Vehicle energy management methods, electronic devices and vehicles

Abstract

This application provides a vehicle energy management method, electronic device, and vehicle, relating to the field of vehicle energy recovery technology. The method includes: acquiring vehicle driving data over a period of time; generating a predicted slope sequence for a future period based on the vehicle driving data using a pre-trained slope prediction model; generating a recommended action sequence and corresponding benefit scores using a benefit evaluation model based on the predicted slope sequence and the current vehicle driving data; and generating an output power sequence for controlling the motor using a pre-trained control model based on the predicted slope sequence, the recommended action sequence, the corresponding benefit scores, and the current vehicle driving data. The output power sequence allows for dynamic adjustment of the vehicle motor's power output and energy recovery intensity, balancing the vehicle's power and economy, and improving the intelligence, adaptability, and efficiency of the vehicle's energy recovery strategy.

Description

Technical Field

[0001] This application relates to the field of vehicle energy recovery technology, and more particularly to a vehicle energy management method, electronic device, and vehicle. Background Technology

[0002] In the field of new energy vehicle technology, energy management strategies can coordinate the power distribution of a vehicle, directly impacting overall vehicle energy consumption, battery life, and driving experience. However, existing new energy vehicles typically employ static power output and energy recovery strategies, resulting in poor adaptive energy recovery capabilities under complex road conditions, thus affecting overall vehicle ride comfort. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a vehicle energy management method, electronic equipment and vehicle.

[0004] To achieve the above objectives, the first aspect of this application provides a vehicle energy management method, comprising:

[0005] Obtain vehicle driving data over a period of time;

[0006] Based on the vehicle driving data, a predicted slope sequence for a future period is generated using a pre-trained slope prediction model.

[0007] Based on the predicted slope sequence and the vehicle driving data at the current moment, a recommended action sequence and corresponding benefit score are generated through a benefit evaluation model.

[0008] Based on the predicted slope sequence, the recommended action sequence and the corresponding profit score, and the vehicle driving data at the current moment, the output power sequence for controlling the motor is generated through the pre-trained control model.

[0009] In some embodiments, the vehicle driving data includes image data, inertial navigation data, and navigation positioning data; the step of generating a predicted slope sequence for a future period based on the vehicle driving data using a pre-trained slope prediction model includes:

[0010] Align the image data, the inertial navigation data, and the navigation and positioning data;

[0011] Based on the aligned image data, the inertial navigation data, and the navigation and positioning data, fusion features corresponding to each frame of image data are generated according to a preset time window;

[0012] Based on the fusion features, the slope prediction model generates a predicted slope sequence for a future period of time.

[0013] In some embodiments, the fused features include image features, inertial navigation features, and navigation and positioning features; the step of generating a predicted slope sequence for a future period based on the fused features and the slope prediction model includes:

[0014] The image features are extracted using the convolutional neural network in the slope prediction model to obtain a two-dimensional feature map;

[0015] The two-dimensional feature map, the inertial navigation feature, and the navigation and positioning feature are concatenated to obtain an embedding sequence;

[0016] The embedded sequence is modeled using the attention mechanism in the slope prediction model to generate the predicted slope sequence.

[0017] In some embodiments, generating a recommended action sequence and corresponding benefit score based on the predicted slope sequence and the vehicle driving data at the current time using a benefit evaluation model includes:

[0018] Based on the predicted slope sequence and the vehicle driving data at the current moment, multiple state nodes are constructed, and adjacent state nodes correspond to a control action and the benefit score generated by the control action.

[0019] Based on the profit score, a control path corresponding to the maximum cumulative profit score is generated among multiple state nodes using a dynamic path planning algorithm. All control actions included in the control path are used as the recommended action sequence, and the recommended action sequence and the corresponding profit score are output.

[0020] In some embodiments, generating an output power sequence for controlling the motor using a pre-trained control model based on the predicted slope sequence, the recommended action sequence and the corresponding reward score, and the vehicle driving data at the current moment includes:

[0021] Based on the predicted slope sequence and the vehicle driving data at the current moment, a recommended driving power is generated through the multilayer perceptron structure in the control model.

[0022] Based on the recommended action sequence and the corresponding benefit score, the predicted energy recovery intensity is generated through the gated loop unit in the control model.

[0023] Based on the recommended drive power and the predicted energy recovery intensity, the output power sequence is generated through the nonlinear mapping layer in the control model.

[0024] In some embodiments, before generating the output power sequence for controlling the motor using a pre-trained control model based on the predicted slope sequence, the recommended action sequence, the corresponding reward score, and the vehicle driving data at the current moment, the process includes:

[0025] Determine the corresponding preset parameter template based on the vehicle's current driving mode;

[0026] Adjust the model parameters of the control model according to the preset parameter template.

[0027] In some embodiments, the method further includes:

[0028] In response to determining that the output power sequence does not meet the preset constraint conditions, the output power sequence is adjusted.

[0029] In some embodiments, the method further includes:

[0030] The output power sequence and the initial power sequence generated by the vehicle control unit are fused according to the current vehicle operating conditions to obtain a fused power sequence, so that the motor operates according to the fused power sequence.

[0031] Based on the same inventive concept, a second aspect of this application also provides 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.

[0032] Based on the same inventive concept, a third aspect of this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method described above.

[0033] As described above, this application provides a vehicle energy management method, electronic device, and vehicle. The method includes: acquiring vehicle driving data over a period of time; and generating a predicted slope sequence for a future period based on the vehicle driving data using a pre-trained slope prediction model. Slope prediction allows for understanding the future trend of road slope changes, which is beneficial for improving the vehicle's energy management adaptability in complex road environments. Based on the predicted slope sequence and the current vehicle driving data, a recommended action sequence and corresponding benefit scores are generated using a benefit evaluation model, achieving a connection between road slope perception and control actions, and providing effective data support for the control model. Based on the predicted slope sequence, the recommended action sequence, the corresponding benefit scores, and the current vehicle driving data, a pre-trained control model generates an output power sequence for controlling the motor. The output power sequence allows for dynamic adjustment of the vehicle motor's power output and energy recovery intensity, balancing the vehicle's power and economy, and improving the intelligence, adaptability, and efficiency of the vehicle's energy recovery strategy. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating the vehicle energy management method according to an embodiment of this application;

[0036] Figure 2 This is a schematic diagram of the structure of a vehicle energy management device according to an embodiment of this application;

[0037] Figure 3 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0039] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application 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 after 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 only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0040] As described in the background section, existing new energy vehicles typically employ static power output and energy recovery strategies. When dealing with complex conditions such as undulating roads, inclines, declines, and low-adhesion surfaces, their control logic often reveals inherent defects of staticity and lag. Vehicle energy management systems generally rely on rule-based or linear interval-based control strategies, and their decision-making mechanisms are essentially "posterior-oriented"—that is, passively adjusting the intensity of power output and energy recovery based solely on the current instantaneous state of the vehicle (such as vehicle speed, accelerator pedal opening, and battery state of charge). This energy management approach lacks real-time response capabilities to complex road structures and vehicle dynamic states, cannot effectively perceive future road gradient trends, and cannot dynamically optimize energy gains based on different operating conditions.

[0041] In addition, existing vehicle energy management systems have limited processing capabilities in multi-source data fusion, making it impossible to accurately model the impact of slope changes on the vehicle's dynamic behavior. This results in unstable energy recovery efficiency, delayed or frequent oscillations in control command response, and affects the overall vehicle's ride comfort and energy efficiency.

[0042] In view of this, this application proposes a vehicle energy management method, which generates an output power sequence for controlling the motor by real-time slope trend perception and vehicle action benefit evaluation, thereby realizing the dynamic response of motor output power, improving the vehicle's adaptive ability under complex road conditions, and also improving the intelligence level and efficiency of vehicle energy recovery.

[0043] The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0044] This application provides a vehicle energy management method, referring to... Figure 1 This includes the following steps:

[0045] Step 102: Obtain vehicle driving data over a period of time.

[0046] Specifically, vehicle driving data is obtained by collecting data from multiple sources of sensors on the vehicle over a period of time. This driving data includes image data, inertial navigation data, navigation and positioning data, electric drive data, and braking data. Image data is acquired through the vehicle's forward-facing onboard camera, with a sampling frequency of 20 frames per second. The image data includes features such as road texture, undulations, and boundaries. A synchronous triggering mechanism is used during image data acquisition, ensuring data is collected on the same hardware clock edge as other sensors, avoiding timing discrepancies. Inertial navigation data is acquired using an automotive-grade six-axis inertial navigation sensor to capture changes in vehicle attitude, including longitudinal acceleration, angular velocity, and pitch angle, with a sampling frequency of at least 50Hz. Navigation and positioning data is collected using GPS (Global Positioning System) and RTK (Real-time kinematic) high-precision positioning equipment, including vehicle position, speed, and heading. Electric drive data is acquired through the vehicle's electric drive system. This system obtains information such as motor output power, motor torque, and energy recovery status via CAN (Controller Area Network) messages reported by the motor controller. Braking data is acquired through the vehicle's braking system, including brake pedal travel and energy recovery level. All vehicle driving data is timestamped by the central domain controller and interpolated or downsampled according to the sampling frequency, ultimately forming synchronized data that includes image data, inertial navigation data, navigation and positioning data, electric drive data, and braking data.

[0047] Step 104: Based on the vehicle driving data, generate a predicted slope sequence for a future period of time using a pre-trained slope prediction model.

[0048] Specifically, slope prediction models can achieve high-precision prediction of continuous slope trends ahead of vehicles. These models are neural network structures that simultaneously possess local texture perception and temporal dependency modeling capabilities. They can not only extract road spatial features from image data but also capture the continuous changing trend of slope over time through sequence modeling structures, ultimately enabling slope prediction for each moment within a future period. For example, a slope prediction model can be a ConvFormer model, which balances the local feature extraction capabilities of CNNs (Convolutional Neural Networks) with the global modeling capabilities of Transformers.

[0049] Step 106: Based on the predicted slope sequence and the vehicle driving data at the current moment, generate a recommended action sequence and corresponding benefit score through the benefit evaluation model.

[0050] Specifically, the benefit assessment model can evaluate the long-term energy benefits of vehicles under different control strategies. It quantifies the level of energy benefits through a benefit score; the higher the benefit score, the greater the energy benefit. The action sequences differ under different control strategies, and these sequences include various vehicle control actions, such as gentle acceleration, speed maintenance, gentle braking, and strong regenerative braking.

[0051] Based on the predicted slope sequence and the vehicle driving data at the current moment, the benefit score corresponding to different action sequences is evaluated through a benefit assessment model. The action sequence that can achieve the best energy efficiency under the current and predicted road condition slope is selected as the recommended action sequence, providing clear indication information for the subsequent control model.

[0052] The benefit evaluation model is a dynamic programming structure based on a state transition graph, comprising multiple state nodes, each representing a possible combination of slope, velocity, and energy states. Each action sequence corresponds to a possible path in the benefit evaluation model, and each path includes multiple state nodes. During path optimization, the path corresponding to the maximum cumulative benefit score is selected as the control path, and the actions corresponding to all state nodes in the control path form a recommended action sequence. The benefit evaluation model outputs the recommended action sequence and its corresponding benefit score.

[0053] A recommended action sequence is a sequence of actions that suggests actions to be performed at multiple future time points. Each suggested action corresponds to a benefit score, which indicates the amount of energy recovered from performing that suggested action.

[0054] Step 108: Based on the predicted slope sequence, the recommended action sequence and the corresponding profit score and the vehicle driving data at the current moment, generate the output power sequence for controlling the motor through the pre-trained control model.

[0055] Specifically, the control model jointly models the predicted slope sequence, recommended action sequence, corresponding benefit scores, and vehicle driving data at the current moment, and outputs an output power sequence suitable for controlling the motor under the current working conditions.

[0056] The control model employs a multi-branch parallel structure, with different branches processing the predicted slope sequence, recommended action sequence and corresponding reward scores, and the vehicle's current driving data, thus achieving data decoupling. The generated output power sequence is a continuous scalar, with positive values ​​representing the motor's forward output power and negative values ​​representing the degree of energy recovery. The motor operation is controlled by this output power sequence to achieve energy management for the vehicle.

[0057] Based on steps 102 to 108 above, this embodiment provides a vehicle energy management method, including: acquiring vehicle driving data over a period of time; and generating a predicted slope sequence for a future period based on the vehicle driving data using a pre-trained slope prediction model. Slope prediction allows for understanding the future trend of road slope changes, which is beneficial for improving the vehicle's energy management adaptability in complex road environments. Based on the predicted slope sequence and the current vehicle driving data, a recommended action sequence and corresponding benefit score are generated using a benefit evaluation model, achieving a connection between road slope perception and control actions, providing effective data support for the control model. Based on the predicted slope sequence, the recommended action sequence, the corresponding benefit score, and the current vehicle driving data, a pre-trained control model generates an output power sequence for controlling the motor. The output power sequence allows for dynamic adjustment of the vehicle motor's power output and energy recovery intensity, balancing the vehicle's power and economy, and improving the intelligence, adaptability, and efficiency of the vehicle's energy recovery strategy.

[0058] In some embodiments, the vehicle driving data includes image data, inertial navigation data, and navigation positioning data; the step of generating a predicted slope sequence for a future period based on the vehicle driving data using a pre-trained slope prediction model includes:

[0059] Align the image data, the inertial navigation data, and the navigation and positioning data;

[0060] Based on the aligned image data, the inertial navigation data, and the navigation and positioning data, fusion features corresponding to each frame of image data are generated according to a preset time window;

[0061] Based on the fusion features, the slope prediction model generates a predicted slope sequence for a future period of time.

[0062] Specifically, a single data source cannot accurately represent changes in road slope. For example, image data is easily affected by lighting, weather, occlusion, and ground texture, limiting its ability to perceive slope information. Inertial navigation data measures the vehicle's own motion and cannot distinguish between pitch changes caused by slope and pitch changes caused by acceleration / braking. Navigation and positioning data has a low update frequency, is prone to signal loss in tunnels and urban canyons, and has limited accuracy. Therefore, this embodiment fuses image data, inertial navigation data, and navigation and positioning data to construct a fused feature with continuity and spatial representation capabilities. This feature serves as the input to the subsequent slope prediction model, improving the model's robustness.

[0063] First, the acquired image data undergoes preprocessing, including distortion correction, cropping, and normalization. Then, the image data, inertial navigation data, and navigation and positioning data are synchronized and registered to ensure that each frame of image data is matched with its corresponding inertial navigation and navigation and positioning data with a strictly corresponding timestamp. Through timestamp alignment, all data is unified to the time reference of the main controller.

[0064] For each frame of image data, the two adjacent frames on the time axis are introduced to form a time window containing five frames of image data. Linear interpolation is used to supplement the inertial navigation (INS) and navigation / positioning (NO) data corresponding to each frame of image data within the time window. Thus, for each frame of image data, a fused sequence is formed, containing five frames of image data, an INS data sequence, and a NOW / NO ...

[0065] To adapt to the input format of the slope prediction model, the fusion sequence needs to be converted into fusion features. For each fusion sequence, a low-dimensional spatial feature map is extracted from the image data using a convolutional neural network. The pitch angle change in the inertial navigation data is embedded as a position encoding signal into the image channel, enhancing the physical meaning of the low-dimensional spatial feature map and forming a fusion feature map. The speed change trend in the positioning and navigation data is used as a slope change cue and input into the fusion feature map to help distinguish between real slopes and attitude changes caused by braking / acceleration, ultimately forming a fusion feature with a temporal structure. For example, if the vehicle pitch angle increases while the speed decreases rapidly, it is likely due to sudden braking on a flat road. If the vehicle pitch angle increases but the speed remains stable or decreases slowly, it is more likely due to going uphill.

[0066] After determining the fusion features, these features are input into the slope prediction model, which generates a predicted slope sequence for a future period. Furthermore, to improve the robustness of the slope prediction model, data augmentation processing can be performed on image data, inertial navigation data, and navigation positioning data. For image data, perturbations such as brightness, shadows, and blur are added to simulate different climate and lighting environments. For inertial navigation data, small noise perturbations are introduced to improve the model's tolerance to hardware biases. For positioning and navigation data, random offsets are applied to enhance the model's robustness to positioning errors. These data augmentation processes can effectively improve the adaptability of the subsequent slope prediction model in complex road environments.

[0067] In this embodiment, by fusing image data, inertial navigation data, and navigation positioning data, the perception dimension of slope in a single frame image data is improved, resulting in fused features with image semantics, vehicle attitude, and speed trends. This provides a coherent basis for slope trend judgment for the slope prediction model, effectively improving the accuracy of slope prediction.

[0068] In some embodiments, the fused features include image features, inertial navigation features, and navigation and positioning features; the step of generating a predicted slope sequence for a future period based on the fused features and the slope prediction model includes:

[0069] The image features are extracted using the convolutional neural network in the slope prediction model to obtain a two-dimensional feature map;

[0070] The two-dimensional feature map, the inertial navigation feature, and the navigation and positioning feature are concatenated to obtain an embedding sequence;

[0071] The embedded sequence is modeled using the attention mechanism in the slope prediction model to generate the predicted slope sequence.

[0072] Specifically, the fused feature time series is fixed at 5 frames, covering the vehicle's trajectory within 1 second. Each frame contains image features, inertial navigation features, and navigation positioning features. The slope prediction model consists of two parts: a convolutional neural network and a multi-layer Transformer structure. The convolutional neural network uses a lightweight structure to extract image features, extracting edges, textures, and spatial variation patterns through multi-layer convolution and normalization operations to capture the undulating contours of the road surface and output a two-dimensional feature map. The two-dimensional feature map is then concatenated with the inertial navigation features and navigation positioning features along the time axis to obtain the embedding sequence.

[0073] The Transformer architecture is based on a multi-layered encoder stack and employs a self-attention mechanism to model the input embedding sequence. It analyzes the intrinsic relationships between data points at each time step in the embedding sequence to capture dynamic patterns of slope changes. When the slope ahead is gradually rising or falling, the Transformer architecture can recognize continuous visual and gestural cues and form a reasonable prediction of the future slope. The output of the Transformer architecture includes a multi-step regression structure, which ultimately maps to specific slope values ​​at multiple future time points through a regression layer, forming a predicted slope sequence. The predicted slope sequence is a continuous scalar sequence, for example, including 10 slope values ​​with a 200ms interval between adjacent values. The output dimension of the predicted slope sequence is configurable and can be adjusted according to the vehicle control response frequency and model latency to ensure real-time requirements are met.

[0074] It should be noted that the slope prediction model employs a supervised learning method during training. The model training data comes from multimodal data samples collected from real roads. Fusion feature samples are obtained from these multimodal data samples. The true slope label corresponding to the fusion feature sample is not a direct measurement value, but rather calculated by fusing high-precision map data and inertial navigation data. The high-precision map provides prior absolute slope, while the inertial navigation data provides verification of vehicle transient motion; the combination of these two forms a reliable true slope label. The loss function during training can use mean squared error or mean absolute error to minimize the difference between the slope sequence predicted by the model and the true slope sequence. Furthermore, regularization constraints on prediction continuity and smoothness are introduced during the slope prediction model training process to reduce high-frequency fluctuations in the predicted slope during control. Simultaneously, a data distribution balancing mechanism is introduced into the training data to improve the model's adaptability to different terrain samples (uphill, downhill, and flat roads), ensuring that the model has a balanced learning ability for all road conditions.

[0075] The slope prediction model possesses modeling capabilities across both spatial and temporal dimensions. Convolutional neural networks capture road geometry features, while the Transformer structure handles temporal continuity, enabling the model to stably output predicted slope sequences even in complex road scenarios. In conditions such as urban elevated roads and mountain curves, compared to traditional convolutional networks, the slope prediction model can more accurately predict changes in road undulations ahead, providing a more forward-looking input signal for the vehicle's energy management system, thereby improving the accuracy of vehicle energy management.

[0076] In some embodiments, generating a recommended action sequence and corresponding benefit score based on the predicted slope sequence and the vehicle driving data at the current time using a benefit evaluation model includes:

[0077] Based on the predicted slope sequence and the vehicle driving data at the current moment, multiple state nodes are constructed, and adjacent state nodes correspond to a control action and the benefit score generated by the control action.

[0078] Based on the profit score, a control path corresponding to the maximum cumulative profit score is generated among multiple state nodes using a dynamic path planning algorithm. All control actions included in the control path are used as the recommended action sequence, and the recommended action sequence and the corresponding profit score are output.

[0079] Specifically, future time is discretized into multiple time steps, each step (e.g., 0.2s) representing a stage and corresponding to a slope value in a predicted slope sequence. For each stage, a set of discrete states is defined based on the vehicle's driving data at the current moment, forming multiple state nodes. Each state node typically includes vehicle speed and cumulative energy change. The vehicle driving data includes vehicle speed, battery state of charge, motor torque, drive axle load, and ground adhesion estimation. Ground adhesion estimation can serve as safety constraint information. For example, on low-traction surfaces, real-time strong energy recovery may cause vehicle skidding, leading to safety hazards. Ground adhesion estimation can help eliminate the generation of high-risk control actions.

[0080] Each state node represents a possible speed or energy state of the vehicle at that moment. The state transition from one state node to the next represents a control action, such as: gentle acceleration, maintaining speed, gentle braking, or strong energy recovery. For each control action, a physical model is used to predict the benefit of performing the action. The physical model includes a power model and an energy consumption model. The power model calculates the driving power required for the action or the power generated by energy recovery. In the energy consumption model, constraints such as electric drive efficiency, battery charging and discharging efficiency, and motor temperature rise are comprehensively considered. Based on the driving power or the power recovered, the net energy benefit of the action is calculated (negative values ​​indicate consumption, positive values ​​indicate recovery). This net energy benefit is the benefit of the action corresponding to the state transition process, and corresponds to a benefit score.

[0081] Starting from the current state node, all possible paths are explored forward along the time axis of the predicted slope sequence. At each stage, the cumulative reward from the starting state node to each state node in the current stage is calculated. Using a dynamic programming algorithm, the path corresponding to the maximum cumulative reward score from the starting state node to the ending state node is found and designated as the control path. All control actions included in the control path are combined to form the recommended action sequence. The recommended action sequence and the reward score corresponding to each control action are output through a reward evaluation model. The recommended action sequence includes actions suggested to be executed at multiple future time points, such as [maintain speed, light recovery, strong recovery, maintain speed, etc.].

[0082] Furthermore, the benefit assessment model can also output an estimated recovered energy index, which represents the total energy recovered by the vehicle after executing the recommended action sequence. The model can also output a risk level index, including braking frequency, motor temperature increase, and battery SOC fluctuation. Braking frequency and motor temperature increase reflect the impact of the recommended action sequence on vehicle hardware lifespan and comfort. Battery SOC fluctuation reflects the impact of the recommended action sequence on battery state.

[0083] The recommended action sequence can not only be used as input for subsequent control models, but also as an emergency control strategy for independently driven vehicle controllers in special scenarios. The benefit evaluation model has a certain degree of adaptability. When the slope prediction model fluctuates or becomes unstable, a dynamic path planning algorithm can be used to find the control path with the optimal global energy benefit, resulting in a series of continuous and smooth recommended action sequences. This eliminates noise interference from the predicted slope sequence, ensuring that the vehicle maintains stable control at all times.

[0084] The benefit evaluation model can accurately determine the recommended sequence of actions that will generate the maximum benefit for the vehicle over a future period, providing a clear optimization objective for the subsequent control network. Combining a physical model and a dynamic path planning algorithm makes the decisions and benefits at each step clearly traceable, meeting automotive-grade requirements for safety and traceability.

[0085] In some embodiments, generating an output power sequence for controlling the motor using a pre-trained control model based on the predicted slope sequence, the recommended action sequence and the corresponding reward score, and the vehicle driving data at the current moment includes:

[0086] Based on the predicted slope sequence and the vehicle driving data at the current moment, a recommended driving power is generated through the multilayer perceptron structure in the control model.

[0087] Based on the recommended action sequence and the corresponding benefit score, the predicted energy recovery intensity is generated through the gated loop unit in the control model.

[0088] Based on the recommended drive power and the predicted energy recovery intensity, the output power sequence is generated through the nonlinear mapping layer in the control model.

[0089] Specifically, the predicted slope sequence reflects the future undulation trend of the road segment, providing a future contextual environment, which is the foundation for the forward-looking nature of vehicle energy control. The recommended action sequence and corresponding benefit score provide optimization goals for energy control. Current vehicle driving data can be used as the basis for adjusting the control model, ensuring that the generated output power sequence is feasible and safe. For example, even if there is a downhill slope ahead, strong energy recovery cannot be performed if the battery SOC in the vehicle driving data is high.

[0090] The control model is a neural network model, comprising three functional branches. The first branch is a multilayer perceptron module, which employs a multilayer perceptron architecture. This architecture processes the predicted gradient sequence and the vehicle's current driving data, outputting a recommended drive power to determine the motor's output requirements. By processing the recommended action sequence and its corresponding reward score separately, the multilayer perceptron module can focus more on scenarios such as acceleration and hill climbing, providing suggestions on the appropriate drive power output from the motor based solely on the predicted gradient sequence and the current vehicle driving data.

[0091] The second functional branch is the energy recovery intensity prediction module. This module employs a gated loop unit structure to process recommended action sequences with temporal information and corresponding benefit scores. The gated loop unit structure has excellent modeling capabilities for the multi-step coherence of energy recovery, avoiding abrupt changes in energy recovery commands. The predicted energy recovery intensity is generated through this module.

[0092] The third functional branch is the policy fusion module, which includes a set of nonlinear mapping layers. The policy fusion module receives the recommended driving power and the predicted energy recovery intensity, performs deep fusion on these two parameters, and generates an output power sequence, thus avoiding the decision conflict problem that occurs in traditional multi-model structures.

[0093] The output power sequence is a continuous scalar, with positive values ​​representing the motor's forward output power and negative values ​​representing the energy recovery intensity. This output power sequence creates a seamless and smooth control, avoiding the vehicle jerking sensation caused by switching between drive and recovery modes in traditional energy control systems.

[0094] After the output power sequence is generated, it will undergo interpolation and boundary constraint processing (such as power upper and lower limit constraints, current safety threshold constraints, and SOC boundary constraints) to ensure that it meets the actual control interface requirements of the vehicle electric drive system and generate a safe and reliable output power sequence as a control command.

[0095] During the training of the control model, a hybrid strategy combining supervised learning and reinforcement learning is employed. In the supervised learning phase, the actual motor output power from real-vehicle data is used as a label to train the model's predictive ability under typical road conditions. In the reinforcement learning phase, a policy gradient optimization method is introduced, using energy minimization, energy recovery maximization, and control smoothness as reward functions to continuously optimize the policy fusion effect in a simulation environment. Control stability constraints are added to the loss function to prevent the control model from outputting high-frequency jitter or physically infeasible output power.

[0096] The control model can adjust the allocation ratio of driving power and energy recovery in real time, rather than statically executing fixed rules. Based on the predicted gradient sequence, recommended action sequence, corresponding benefit scores, and current vehicle driving data, it finds the globally optimal balance point between power, economy, and comfort, and then generates the output power sequence. The neural network structure of the control model has an extremely fast forward propagation speed, achieving millisecond-level response. During the training phase, the smoothness of the control model's output commands is ensured through the reasonable setting of reward and loss functions, fundamentally eliminating the lag and oscillation of control commands.

[0097] In some embodiments, before generating the output power sequence for controlling the motor using a pre-trained control model based on the predicted slope sequence, the recommended action sequence, the corresponding reward score, and the vehicle driving data at the current moment, the process includes:

[0098] Determine the corresponding preset parameter template based on the vehicle's current driving mode;

[0099] Adjust the model parameters of the control model according to the preset parameter template.

[0100] Specifically, since the energy control strategies differ under different driving modes (e.g., energy-saving mode prioritizes energy recovery, while sport mode prioritizes power response), this embodiment predetermines different preset parameter templates for different driving modes. Each preset parameter template defines a set of model parameters corresponding to the control model. By predicting the parameter templates, the model parameters in the control model can be switched, such as changing certain key weight parameters, thereby achieving control model switching under different driving modes without changing the control model structure.

[0101] Therefore, before calling the control model, a corresponding preset parameter template is determined based on the current driving mode. The model parameters of the control model are adjusted through the preset parameter template. Then, the output power sequence is generated based on the control model after parameter adjustment, so that the output power sequence can better match the current driving mode of the vehicle and provide the vehicle with a better energy management strategy.

[0102] In some embodiments, the method further includes:

[0103] In response to determining that the output power sequence does not meet the preset constraint conditions, the output power sequence is adjusted.

[0104] Specifically, to further ensure the controllability, adjustability, and stability of the output power sequence under any scenario, this embodiment sets up a strategy supervision module to determine whether the output power sequence meets preset constraints. If the output power sequence meets the preset constraints, it indicates that the output power sequence is a safe and stable control command; if the output power sequence does not meet the preset constraints, it indicates that the output power sequence is an unsafe or unstable control command.

[0105] The preset constraints include a series of hard boundaries set for the power battery, motor control system, and braking system. These include the battery state of charge range, the upper limit of the motor's allowable output power, the maximum energy recovery intensity, and acceleration / deceleration comfort limits. These constraints allow for the filtering of the output power sequence. If the output power sequence exceeds the range of the preset constraints, it needs to be adjusted, such as by using amplitude limiting or strategy replacement. The method in this embodiment further ensures that the output power sequence meets the safety operating standards of the vehicle control system, possessing controllability, adjustability, and stability in complex real-world environments.

[0106] In some embodiments, the method further includes:

[0107] The output power sequence and the initial power sequence generated by the vehicle control unit are fused according to the current vehicle operating conditions to obtain a fused power sequence, so that the motor operates according to the fused power sequence.

[0108] Specifically, in actual vehicle control, to ensure compatibility between the control model's output and the VCU (Vehicle Control Unit) power scheduling system, a soft fusion mechanism is implemented in this embodiment. Specifically, the output power sequence generated by the control model does not directly replace VCU decisions; instead, it is synthesized with the VCU's initial power sequence through interpolation. The fusion ratio can be dynamically adjusted according to the current operating conditions. For example, when the gradient changes rapidly, the control model output takes precedence, increasing the proportion of the output power sequence; under stable operating conditions, the VCU's initial power sequence takes precedence, increasing the proportion of the initial power sequence. During normal vehicle operation, the vehicle is controlled by the VCU. If the system were to directly switch to controlling the motor via the output power sequence, the user would clearly perceive a change in vehicle control, leading to low user acceptance. The fusion method in this embodiment improves the interpretability and user acceptance of the vehicle control system, enhancing driving comfort and safety.

[0109] Building upon the aforementioned embodiments, to ensure the real-time performance and stability of the data stream, a specific data synchronization buffer structure is established between different models. The slope prediction model and the benefit evaluation model are calculated 5 times / second, while the control model is updated 10 times / second. By designing a sliding time window structure to store the predicted slope sequence, the recommended action sequence, and the corresponding benefit score, the control model can retrieve and fuse data from the most recent time window at any time. This ensures both the real-time performance of the control response and avoids policy response lag caused by the low frame rate of the slope prediction model and the benefit evaluation model's output data.

[0110] In the multi-model fusion mechanism of this application, the slope prediction model, benefit evaluation model, and control model are optimized and deployed independently, while efficient collaboration between models is achieved through information interfaces and standardized input formats. When updating models or replacing strategies, it is not necessary to retrain all models; only the parameters of the corresponding model need to be updated to achieve flexible iterative upgrades. Especially in automotive-grade deployments, this decoupled structure greatly facilitates algorithm testing and verification, functional safety assessment, and strategy switching control.

[0111] To address perception anomalies or strategy failures in complex scenarios, a degradation mechanism is incorporated into the fusion of the three models. When the predicted slope sequence fluctuates drastically or the benefit assessment shows a non-convergent trend, the control model automatically invokes a set of preset empirical control parameters to cover the output strategy under typical operating conditions, ensuring the safe operation of the vehicle. This degradation mechanism enhances the fault tolerance of the vehicle control system under extreme road conditions or sensor failures, meeting the actual requirements of automakers for stability and safety.

[0112] Through multi-model fusion design, a closed-loop linkage from perception to assessment to control is achieved, which not only has high scalability but also good maintainability. The multi-model fusion design not only improves the adaptability of the control strategy under dynamic road conditions but also provides a technical foundation and integration convenience for subsequent deployment and mass production.

[0113] Regarding redundant control logic, multiple strategy switching triggers are set up, such as triggers for abnormal fluctuations in slope prediction, non-convergence of revenue scoring, battery SOC threshold, and sensor signal interruption. When any of these situations occur, the energy control system immediately abandons the model output and switches to a set of preset hard rule control logic. This logic covers typical scenarios such as common undulations, downhill slopes, and low-adhesion conditions, and implements static power allocation based on engineering calibration experience. The redundant modules corresponding to the redundant control logic are completely independent of any model inference path, deployed in the local controller, and possess high reliability and low latency characteristics, ensuring that the vehicle still has basic control capabilities under extreme conditions.

[0114] The energy control system also provides a strategy output monitoring interface, which can record and visualize the output power sequence of the control model, facilitating verification and playback by vehicle testing technicians during development. The traceability of the control model's output behavior is crucial for meeting vehicle functional safety analysis and software update compliance requirements. The design of the strategy output monitoring interface reflects the high level of emphasis placed on safety, stability, and strategy controllability in automotive-grade control systems.

[0115] In this application, the model execution platform uses the vehicle's central domain controller, which features a high-performance multi-core SoC (Multi-Core System on Chip) capable of supporting parallel tasks such as image processing, neural network inference, and real-time control decision-making. During deployment, the slope prediction model is deployed in a sub-processor equipped with a tensor acceleration engine to handle the convolution and Transformer computations from camera images. Since this module needs to process image sequences within a time window, image cache management and asynchronous computation scheduling mechanisms must be enabled during deployment to ensure processing latency is kept below 200 milliseconds. The profit evaluation model, a low-density but logically complex strategy model, is deployed on a general-purpose CPU core, employing a periodic calculation mechanism with a computation frequency of approximately five times per second. Its output results are cached for subsequent access by the control model inference module.

[0116] As the final control command generation module, the control model has the highest real-time requirements. Deployed in a control thread with RTOS (Real-Time Operated System) support, it operates at a frequency of no less than ten times per second. Its output must be transmitted to the vehicle's VCU or electric drive controller via a low-latency shared memory mechanism. The entire model chain, from vehicle driving data input to control output, has a maximum response latency controlled within 300 milliseconds, meeting the closed-loop control cycle requirements of mainstream passenger vehicle powertrain control systems.

[0117] In terms of system communication, the model exchanges data with each controller via the CAN-FD (CAN with Flexible Data Rate) bus. The forward-facing camera image input signal is connected to the central controller via Ethernet, while inertial navigation, GPS positioning, battery status, motor output, and current information are reported in real time via CAN-FD. The output power sequence from the control model must be packaged according to the vehicle's VCU data protocol, including fields such as control intent identifier, power command, motor torque distribution intent, and recovery level, and then encoded before being written to the controller's target address. To meet functional safety requirements, each control signal must be accompanied by a CRC checksum before and after writing, and an acknowledgment mechanism must be established.

[0118] In terms of software architecture, all models are deployed as containerized modules, supporting dynamic loading and version rollback, facilitating hot model updates via the vehicle's OTA system. The container integrates a model engine, parameter manager, monitoring module, and redundancy detection module. The parameter manager is used for runtime hyperparameter tuning of the model, while the monitoring module monitors model runtime latency, input validity, output boundaries, and other information in real time. If an anomaly is detected, it triggers a model shutdown signal and sends a policy degradation prompt to the VCU.

[0119] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in 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 in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0120] It should be noted that some embodiments of this application have been described above. In some cases, the actions or steps described in the above embodiments can be performed in a different order than that shown in the above embodiments and the desired result can still be achieved. In addition, the processes depicted in the accompanying 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.

[0121] Based on the same inventive concept, corresponding to any of the above-described embodiments, this application also provides a vehicle energy management device.

[0122] refer to Figure 2 The vehicle energy management device includes:

[0123] The acquisition module 202 is configured to acquire vehicle driving data over a period of time;

[0124] Prediction module 204 is configured to generate a predicted slope sequence for a future period of time based on the vehicle driving data and a pre-trained slope prediction model.

[0125] The evaluation module 206 is configured to generate a recommended action sequence and a corresponding benefit score based on the predicted slope sequence and the vehicle driving data at the current time through a benefit evaluation model.

[0126] The control module 208 is configured to generate an output power sequence for controlling the motor based on the predicted slope sequence, the recommended action sequence and the corresponding benefit score and the vehicle driving data at the current moment, using a pre-trained control model.

[0127] In some embodiments, the vehicle driving data includes image data, inertial navigation data, and navigation positioning data; the prediction module 204 is configured to perform data alignment on the image data, the inertial navigation data, and the navigation positioning data; based on the data-aligned image data, the inertial navigation data, and the navigation positioning data, generate fusion features corresponding to each frame of image data according to a preset time window; and based on the fusion features, generate a predicted slope sequence for a future period of time through the slope prediction model.

[0128] In some embodiments, the fused features include image features, inertial navigation features, and navigation and positioning features; the prediction module 204 is configured to extract the image features through a convolutional neural network in the slope prediction model to obtain a two-dimensional feature map; to concatenate the two-dimensional feature map, the inertial navigation features, and the navigation and positioning features to obtain an embedding sequence; and to model the embedding sequence using the attention mechanism in the slope prediction model to generate the predicted slope sequence.

[0129] In some embodiments, the evaluation module 206 is configured to construct multiple state nodes based on the predicted slope sequence and the vehicle driving data at the current time, with adjacent state nodes corresponding to a control action and the benefit score generated by the control action; based on the benefit score, a control path corresponding to the maximum cumulative benefit score is generated among the multiple state nodes through a dynamic path planning algorithm, and all control actions included in the control path are used as the recommended action sequence, and the recommended action sequence and the corresponding benefit score are output.

[0130] In some embodiments, the control module 208 is configured to generate a recommended driving power based on the predicted slope sequence and the vehicle driving data at the current time, through a multilayer perceptron structure in the control model; generate a predicted energy recovery intensity based on the recommended action sequence and the corresponding benefit score, through a gated loop unit in the control model; and generate the output power sequence based on the recommended driving power and the predicted energy recovery intensity, through a nonlinear mapping layer in the control model.

[0131] In some embodiments, before generating the output power sequence for controlling the motor using a pre-trained control model based on the predicted slope sequence, the recommended action sequence, the corresponding benefit score, and the vehicle driving data at the current moment, an adjustment module is further included, configured to determine a corresponding preset parameter template according to the current driving mode of the vehicle; and adjust the model parameters of the control model according to the preset parameter template.

[0132] In some embodiments, the adjustment module is configured to adjust the output power sequence in response to determining that the output power sequence does not meet a preset constraint condition.

[0133] In some embodiments, a fusion module is further included, configured to fuse the output power sequence and the initial power sequence generated by the vehicle control unit according to the current vehicle operating conditions to obtain a fused power sequence, so that the motor operates according to the fused power sequence.

[0134] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0135] The apparatus of the above embodiments is used to implement the corresponding vehicle energy management method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0136] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application 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 energy management method described in any of the above embodiments.

[0137] Figure 3 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0138] 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.

[0139] 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.

[0140] 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.

[0141] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0142] 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.

[0143] 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.

[0144] The electronic devices described above are used to implement the corresponding vehicle energy management methods in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0145] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a vehicle, which includes the electronic equipment described in any of the above embodiments.

[0146] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the vehicle energy management method as described in any of the above embodiments.

[0147] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0148] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the vehicle energy management 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.

[0149] Based on the same concept, corresponding to any of the above embodiments, this application also provides a computer program product, including computer program instructions, which, when run on a computer, cause the computer to perform the method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0150] 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 application is limited to these examples; under the concept of this application, 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 application as described above, which are not provided in detail for the sake of brevity.

[0151] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application 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 application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0152] Although this application 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.

[0153] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A vehicle energy management method, characterized by, The method comprises the following steps: acquiring vehicle driving data in a period of time; the vehicle driving data comprises image data, inertial navigation data and navigation positioning data; based on the vehicle driving data, a slope prediction model pre-trained is used to generate a predicted slope sequence in a future period of time, comprising: based on the image data, the inertial navigation data and the navigation positioning data after data alignment, a fusion feature corresponding to each frame of image data is generated according to a preset time window; based on the fusion feature, the slope prediction model is used to generate a predicted slope sequence in a future period of time; based on the predicted slope sequence and the vehicle driving data at the current time, a reward evaluation model is used to generate a recommended action sequence and a corresponding reward score; based on the predicted slope sequence, the recommended action sequence and the corresponding reward score, and the vehicle driving data at the current time, a control model pre-trained is used to generate an output power sequence for controlling the motor.

2. The method of claim 1, wherein, The fusion feature comprises image features, inertial navigation features and navigation positioning features; the slope prediction model is used to generate a predicted slope sequence in a future period of time based on the fusion feature, comprising: a convolutional neural network in the slope prediction model is used to extract the image features to obtain a two-dimensional feature map; the two-dimensional feature map, the inertial navigation features and the navigation positioning features are spliced to obtain an embedding sequence; an attention mechanism in the slope prediction model is used to model the embedding sequence to generate the predicted slope sequence.

3. The method of claim 1, wherein, The reward evaluation model is used to generate a recommended action sequence and a corresponding reward score based on the predicted slope sequence and the vehicle driving data at the current time, comprising: based on the predicted slope sequence and the vehicle driving data at the current time, a plurality of state nodes are constructed, and adjacent state nodes correspond to a control action and a reward score generated by the control action; based on the reward score, a dynamic path planning algorithm is used to generate a control path corresponding to a maximum cumulative reward score in the plurality of state nodes, all control actions included in the control path are taken as the recommended action sequence, and the recommended action sequence and the corresponding reward score are output.

4. The method of claim 1, wherein, The control model pre-trained is used to generate an output power sequence for controlling the motor based on the predicted slope sequence, the recommended action sequence and the corresponding reward score, and the vehicle driving data at the current time, comprising: based on the predicted slope sequence and the vehicle driving data at the current time, a multi-layer perceptron structure in the control model is used to generate a recommended driving power; based on the recommended action sequence and the corresponding reward score, a gated recurrent unit in the control model is used to generate a predicted energy recovery intensity; based on the recommended driving power and the predicted energy recovery intensity, a nonlinear mapping layer in the control model is used to generate the output power sequence.

5. The method of claim 1, wherein, Before the control model pre-trained is used to generate an output power sequence for controlling the motor based on the predicted slope sequence, the recommended action sequence and the corresponding reward score, and the vehicle driving data at the current time, comprising: determine a corresponding preset parameter template according to a current driving mode of the vehicle; adjust model parameters of the control model according to the preset parameter template.

6. The method of claim 1, wherein, The method further comprises: in response to determining that the output power sequence does not satisfy a preset constraint condition, adjusting the output power sequence.

7. The method of claim 1, wherein, The method further comprises: fusing the output power sequence and an initial power sequence generated by a vehicle control unit according to a current vehicle working condition to obtain a fused power sequence, so that the motor operates according to the fused power sequence.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor implements the method of any one of claims 1 to 7 when executing the computer program.

9. A vehicle characterized by comprising: The vehicle comprises the electronic device of claim 8. The processor implements the method of any one of claims 1 to 7 when executing the computer program. The vehicle comprises the electronic device of claim 8.

Images