based on the torque time-series data of the preset duration; and the predicted peak torque is determined based on the torque growth characteristics and torque decay characteristics.

[0024] Step 103: Using a pre-trained terrain recognition model, determine the terrain resistance parameters based on the environmental data.

[0025] In practice, the terrain recognition model is a pre-trained model used to identify the resistance generated by terrain on vehicles. Specifically, the generator in the terrain recognition model adds a noise vector to the environmental data to obtain a terrain feature map and a resistance feature vector; the discriminator in the terrain recognition model performs feature determination processing on the terrain feature map to obtain a first probability distribution, and performs determination processing on the resistance feature vector to obtain a second probability distribution; the terrain type with the highest probability distribution in the first probability distribution is determined from the terrain feature map, and the terrain resistance parameter with the highest probability distribution in the second probability distribution is determined from the resistance feature vector.

[0026] Step 104: Using a pre-trained risk prediction model, determine the load risk level of the vehicle transmission structure based on the predicted peak torque and the terrain resistance parameters, and control the vehicle based on the load risk level.

[0027] In practice, the risk prediction model is a pre-trained model used to predict the load risk of the vehicle's transmission structure. The vehicle transmission structure can be the vehicle differential and axle structure. Specifically, the predicted peak torque and terrain resistance parameters are aligned to obtain a joint feature vector; using the pre-trained risk prediction model, key features correlated with the load risk of the vehicle's transmission structure are extracted from the joint feature vector; and the load risk level of the vehicle's transmission structure is determined based on these key features.

[0028] Specifically, the predicted stress on the vehicle's transmission structure is determined based on the predicted peak torque and terrain resistance parameters. The load risk level corresponding to the predicted stress is determined based on pre-stored mapping relationships.

[0029] For example, the pre-stored mapping relationship is as follows: the first stress range corresponds to a low-risk level, the second stress range corresponds to a medium-risk level, and the third stress range corresponds to a high-risk level. In the above scenario, when the predicted stress falls within the first stress range, the load risk level is determined to be low-risk. When the predicted stress falls within the second stress range, the load risk level is determined to be medium-risk. When the predicted stress falls within the third stress range, the load risk level is determined to be high-risk.

[0030] This disclosure embodiment uses a torque prediction model to determine the predicted peak torque, a terrain identification model to determine the terrain resistance parameters, and a risk prediction model to determine the load risk level, thereby achieving transmission structure load risk level prediction based on the collaboration of multiple models using time series and terrain perception.

[0031] Through the above embodiments, vehicle status data and surrounding environmental data are acquired. Using a pre-trained torque prediction model, the predicted peak torque is determined based on the status data, accurately predicting upcoming large torque demands. Using a pre-trained terrain recognition model, terrain resistance parameters are determined based on environmental data, transforming complex terrain features into specific resistance parameters and quantifying the impact of different terrains on vehicle operation. Using a pre-trained risk prediction model, the load risk level of the vehicle's transmission structure is determined based on the predicted peak torque and terrain resistance parameters, and vehicle control is implemented according to this load risk level. Thus, based on real-time changes in vehicle status and terrain conditions, the load risk level of the vehicle's transmission structure can be accurately predicted. Based on the load risk level, it is possible to accurately determine whether there is a risk of damage to the transmission structure, thereby enabling precise vehicle control and timely protective measures to ensure the safety of the transmission structure and guarantee vehicle operation.

[0032] In some embodiments, step 102 includes: Step 1021: Determine torque time-series data of a preset duration from the state data, and input the torque time-series data into the pre-trained torque prediction model.

[0033] In practice, the torque prediction model can be obtained by pre-training a Bidirectional Gated Recurrent Unit (Bi-GRU). The Bi-GRU can simultaneously capture forward and backward features in time series data, making it suitable for processing torque time series data with asymmetric variation trends and indistinct periodicity.

[0034] The state data is divided into torque time series data of a preset duration according to a preset time window. For example, if the preset time window is 10 seconds, the state data is divided into multiple 10-second torque time series data.

[0035] The torque prediction model comprises an input layer, a hidden layer, and an output layer. The input layer receives torque time-series data for a preset duration. Simultaneously, the input layer also receives vehicle speed change rate, throttle opening change rate, and longitudinal acceleration for a preset duration, enhancing the model's ability to perceive torque response mechanisms under different operating conditions.

[0036] Step 1022: Using a pre-trained torque prediction model, extract torque change features from the torque time series data, and determine the predicted torque peak value based on the torque change features.

[0037] In practice, a pre-trained torque prediction model is used to extract torque growth and decay characteristics from torque time-series data, which are then used as torque change characteristics. Based on these torque change characteristics, the predicted torque intensity and predicted growth slope are determined, and the predicted torque peak value is determined based on these characteristics.

[0038] The above method determines torque time-series data of a preset duration from the state data and inputs this data into a pre-trained torque prediction model. Using the pre-trained model, torque change features are extracted from the time-series data, and the predicted torque peak value is determined based on these features. This allows for rapid and accurate determination of the predicted torque peak value, enabling early prediction of the peak torque during vehicle operation.

[0039] In some embodiments, step 1022 includes: Step 1022A: Using a pre-trained torque prediction model, extract torque growth features and torque decay features from the torque time series data, and use the torque growth features and torque decay features as torque change features.

[0040] In specific implementation, the hidden layer of the torque prediction model includes a forward-gated recurrent subnetwork and a backward-gated recurrent subnetwork. Specifically, the forward-gated recurrent subnetwork is used to extract features from the original sequence of torque time-series data to obtain forward features; the backward-gated recurrent subnetwork is used to extract features from the reverse sequence of torque time-series data to obtain backward features; based on the forward and backward features, torque growth and torque decay features are determined, and these features are used as torque change features.

[0041] Step 1022B: Determine the predicted torque intensity and predicted growth slope based on the torque change characteristics, and determine the predicted torque peak value based on the predicted torque intensity and the predicted growth slope.

[0042] In practical implementation, at the output layer of the torque prediction model, a fully connected layer determines the predicted torque intensity and predicted growth slope based on the torque change characteristics. The predicted torque intensity is the predicted torque intensity at each time step, and the predicted growth slope is the predicted growth slope at each time step. The predicted torque intensity and predicted growth slope are used to assess the impending torque surge trend.

[0043] In the output layer of the torque prediction model, the predicted torque peak value is determined based on the predicted torque intensity and the predicted growth slope. Thus, the torque prediction model acts as a torque trend prediction engine during vehicle operation, proactively identifying high-risk torque peaks, and thus serving as the first layer of risk filtering.

[0044] The above approach utilizes a pre-trained torque prediction model to extract torque growth and decay characteristics from torque time-series data, using these characteristics as torque variation features. This allows for a comprehensive determination of torque variation characteristics. Based on these features, torque intensity and growth slope are determined, and the predicted torque peak value is then calculated using these parameters. This enables rapid and accurate determination of the predicted torque peak value, allowing for early prediction of the peak torque during vehicle operation.

[0045] In some embodiments, the pre-training process of the torque prediction model includes: Step 102A: Obtain sample torque data of the vehicle under various driving conditions, and divide the sample torque data into torque training data and torque verification data.

[0046] In practice, the sample torque data can be high-frequency torque data samples actually collected by the automaker under various driving conditions. These various driving conditions include: low-speed high-resistance conditions, high-speed low-adhesion conditions, and high-frequency acceleration and deceleration conditions.

[0047] The sample torque data is divided into torque training data and torque validation data. Specifically, to improve the model's generalization ability, the sample torque data is uniformly sampled according to driving conditions to obtain torque training data and torque validation data.

[0048] Step 102B: The first neural network is trained using the torque training data to obtain an updated first neural network.

[0049] In a specific implementation, the first neural network can be a bidirectional gated recurrent unit (Bi-GRU). The updated first neural network is obtained by training the first neural network using torque training data.

[0050] Specifically, torque training data is input into a first neural network, which then determines a first predicted torque peak value based on the training data. A first actual torque peak value is determined from the training data, and a first loss function is determined based on the first predicted torque peak value and the first actual torque peak value. The model parameters of the first neural network are adjusted based on the first loss function to obtain an updated first neural network.

[0051] Step 102C: Use the torque verification data to verify the updated first neural network to obtain the second loss function of the updated first neural network, and determine the decrease of the second loss function within a preset number of training periods.

[0052] In practice, the updated first neural network is validated using torque verification data. If the updated first neural network passes the validation, it is used as the torque prediction model.

[0053] Specifically, the torque verification data is input into the updated first neural network, which then determines the second predicted torque peak value based on the torque verification data. The second actual torque peak value is determined from the torque verification data. A second loss function is determined based on the second predicted torque peak value and the second actual torque peak value. The rate of decrease of the second loss function within a preset number of training periods is then determined, and training is stopped based on the rate of decrease.

[0054] Step 102D: In response to determining that the decrease in the second loss function is less than a preset first magnitude threshold, the updated first neural network is used as the torque prediction model.

[0055] In practice, when the decrease in the second loss function is less than a preset first magnitude threshold, an early stopping mechanism is triggered, and the updated first neural network is used as the torque prediction model. This way, if the second loss function of the torque verification data does not decrease within multiple consecutive training cycles, training is stopped early to prevent the updated first neural network from overlearning noise or specific patterns in the torque training data. This improves the generalization ability of the torque prediction model and prevents overfitting.

[0056] Early stopping is a simple and effective way to prevent overfitting. The core idea is to monitor the performance of validation data and terminate training early when the model performance no longer improves, so as to avoid the model performing well on training data but having poor generalization ability on test data due to excessive training time.

[0057] The above scheme acquires sample torque data of vehicles under various driving conditions and divides this data into torque training data and torque validation data. The torque training data is used to train a first neural network, resulting in an updated first neural network. The torque validation data is then used to validate the updated first neural network, yielding a second loss function. The magnitude of the decrease in the second loss function over a preset number of training periods is determined. If the magnitude of the decrease in the second loss function is less than a preset first magnitude threshold, the updated first neural network is used as the torque prediction model. This approach prevents the updated first neural network from overlearning noise or specific patterns in the torque training data if the second loss function fails to decrease over several consecutive training periods, thereby improving the generalization ability of the torque prediction model and preventing overfitting.

[0058] In some embodiments, step 103 includes: Step 1031: Using the generator in the terrain recognition model, add a noise vector to the environmental data to obtain a terrain feature map and a resistance feature vector.

[0059] In practice, the terrain recognition model can be pre-trained from a Generative Adversarial Network (GAN). A GAN is a deep learning model consisting of a generator and a discriminator, capable of efficiently simulating real-world data distributions.

[0060] By using the generator in the terrain recognition model, noise vectors are added to the environmental data to obtain terrain feature maps and resistance feature vectors. The terrain feature map is the terrain feature map with noise vectors added, and the resistance feature vector is the resistance feature vector with noise vectors added.

[0061] Step 1032: Using the discriminator in the terrain recognition model, perform feature determination processing on the terrain feature map to obtain a first distribution probability, and perform determination processing on the resistance feature vector to obtain a second distribution probability.

[0062] In practice, the discriminator in the terrain recognition model is used to process the terrain feature map through convolutional layers and fully connected layers to obtain the first distribution probability, and the resistance feature vector is processed through a multilayer perceptron (MLP) to obtain the second distribution probability.

[0063] Step 1033: Determine the terrain type with the highest distribution probability from the terrain feature map, determine the terrain resistance feature with the highest distribution probability from the resistance feature vector, and use the terrain type and the terrain resistance feature as the terrain resistance parameter.

[0064] In practice, the terrain type with the highest probability of distribution is determined from the terrain feature map, and the terrain resistance feature with the highest probability of distribution is determined from the resistance feature vector. The terrain type and terrain resistance feature are then used as terrain resistance parameters. In this way, the pre-trained terrain recognition model can accurately identify the terrain resistance parameters around the vehicle.

[0065] The above scheme utilizes the generator in the terrain recognition model to add noise vectors to environmental data, obtaining terrain feature maps and resistance feature vectors. The discriminator in the terrain recognition model then performs feature determination processing on the terrain feature maps to obtain a first probability distribution, and performs determination processing on the resistance feature vectors to obtain a second probability distribution. From the terrain feature maps, the terrain type with the highest first probability distribution is determined, and from the resistance feature vectors, the terrain resistance feature with the highest second probability distribution is determined. These terrain type and terrain resistance feature are then used as terrain resistance parameters. This allows for accurate identification of surrounding terrain resistance parameters.

[0066] In some embodiments, the pre-training process of the terrain recognition model includes: Step 103A: Obtain sample environmental data of the vehicle under various driving conditions, and use the sample environmental data to train the discriminator in the second neural network to obtain an updated discriminator.

[0067] In practice, to address the issue of uneven distribution of extreme samples in the training set, sample environmental data of vehicles under various driving conditions is acquired, covering sample environmental data under various complex terrain conditions. Specifically, based on a large amount of image data, LiDAR point cloud data, and acceleration data of the surrounding environment collected on-site at the off-road test track built by the automaker, real terrain-resistance correlated sample environmental data is constructed.

[0068] The sample environmental data includes: structured parameters, unstructured images, and actual stresses on the vehicle's transmission structure. Structured parameters include at least one of the following: slope, tire sag depth, tire sag area, and particulate density. Unstructured images include: terrain surface material, terrain surface color, and terrain surface texture. Actual stresses include: wheel-side resistance, acceleration, and vehicle speed fluctuations.

[0069] The second neural network can be a generative adversarial network (GAN). Sample environmental data is used to train the discriminator in the GAN, enabling the discriminator to identify the probability distribution of real terrain conditions.

[0070] Step 103B: Obtain the sample noise vector, and use the sample noise vector to train the generator in the second neural network to obtain the updated generator.

[0071] In specific implementation, the second neural network can be a Generative Adversarial Network (GAN). The generator in the GAN is trained using sample noise vectors and base vehicle parameters to obtain an updated generator. The sample noise vectors can be a set of low-dimensional random noise vectors. The basic vehicle parameters include at least one of the following: axle load, tire type, and powertrain configuration. The generator output is a fake sample that fits the terrain drag features; the fake sample includes a terrain feature map in image form or a structured drag feature vector.

[0072] Step 103C: Construct a terrain recognition model based on the updated discriminator and the updated generator.

[0073] In practice, the generator parameters are continuously optimized through adversarial training to enable the generator to generate realistic, reasonably distributed, and physically plausible extreme terrain scenarios. These extreme terrain scenarios can be scenarios with rapidly changing adhesion coefficients on steep slopes or scenarios with mixed gravel and mud surfaces.

[0074] After the terrain recognition model is trained, it can generate a batch of extended data sets for training subsequent risk prediction models. This effectively covers extreme resistance distribution ranges that are difficult to obtain from actual measurements, significantly enhancing the risk prediction model's ability to identify rare scenarios and improving its prediction stability and generalization ability under non-equilibrium conditions. Through generative modeling, terrain-resistance combinations that are difficult to simulate in the physical world are supplemented in a data-driven manner, providing a stable and reliable data expansion solution for the risk prediction model and effectively solving the prediction blind spot problem caused by the lack of extreme data.

[0075] The above scheme acquires sample environmental data of vehicles under various driving conditions. This sample environmental data is then used to train the discriminator in the second neural network, resulting in an updated discriminator capable of recognizing the probability distribution of real terrain conditions. Sample noise vectors are then acquired and used to train the generator in the second neural network, resulting in an updated generator capable of generating realistic, reasonably distributed, and physically plausible extreme terrain scenes. A terrain recognition model is then constructed based on the updated discriminator and the updated generator.

[0076] In some embodiments, the overload prediction task for off-road vehicles involves highly complex data types, including both structured numerical time-series signals such as torque, vehicle speed, and acceleration, and unstructured high-dimensional information such as terrain images, point cloud data, and generated samples. To achieve input data uniformity, multi-source data features are encoded and aligned, enabling different types of multi-source data features to collaboratively enter the risk prediction model.

[0077] First, in the structured feature processing section, engine output torque, wheel torque, vehicle speed, throttle opening, longitudinal acceleration, lateral acceleration, wheel speed, and differential temperature are used as input variables. Time-series data of a preset duration are extracted according to a preset time window, and a unified sampling frequency is used to construct time-series samples using a sliding window mechanism. For high-frequency changing signals (e.g., torque), statistical feature enhancement is employed to extract features such as maximum value, volatility, rise slope, and peak duration to improve the sensitivity of the torque prediction model in identifying sudden torque peaks.

[0078] Secondly, in the unstructured feature processing section, the environmental image is input into a pre-trained convolutional encoder for feature extraction, yielding semantic vectors such as surface type, obstacle distribution, and texture complexity. Simultaneously, dimensionality reduction projection and voxelization are applied to the point cloud data, and local geometric features are extracted using architectures such as Point Network (PointNet). These high-dimensional features are uniformly encoded into fixed-dimensional vectors, concatenated with structured features, and then input into the risk prediction network.

[0079] To ensure temporal consistency of data from different sources, the system incorporates a time alignment module. This module precisely matches image sampling frames, LiDAR frames, and vehicle sensor signals using timestamps, constructing a multimodal joint feature set for each sample. Simultaneously, a missing-filling strategy addresses the issue of image frame rates being lower than sensor signal frequencies, ensuring that multiple sample data points contain complete torque state, vehicle dynamic response, and terrain environment information. The final output is a joint feature vector encompassing vehicle powertrain state, terrain attributes, and driving behavior characteristics. This vector serves as input to the risk prediction model, laying the foundation for deep collaborative prediction of terrain resistance and torque impact. This approach achieves the fusion preprocessing of structured temporal data and unstructured visual spatial data, maximizing feature dimensionality while ensuring information integrity. This meets the real-time requirements of the risk prediction model and also possesses high scalability, compatible with future inputs from more types of sensors.

[0080] In some embodiments, step 104 includes: Step 1041: Using a pre-trained risk prediction model, the predicted peak torque and the terrain resistance parameters are aligned to obtain a joint feature vector.

[0081] In practice, the risk prediction model can be pre-trained from a Tabular Neural Network (TabNet). TabNet can effectively process high-dimensional joint feature vectors resulting from the fusion of structured and unstructured features, and achieve accurate classification of load risk levels.

[0082] TabNet is a deep learning model specifically optimized for structured data. It integrates attention mechanisms and interpretable feature selection strategies, enabling it to dynamically select key features relevant to load risk prediction while maintaining model performance. The joint feature vector includes: torque time-series data, vehicle state data, terrain image features, and terrain point cloud features.

[0083] Step 1042: Using the feature transformer in the risk prediction model, extract key features that are associated with the load risk of the vehicle transmission structure from the joint feature vector.

[0084] In practice, the risk prediction model includes an input layer, a hidden layer, and an output layer.

[0085] The input layer of the risk prediction model initializes the joint feature vector based on the importance of various features in the data collected by the automakers. In the hidden layer of the risk prediction model, multiple feature transformers and decision blocks are stacked and combined to extract deep interaction relationships.

[0086] The Decision Block consists of multiple Decision Steps. In each Decision Step, a sparse attention mechanism is used to determine the correlation between each feature. By performing a nonlinear transformation on the correlation, the tabular neural network can automatically learn key features that are highly correlated with the load risk of the vehicle's transmission structure in an end-to-end manner.

[0087] Step 1043: Determine the load risk level of the vehicle transmission structure based on the key features.

[0088] In practice, at the output layer of the risk prediction model, the load risk level of the vehicle transmission structure is determined based on key features, and the risk prediction model outputs a classification result representing the risk level of the differential and the axle.

[0089] The load risk levels include: low risk, medium risk, and high risk. A low risk level indicates the vehicle's transmission structure is in a safe condition; a medium risk level indicates the transmission structure is under slight overload; and a high risk level indicates the transmission structure is under severe overload. Furthermore, the load risk levels can be customized with more granular labels according to the automaker's needs.

[0090] The above approach utilizes a pre-trained risk prediction model to align the predicted peak torque and terrain resistance parameters, obtaining a joint feature vector. The feature transformer within the risk prediction model then extracts key features correlated with the load risk of the vehicle's transmission structure from this joint feature vector. The load risk level of the vehicle's transmission structure is determined based on these key features. This comprehensive approach, combining the predicted peak torque and terrain resistance parameters, enables a more accurate prediction of the vehicle's transmission structure's load risk level.

[0091] In some embodiments, the pre-training process of the risk prediction model includes: Step 104A: Obtain sample torque peak data and sample terrain resistance data, and perform alignment processing on the sample torque peak data and the sample terrain resistance data to obtain sample joint features.

[0092] In practice, peak torque data and terrain resistance data are acquired and input into a third neural network. The third neural network then aligns the peak torque data and terrain resistance data to obtain joint features. During the training phase of the risk prediction model, labeled historical measured datasets are used for supervised learning.

[0093] Step 104B: Using the third neural network, determine the predicted load risk of the vehicle transmission structure based on the joint features of the samples, and determine the cross-entropy loss function based on the predicted load risk.

[0094] In practice, the third neural network can be a tabNet. The updated third neural network is obtained by training the third neural network using the joint features of the samples.

[0095] A third neural network is used to determine the predicted load risk of the vehicle's transmission structure based on the joint features of the samples. The actual load risk corresponding to the sample peak torque data and sample terrain resistance data is obtained, and the cross-entropy loss function is determined based on the predicted load risk and the actual load risk.

[0096] Step 104C: Adjust the model parameters of the third neural network based on the cross-entropy loss function to obtain the updated third neural network.

[0097] In practice, the cross-entropy loss function, combined with an early stopping mechanism, is used to adjust the model parameters of the third neural network, resulting in an updated third neural network. Simultaneously, class balancing weights are introduced to address the uneven distribution of sample numbers across different load risk levels. Specifically, adjustments are made to model parameters such as feature sparsity control parameters, the number of decision steps, and the attention temperature coefficient of the third neural network, enabling the risk prediction model to maintain its generalization ability while improving its discrimination capability under rare and extreme conditions.

[0098] Step 104D: In response to determining that the decrease in the cross-entropy loss function is less than a preset second magnitude threshold within a preset number of training periods, the updated third neural network is used as the risk prediction model.

[0099] In practice, if the decrease in the cross-entropy loss function of the updated third neural network is less than a preset second magnitude threshold within a preset number of training cycles, an early stopping mechanism is triggered, and the updated third neural network is used as the risk prediction model. This way, training is stopped early when the cross-entropy loss function fails to decrease over several consecutive training cycles, preventing the updated third neural network from overlearning noise or specific patterns in the joint features of the samples. This improves the generalization ability of the risk prediction model and prevents overfitting.

[0100] Early stopping is a simple and effective way to prevent overfitting. The core idea is to monitor the performance of validation data and terminate training early when the model performance no longer improves, so as to avoid the model performing well on training data but having poor generalization ability on test data due to excessive training time.

[0101] In addition, to meet the response speed requirements of industrial deployment, the risk prediction model is pruned and optimized after training to make it adaptable to the deployment requirements of embedded platforms.

[0102] Through the risk prediction model, the system not only achieves high-precision identification of vehicle transmission structure load risks, but also possesses good feature interpretability, making it easy for engineers to trace the basis for the risk prediction model's judgments. This provides valuable load risk prediction data and strategy support for automotive R&D departments. A deep attention mechanism and feature selection path control are introduced on the basis of traditional models, overcoming the problem that structured models struggle to handle high-dimensional heterogeneous features. This is a key discrimination module in the algorithm fusion system of this disclosure embodiment.

[0103] The above scheme acquires sample peak torque data and sample terrain resistance data, and aligns these data to obtain sample joint features. Using a third neural network, the predicted load risk of the vehicle transmission structure is determined based on the sample joint features, and a cross-entropy loss function is determined based on the predicted load risk. The model parameters of the third neural network are adjusted based on the cross-entropy loss function to obtain an updated third neural network. If the decrease in the cross-entropy loss function is less than a preset second threshold within a preset number of training periods, the updated third neural network is used as the risk prediction model. This way, training is stopped early when the cross-entropy loss function does not decrease within several consecutive training periods, preventing the updated third neural network from overlearning noise or specific patterns in the sample joint features, thereby improving the generalization ability of the risk prediction model and preventing overfitting.

[0104] In some embodiments, in order to accurately predict the overload risk of off-road vehicle differentials and axles under complex terrain conditions, a multi-model fusion mechanism with torque prediction model, terrain recognition model and risk prediction model as the core not only realizes information flow and functional complementarity between models in structure, but also constructs a comprehensive discrimination system with time sensitivity, terrain perception capability and classification accuracy through loss collaboration in the training stage and feature transfer in the inference stage.

[0105] First, the input data is modularly split, with torque time-series data fed separately into the torque prediction model to extract the predicted torque peak. The terrain recognition model receives environmental data and random noise vectors from the vehicle's surroundings. During the training phase, it generates representative extreme terrain drag feature samples. The generator output and discriminator feedback signals enrich the training set. Simultaneously, the generated pseudo-samples, after feature extraction by the image encoder, are fed into the subsequent processing path along with the measured terrain image features. The risk prediction model acts as the main discriminator, receiving three types of fused information as input: the predicted torque peak output from the torque prediction model, the original vehicle operating condition variables and driving behavior characteristics, and the output terrain drag parameters.

[0106] To enhance the collaborative capabilities between models, the system designs a unified loss function framework during the training phase, including the time series prediction error of the torque prediction model, the adversarial loss of the terrain recognition model (alternating optimization of the generator and discriminator), and the classification loss of the risk prediction model. The various losses are combined proportionally, and the classification loss weight is dynamically adjusted to guide the optimization direction towards the actual prediction accuracy.

[0107] The training process employs a joint training strategy. In the initial stage, each model undergoes independent pre-training. After convergence, each model enters a collaborative training phase, where a unified optimizer iteratively updates all model parameters. During the inference phase, each model independently processes the input data, and the encoded results are aggregated into the risk prediction model through an intermediate feature sharing mechanism, achieving end-to-end risk classification output.

[0108] The fusion mechanism not only enhances the coverage of feature dimensions that are easily overlooked by single models, but also strengthens the robustness of the risk prediction model to extreme inputs through simulation of abnormal operating conditions. Compared with the traditional serial model architecture, the embodiments disclosed in this disclosure emphasize loss resonance in the training phase and feature alignment in the inference phase, forming a highly integrated intelligent prediction architecture with strong generalization, modular scalability, and engineering deployment adaptability. Through the above fusion mechanism, automakers can build a dedicated, sustainably evolving intelligent overload risk assessment system for off-road platforms, achieving proactive protection of key vehicle transmission structures and a data-driven reliability design closed loop.

[0109] In some embodiments, step 104 includes: Step 1045: In response to determining that the load risk level is a low risk level, the vehicle is controlled to maintain its current state.

[0110] In practice, when the load risk level is low, it means that the vehicle's transmission structure is in a safe state. The vehicle is then controlled to maintain its current state, and the load risk level of the vehicle's transmission structure is continuously predicted.

[0111] Specifically, if the predicted peak torque is within the normal range, the terrain resistance parameters do not show any abnormal areas, and the predicted stress on the vehicle transmission structure is within the safe first stress range, then the load risk level is determined to be low risk level.

[0112] For example, a maximum stress threshold is preset, and the first stress range is less than 60% of the maximum stress threshold. When the predicted stress on the vehicle transmission structure is less than 60% of the maximum stress threshold, the load risk level of the vehicle transmission structure is determined to be low risk level.

[0113] Step 1046: In response to determining that the load risk level is a medium risk level, a warning prompt for the load risk of the vehicle transmission structure is triggered, and the load risk event is recorded.

[0114] In practice, when the load risk level is medium risk, it indicates that the vehicle's transmission structure is in a state of slight overload. In this case, an early warning of load risk of the vehicle's transmission structure is triggered, and the load risk event is recorded.

[0115] Specifically, when the predicted peak torque exhibits short-term drastic torque fluctuations, the terrain resistance parameter shows a sharp upward trend but has not yet reached the theoretical failure threshold, and the predicted stress on the vehicle transmission structure is within the second stress range of mild overload, the load risk level is determined to be medium risk.

[0116] For example, a maximum stress threshold is preset, and the second stress range is 60% to 80% of the maximum stress threshold. When the predicted stress on the vehicle transmission structure is within the second stress range corresponding to 60% to 80% of the maximum stress threshold, the load risk level of the vehicle transmission structure is determined to be a medium risk level.

[0117] Step 1047: In response to determining that the load risk level is a high risk level, an alarm for the load risk of the vehicle transmission structure is triggered, and the current power parameters of the vehicle are controlled within the preset power parameter range.

[0118] In practice, when the load risk level is high, indicating that the vehicle's transmission structure is under severe overload, an alarm for the load risk of the vehicle's transmission structure is triggered, and the vehicle's current power parameters are controlled within a preset power parameter range. This preset power parameter range is designed to prevent damage to the vehicle's transmission structure.

[0119] The preset power parameter ranges include: preset torque range and preset vehicle speed range. Specifically, when the load risk level of the vehicle's transmission structure is high, a warning message for the load risk of the vehicle's transmission structure is triggered, and the vehicle's current torque is controlled within the preset torque range. Alternatively, when the load risk level of the vehicle's transmission structure is high, a warning message for the load risk of the vehicle's transmission structure is triggered, and the vehicle's current speed is controlled within the preset speed range.

[0120] Specifically, when the predicted peak torque shows a continuous upward trend, the terrain resistance parameters have concentrated risk areas and the path has risk behaviors such as getting stuck, slipping, and high-speed impact, the load on the vehicle transmission structure is close to the limit, and the predicted stress on the vehicle transmission structure is within the third stress range of heavy overload, then the load risk level is determined to be high risk level.

[0121] For example, a maximum stress threshold is preset, and the third stress range is greater than 80% of the maximum stress threshold. When the predicted stress on the vehicle transmission structure is greater than 80% of the maximum stress threshold, the load risk level of the vehicle transmission structure is determined to be a high-risk level.

[0122] To improve the accuracy of the judgment, the system introduces a joint feature boundary mechanism in the judgment logic to establish a two-dimensional mapping relationship between terrain resistance parameters and predicted torque peak. When the surface disturbance frequency, texture complexity, slope steepness and other indicators in the terrain resistance parameters exceed the set threshold, and the predicted torque peak is accompanied by a rapid increase in the slope, a high-risk level is automatically triggered.

[0123] Furthermore, to adapt to different vehicle models and user preferences, the system allows automakers' R&D teams to customize risk thresholds based on historical test data, ensuring the same algorithm's adaptability across different vehicle platforms. All risk level labels are timestamped, outputting structured records for easy integration with embedded systems or interface with onboard diagnostic systems.

[0124] To enhance engineering usability, the system incorporates output signal redundancy verification and anomaly marking mechanisms to ensure reliable load risk levels are still output even in the event of sensor failure or data loss. Through this discrimination logic, the entire prediction system achieves a complete closed loop from sensor data acquisition, feature extraction, multi-model fusion to final risk decision-making. This supports automakers in developing proactive protection strategies for critical structural components such as differentials and axles, while providing traceable and quantifiable data support for after-sales maintenance and vehicle reliability improvement. The innovation of this logic lies in integrating temporal trends and spatial structural characteristics into risk level decisions. This ensures that risk judgment does not rely solely on a single model output but rather on multi-model cross-validation to form a more robust decision path, significantly enhancing the reliability and interpretability of judgments under extreme off-road conditions.

[0125] Through the above scheme, when the load risk level is low, the vehicle is kept in its current state, continuously monitoring the load risk level of the vehicle's transmission structure. When the load risk level is medium, a warning message for the load risk of the vehicle's transmission structure is triggered, and the load risk event is recorded, allowing the driver to promptly understand whether the vehicle's transmission structure is slightly overloaded, enabling the driver to control the vehicle and reduce the load risk of the transmission structure. When the load risk level is high, an alarm message for the load risk of the vehicle's transmission structure is triggered, and the vehicle's current power parameters are kept within a preset range, limiting the vehicle's current power parameters to prevent damage to the vehicle's transmission structure.

[0126] Through the above embodiments, vehicle status data and surrounding environmental data are acquired. Using a pre-trained torque prediction model, the predicted peak torque is determined based on the status data, accurately predicting upcoming large torque demands. Using a pre-trained terrain recognition model, terrain resistance parameters are determined based on environmental data, transforming complex terrain features into specific resistance parameters and quantifying the impact of different terrains on vehicle operation. Using a pre-trained risk prediction model, the load risk level of the vehicle's transmission structure is determined based on the predicted peak torque and terrain resistance parameters, and vehicle control is implemented according to this load risk level. Thus, based on real-time changes in vehicle status and terrain conditions, the load risk level of the vehicle's transmission structure can be accurately predicted. Based on the load risk level, it is possible to accurately determine whether there is a risk of damage to the transmission structure, thereby enabling precise vehicle control and timely protective measures to ensure the safety of the transmission structure and guarantee vehicle operation.

[0127] It should be noted that the embodiments of this disclosure can also be further described in the following ways: Figure 2 This is a flowchart illustrating differential and axle overload prediction in the vehicle control method of this disclosure embodiment. Figure 2 As shown, the process of predicting differential and axle overload in the vehicle control method includes: Step 1: Data collection and processing of dynamic operating conditions of off-road vehicles.

[0128] Acquire vehicle status data and environmental data surrounding the vehicle. Vehicle status data includes at least one of the following: engine output torque, wheel torque, vehicle speed, gear position, throttle opening, steering angle, longitudinal acceleration, lateral acceleration, and wheel slip ratio. Environmental data surrounding the vehicle includes environmental parameters and environmental images.

[0129] Step 2, Torque peak evolution modeling: Bi-Gated cyclic unit (Bi-GRU).

[0130] The state data is converted into torque time-series data of a preset duration; the torque time-series data of the preset duration is input into a pre-trained torque prediction model, and the torque growth characteristics and torque decay characteristics are determined by the pre-trained torque prediction model based on the torque time-series data of the preset duration; the predicted torque peak value is determined based on the torque growth characteristics and torque decay characteristics.

[0131] Step 3, Terrain resistance sample generation mechanism: Extreme condition enhancement method based on generative adversarial network (GAN).

[0132] Using the generator in the terrain recognition model, noise vectors are added to the environmental data to obtain terrain feature maps and resistance feature vectors; using the discriminator in the terrain recognition model, feature determination processing is performed on the terrain feature maps to obtain the first distribution probability, and determination processing is performed on the resistance feature vectors to obtain the second distribution probability; the terrain type with the highest first distribution probability is determined from the terrain feature maps, and the terrain resistance parameter with the highest second distribution probability is determined from the resistance feature vectors.

[0133] Step 4, Structured and unstructured multi-source feature encoding: input variable construction and alignment mechanism.

[0134] To achieve uniformity in input data, multi-source data features are encoded and aligned, enabling different types of multi-source data features to be collaboratively incorporated into the risk prediction model.

[0135] Step 5, Multimodal fusion classification model: TabNet structure design and parameter tuning mechanism.

[0136] The predicted peak torque and terrain resistance parameters are aligned to obtain a joint feature vector; a pre-trained risk prediction model is used to extract key features that are related to the load risk of the vehicle transmission structure from the joint feature vector; and the load risk level of the vehicle transmission structure is determined based on the key features.

[0137] Step 6, Design of multi-model fusion mechanism: Joint inference and loss co-training strategy of Bi-GRU, GAN and TabNet.

[0138] The multi-model fusion mechanism, which is based on torque prediction model, terrain recognition model and risk prediction model, not only realizes information flow and functional complementarity between models in structure, but also constructs a comprehensive discrimination system with time sensitivity, terrain perception capability and classification accuracy through loss collaboration in the training stage and feature transfer in the inference stage.

[0139] Step 7: Risk level determination logic based on terrain-torque feature space.

[0140] When the load risk level is low, the vehicle is kept in its current state; when the load risk level is medium, a warning is triggered for the load risk of the vehicle's transmission structure, and the load risk event is recorded; when the load risk level is high, an alarm is triggered for the load risk of the vehicle's transmission structure, and the vehicle's current power parameters are controlled within the preset power parameter range.

[0141] 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.

[0142] 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.

[0143] Based on the same inventive concept, corresponding to any of the above-described embodiments, this disclosure also provides a vehicle control device.

[0144] refer to Figure 3 The vehicle control device includes: The acquisition module 301 is configured to acquire vehicle status data and environmental data around the vehicle. The torque prediction module 302 is configured to use a pre-trained torque prediction model to determine the predicted torque peak based on the state data. The terrain recognition module 303 is configured to determine terrain resistance parameters based on the environmental data using a pre-trained terrain recognition model. The control module 304 is configured to use a pre-trained risk prediction model to determine the load risk level of the vehicle transmission structure based on the predicted peak torque and the terrain resistance parameters, and to control the vehicle based on the load risk level.

[0145] In some embodiments, the torque prediction module 302 includes: The torque time series data determination unit is configured to determine torque time series data of a preset duration from the state data and input the torque time series data into a pre-trained torque prediction model. The feature extraction unit is configured to extract torque change features from the torque time series data using a pre-trained torque prediction model, and determine the predicted torque peak value based on the torque change features.

[0146] In some embodiments, the feature extraction unit includes: The torque change feature determination subunit is configured to use a pre-trained torque prediction model to extract torque growth features and torque decay features from the torque time series data, and use the torque growth features and the torque decay features as torque change features. The predicted torque peak determination subunit is configured to determine the predicted torque intensity and the predicted growth slope based on the torque change characteristics, and to determine the predicted torque peak value based on the predicted torque intensity and the predicted growth slope.

[0147] In some embodiments, the apparatus further includes a torque prediction model training module, the torque prediction model training module comprising: The sample torque data acquisition unit is configured to acquire sample torque data of the vehicle under various driving conditions, and to divide the sample torque data into torque training data and torque verification data. The first neural network update unit is configured to train the first neural network using the torque training data to obtain the updated first neural network. The descent magnitude determination unit is configured to use the torque verification data to perform verification processing on the updated first neural network to obtain the second loss function of the updated first neural network, and to determine the descent magnitude of the second loss function within a preset number of training periods; The torque prediction model determination unit is configured to use the updated first neural network as the torque prediction model in response to determining that the decrease magnitude of the second loss function is less than a preset first magnitude threshold.

[0148] In some embodiments, the terrain recognition module 303 includes: The feature generation unit is configured to use the generator in the terrain recognition model to add a noise vector to the environmental data to obtain a terrain feature map and a resistance feature vector. The feature determination unit is configured to use the discriminator in the terrain recognition model to perform feature determination processing on the terrain feature map to obtain a first distribution probability, and to perform determination processing on the resistance feature vector to obtain a second distribution probability; The terrain resistance parameter determination unit is configured to determine the terrain type with the highest distribution probability in the terrain feature map, determine the terrain resistance feature with the highest distribution probability in the resistance feature vector, and use the terrain type and the terrain resistance feature as the terrain resistance parameter.

[0149] In some embodiments, the apparatus further includes a terrain recognition model training module, the terrain recognition model training module comprising: The discriminator training unit is configured to acquire sample environment data of the vehicle under various driving conditions, and use the sample environment data to train the discriminator in the second neural network to obtain an updated discriminator; The generator training unit is configured to acquire a sample noise vector and use the sample noise vector to train the generator in the second neural network to obtain an updated generator. The terrain recognition model determination unit is configured to construct a terrain recognition model based on the updated discriminator and the updated generator.

[0150] In some embodiments, the control module 304 includes: The joint feature vector determination unit is configured to use a pre-trained risk prediction model to align the predicted peak torque and the terrain resistance parameters to obtain a joint feature vector; The key feature determination unit is configured to use the feature transformer in the risk prediction model to extract key features that are associated with the load risk of the vehicle transmission structure from the joint feature vector. The load risk level determination unit is configured to determine the load risk level of the vehicle transmission structure based on the key features.

[0151] In some embodiments, the apparatus further includes a risk prediction model training module, the risk prediction model training module comprising: The sample joint feature determination unit is configured to acquire sample torque peak data and sample terrain resistance data, and perform alignment processing on the sample torque peak data and sample terrain resistance data to obtain sample joint features; The cross-entropy loss function determination unit is configured to use the third neural network to determine the predicted load risk of the vehicle transmission structure based on the joint features of the samples, and to determine the cross-entropy loss function based on the predicted load risk. The third neural network update unit is configured to adjust the model parameters of the third neural network based on the cross-entropy loss function to obtain the updated third neural network. The risk prediction model determination unit is configured to use the updated third neural network as the risk prediction model in response to determining that the decrease in the cross-entropy loss function is less than a preset second magnitude threshold within a preset number of training periods.

[0152] In some embodiments, the control module 304 includes: The first control unit is configured to control the vehicle to maintain its current state in response to determining that the load risk level is a low risk level. The second control unit is configured to trigger an early warning of load risk in the vehicle transmission structure and record the load risk event in response to determining that the load risk level is medium risk level. The third control unit is configured to trigger an alarm for load risk of the vehicle transmission structure in response to determining that the load risk level is high, and to control the current power parameters of the vehicle within a preset power parameter range.

[0153] 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.

[0154] 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.

[0155] 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.

[0156] Figure 4 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.

[0157] 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.

[0158] 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.

[0159] 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.

[0160] The communication interface 1040 is used to connect the 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 (Universal Serial Bus), network cable, etc.) or wireless means (such as mobile network, WIFI (Wireless Fidelity), Bluetooth, etc.).

[0161] 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.

[0162] 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.

[0163] 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.

[0164] 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.

[0165] 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 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.

[0166] 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.

[0167] 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, or electronic device, or storage medium in the above embodiments, wherein the vehicle device implements the vehicle control method described in any of the above embodiments.

[0168] 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.

[0169] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides a computer program product, including computer program instructions. When the computer program instructions are run on a computer, the computer causes the computer to execute the vehicle control method as described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0170] 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.

[0171] 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.

[0172] 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.

[0173] 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.

[0174] 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 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.

[0175] 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.

[0176] 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.

[0177] This disclosure is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this disclosure. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the protection scope of this disclosure.

Claims

1. A vehicle control method, characterized in that, The method includes: Acquire vehicle status data and environmental data surrounding the vehicle; Using a pre-trained torque prediction model, the predicted peak torque is determined based on the state data; Using a pre-trained terrain recognition model, terrain resistance parameters are determined based on the environmental data; Using a pre-trained risk prediction model, the load risk level of the vehicle's transmission structure is determined based on the predicted peak torque and the terrain resistance parameters, and the vehicle is controlled according to the load risk level.

2. The method according to claim 1, characterized in that, The step of using a pre-trained torque prediction model to determine the predicted torque peak value based on the state data includes: Determine torque time-series data of a preset duration from the state data, and input the torque time-series data into a pre-trained torque prediction model; Using a pre-trained torque prediction model, torque change features are extracted from the torque time series data, and the predicted torque peak value is determined based on the torque change features.

3. The method according to claim 2, characterized in that, The step of using a pre-trained torque prediction model to extract torque variation features from the torque time-series data and determining the predicted torque peak value based on the torque variation features includes: Using a pre-trained torque prediction model, torque growth features and torque decay features are extracted from the torque time series data, and the torque growth features and torque decay features are used as torque change features. Based on the torque change characteristics, the predicted torque intensity and predicted growth slope are determined, and the predicted torque peak value is determined based on the predicted torque intensity and the predicted growth slope.

4. The method according to claim 1, characterized in that, The pre-training process of the torque prediction model includes: Acquire sample torque data of the vehicle under various driving conditions, and divide the sample torque data into torque training data and torque verification data; The first neural network is trained using the torque training data to obtain an updated first neural network. The updated first neural network is validated using the torque verification data to obtain the second loss function of the updated first neural network, and the decrease rate of the second loss function within a preset number of training periods is determined. In response to determining that the decrease in the second loss function is less than a preset first magnitude threshold, the updated first neural network is used as the torque prediction model.

5. The method according to claim 1, characterized in that, The step of using a pre-trained terrain recognition model to determine terrain resistance parameters based on the environmental data includes: Using the generator in the terrain recognition model, a noise vector is added to the environmental data to obtain a terrain feature map and a resistance feature vector; Using the discriminator in the terrain recognition model, the terrain feature map is processed to obtain a first distribution probability, and the resistance feature vector is processed to obtain a second distribution probability. The terrain type with the highest distribution probability is determined from the terrain feature map, and the terrain resistance feature with the highest distribution probability is determined from the resistance feature vector. The terrain type and the terrain resistance feature are used as the terrain resistance parameter.

6. The method according to claim 1, characterized in that, The pre-training process of the terrain recognition model includes: Acquire sample environmental data of the vehicle under various driving conditions, and use the sample environmental data to train the discriminator in the second neural network to obtain an updated discriminator; Obtain the sample noise vector, and use the sample noise vector to train the generator in the second neural network to obtain the updated generator; A terrain recognition model is constructed based on the updated discriminator and the updated generator.

7. The method according to claim 1, characterized in that, The method of determining the load risk level of the vehicle transmission structure using a pre-trained risk prediction model based on the predicted peak torque and the terrain resistance parameters includes: Using a pre-trained risk prediction model, the predicted peak torque and the terrain resistance parameters are aligned to obtain a joint feature vector; Using the feature transformer in the risk prediction model, key features that are correlated with the load risk of the vehicle transmission structure are extracted from the joint feature vector; The load risk level of the vehicle transmission structure is determined based on the aforementioned key features.

8. The method according to claim 1, characterized in that, The pre-training process of the risk prediction model includes: Acquire sample torque peak data and sample terrain resistance data, and align the sample torque peak data and sample terrain resistance data to obtain sample joint features; Using the third neural network, the predicted load risk of the vehicle transmission structure is determined based on the joint features of the samples, and the cross-entropy loss function is determined based on the predicted load risk. The updated third neural network is obtained by adjusting the model parameters of the third neural network based on the cross-entropy loss function. In response to the determination that the decrease in the cross-entropy loss function is less than a preset second magnitude threshold within a preset number of training periods, the updated third neural network is used as the risk prediction model.

9. The method according to claim 1, characterized in that, The control of the vehicle based on the load risk level includes: In response to determining that the load risk level is low risk, the vehicle is controlled to maintain its current state; In response to determining that the load risk level is medium risk, an early warning of load risk in the vehicle transmission structure is triggered, and the load risk event is recorded. In response to determining that the load risk level is high risk, an alarm for the load risk of the vehicle transmission structure is triggered, and the current power parameters of the vehicle are controlled within the preset power parameter range.

10. A vehicle, characterized in that, The device includes an electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the program, implements the method as claimed in any one of claims 1 to 9.

Images