Vehicle control system and method based on VCU and VCU

Abstract

The invention provides a vehicle control system and method based on a VCU and the VCU, the system comprises a signal acquisition module and a micro-control chip with a neural network hardware acceleration function, the neural network hardware acceleration function of the micro-control chip provides exclusive computing power support for multi-dimensional time sequence data processing, the data operation efficiency is greatly improved, and the system can be applied to the field of multi-dimensional time sequence data processing. And in cooperation with the characteristics of the preset neural network model, the long-term dependency relationship between the vehicle operation state signals can be fully captured, the limitation of a conventional algorithm is effectively made up, and the prediction precision and real-time performance of the torque expected demand are improved. On the basis, an energy coefficient correction module dynamically corrects an energy recovery coefficient based on a torque expected demand, and an acceleration parameter correction module synchronously optimizes related parameters of a motor controller acceleration model, so that the VCU can adapt to a control strategy in advance according to a driving scene, and the problem that a traditional VCU is poor in intelligent control effect is effectively solved.

Description

Technical Field

[0001] This application relates to the field of vehicle intelligent control technology, and in particular to a vehicle control system, method and VCU based on VCU. Background Technology

[0002] In the current new energy vehicle field, the core control of vehicle performance and safety relies on the VCU (Vehicle Control Unit)'s ability to interpret driving intentions, distribute energy, and coordinate systems. As the core of vehicle control, the VCU needs to predict the vehicle's expected torque demand to optimize energy recovery strategies and motor acceleration parameters, thereby improving range and driving comfort. However, traditional VCUs use microcontrollers that lack independent hardware acceleration support. Their conventional algorithms struggle to efficiently process the multi-dimensional time-series data generated by the vehicle and cannot fully capture the long-term dependencies between data points. This results in insufficient accuracy and real-time performance in predicting expected torque demand, leading to poor intelligent vehicle control by the VCU. Summary of the Invention

[0003] In view of the above problems, in order to improve the intelligent control effect of VCU on vehicles, this application provides a fault scenario screening method, system, electronic device and medium.

[0004] The embodiments of this application disclose the following technical solutions:

[0005] In a first aspect, embodiments of this application provide a vehicle control system based on a VCU, applied to a VCU. The system includes: a signal acquisition module and a microcontroller chip with neural network hardware acceleration function; the microcontroller chip includes a torque prediction module, an energy coefficient correction module, and an acceleration parameter correction module; the vehicle includes a motor controller, and the acceleration controller controls the vehicle speed through a built-in acceleration model.

[0006] The signal acquisition module is used to acquire vehicle operating status signals;

[0007] The torque prediction module is used to predict torque demand based on the vehicle operating status signal and a preset neural network model, and determine the expected torque demand of the vehicle in a preset future time period.

[0008] The energy coefficient correction module is used to correct the energy recovery coefficient in the energy recovery strategy algorithm according to the expected torque demand, so as to obtain the corrected energy recovery coefficient.

[0009] The acceleration parameter correction module is used to correct the acceleration parameters in the acceleration model according to the expected torque requirement, so as to obtain the corrected acceleration parameters.

[0010] In one possible implementation, the energy coefficient correction module is specifically used for:

[0011] Obtain the initial energy recovery coefficient of the energy recovery strategy algorithm;

[0012] Based on the predicted braking intensity within the expected torque demand, determine the coefficient correction range for the initial energy recovery coefficient;

[0013] The initial energy recovery coefficient is adjusted according to the coefficient correction range to obtain the corrected energy recovery coefficient.

[0014] In one possible implementation, the acceleration parameter correction module is specifically used for:

[0015] Receive the current acceleration parameters fed back by the motor controller;

[0016] Based on the power output target within the expected torque demand, determine the adaptation deviation of the current acceleration parameters relative to the power output target;

[0017] Based on the adaptation deviation, a correction value is determined for the current acceleration parameter, and the current acceleration parameter is adjusted using the correction value to obtain the corrected acceleration parameter.

[0018] In one possible implementation, the preset neural network model is an LSTM model, and the deployment steps for the hardware acceleration function of the neural network using the LSTM model include:

[0019] The input sample is determined to be an input vector composed of the vehicle operating state signal, and the input layer, fully hidden layer and output layer of the model are configured; the activation function of the fully hidden layer is the ReLU function, the activation function of the output layer is the sigmoid function, and the output of the output layer is the expected torque requirement;

[0020] Configure the model's dropout regularization coefficient, the model's optimization algorithm as the Adam algorithm, and the model's loss function as mean squared absolute percentage error and the corresponding learning rate;

[0021] Based on the road condition type represented by different vehicle operating state signals, the dropout regularization coefficient, the learning rate, and the number of neurons in the fully hidden layer are calibrated individually to obtain multiple optimized LSTM models corresponding to different road condition types.

[0022] An executable file is generated based on an optimized LSTM model corresponding to each of the multiple different road conditions, and the executable file is downloaded to the microcontroller chip.

[0023] In one possible implementation, the LSTM model includes: an input gate, a forget gate, and an output gate;

[0024] The input gate is used to control the proportion of data stored in the input vector at the current moment, and to update the unit state based on the storage proportion;

[0025] The forget gate is used to control the retention ratio of the unit state at the previous time step;

[0026] The output gate is used to control the output ratio of the current unit state and perform nonlinear transformation on the output information to obtain the expected torque requirement.

[0027] In one possible implementation, the vehicle operating status signal includes a steering wheel angular velocity signal, an accelerator pedal angle signal, a brake pedal angle signal, a battery SOC value signal, a vehicle speed signal, and a road condition related signal; the road condition related signal includes a slope angle signal, a road surface friction coefficient signal, a road curvature signal, and a road traffic density signal.

[0028] In one possible implementation, the microcontroller chip is an S32K5 chip.

[0029] Secondly, this application provides a vehicle control method based on a VCU, which is applied to a VCU, wherein the VCU includes a microcontroller chip with neural network hardware acceleration function; the vehicle includes a motor controller, and the acceleration controller controls the vehicle speed through a built-in acceleration model.

[0030] Acquire vehicle operating status signals;

[0031] Based on the vehicle operating status signal and the preset neural network model, torque demand is predicted to determine the expected torque demand of the vehicle within a preset future time period.

[0032] Based on the expected torque demand, the energy recovery coefficient in the energy recovery strategy algorithm is corrected to obtain the corrected energy recovery coefficient;

[0033] Based on the expected torque requirement, the acceleration parameters in the acceleration model are corrected to obtain the corrected acceleration parameters.

[0034] In one possible implementation, the preset neural network model is an LSTM model, and the deployment steps for the hardware acceleration function of the neural network using the LSTM model include:

[0035] The input sample is determined to be an input vector composed of the vehicle operating state signal, and the input layer, fully hidden layer and output layer of the model are configured; the activation function of the fully hidden layer is the ReLU function, the activation function of the output layer is the sigmoid function, and the output of the output layer is the expected torque requirement;

[0036] Configure the model's dropout regularization coefficient, the model's optimization algorithm as the Adam algorithm, and the model's loss function as mean squared absolute percentage error and the corresponding learning rate;

[0037] Based on the road condition type represented by different vehicle operating state signals, the dropout regularization coefficient, the learning rate, and the number of neurons in the fully hidden layer are calibrated individually to obtain multiple optimized LSTM models corresponding to different road condition types.

[0038] An executable file is generated based on an optimized LSTM model corresponding to each of the multiple different road conditions, and the executable file is downloaded to the microcontroller chip.

[0039] Thirdly, embodiments of this application provide a VCU for implementing any possible VCU-based vehicle control method in the second aspect.

[0040] Compared to existing technologies, this application offers the following advantages: This application provides a vehicle control system, method, and VCU based on a VCU. The system includes a signal acquisition module and a microcontroller chip with neural network hardware acceleration capabilities. The microcontroller chip's neural network hardware acceleration function provides dedicated computing power support for multi-dimensional time-series data processing, significantly improving data processing efficiency. Combined with the characteristics of a preset neural network model, it can fully capture the long-term dependencies between vehicle operating state signals, effectively overcoming the limitations of conventional algorithms and improving the prediction accuracy and real-time performance of torque expectation requirements. Furthermore, the energy coefficient correction module dynamically corrects the energy recovery coefficient based on torque expectation requirements, and the acceleration parameter correction module synchronously optimizes the relevant parameters of the motor controller acceleration model. This enables the VCU to adapt the control strategy in advance according to the driving scenario, not only solving the problem of poor intelligent control performance in traditional VCUs but also achieving synergy between energy recovery and power output, further ensuring the stability and intelligence level of vehicle control. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A schematic diagram of a vehicle control system based on a VCU provided in this application embodiment;

[0043] Figure 2 A schematic diagram of another VCU-based vehicle control system provided for an embodiment of this application;

[0044] Figure 3 A flowchart illustrating an energy recovery coefficient correction method provided in this application embodiment;

[0045] Figure 4 A flowchart illustrating an acceleration parameter correction method provided in an embodiment of this application;

[0046] Figure 5 This application provides a schematic diagram of a process for deploying neural network hardware acceleration functionality on an LSTM model, as illustrated in an embodiment of the present application.

[0047] Figure 6 This is a flowchart illustrating a vehicle control method based on a VCU, provided as an embodiment of this application. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and accompanying drawings. It should be particularly noted that the embodiments described in this application are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0049] 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.

[0050] As described earlier, in the current new energy vehicle field, the core control of vehicle performance and safety relies on the VCU's (Vehicle Control Unit) capabilities for interpreting driving intentions, distributing energy, and coordinating systems. The VCU, as the core of vehicle control, needs to predict the vehicle's expected torque demand to optimize energy recovery strategies and motor acceleration parameters, thereby improving range and driving comfort. However, traditional VCUs use microcontrollers that lack independent hardware acceleration support. Their conventional algorithms struggle to efficiently process the multi-dimensional time-series data generated by the vehicle, failing to fully capture long-term dependencies between data points. This results in insufficient accuracy and real-time performance in predicting expected torque demand, leading to poor intelligent vehicle control by the VCU.

[0051] Based on this, this application provides a vehicle control system, method, and VCU based on a VCU. The system includes a signal acquisition module and a microcontroller chip with neural network hardware acceleration capabilities. The microcontroller chip's neural network hardware acceleration function provides dedicated computing power support for multi-dimensional time-series data processing, significantly improving data processing efficiency. Combined with the characteristics of a preset neural network model, it can fully capture the long-term dependencies between vehicle operating state signals, effectively overcoming the limitations of conventional algorithms and improving the prediction accuracy and real-time performance of torque expectation requirements. Furthermore, the energy coefficient correction module dynamically corrects the energy recovery coefficient based on torque expectation requirements, and the acceleration parameter correction module synchronously optimizes the relevant parameters of the motor controller acceleration model. This enables the VCU to adapt the control strategy in advance according to the driving scenario, not only solving the problem of poor intelligent control performance in traditional VCUs but also achieving synergy between energy recovery and power output, further ensuring the stability and intelligence level of vehicle control.

[0052] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0053] See Figure 1 and Figure 2 , Figure 1 This is a schematic diagram of a vehicle control system based on a VCU, provided as an embodiment of this application. Figure 2 This is a schematic diagram of another VCU-based vehicle control system provided in an embodiment of this application. Wherein, Figure 2 To adapt to the exemplary VCU in this embodiment, in Figure 2 In this system, the microcontroller chip within the VCU is an S32K5 (corresponding to the neural network hardware in the previous embodiment of this application). The peripheral components shown in the figure include various types of input analog signal monitoring modules, corresponding to the signal acquisition module in this embodiment, responsible for acquiring vehicle operating status signals. The S32K5, as the core, integrates torque prediction, energy coefficient correction, and acceleration parameter correction functions. It can predict the expected torque demand based on the acquired multi-source signals and correct the energy recovery coefficient and acceleration parameters. Components such as the CAN communication module support its interaction with the vehicle motor controller, enabling the VCU to control various controllers within the vehicle.

[0054] See Figure 1 The VCU in this embodiment includes a signal acquisition module 100, a microcontroller chip 200, a torque prediction module 300, an energy coefficient correction module 400, and an acceleration parameter correction module 500. The microcontroller chip 200 features neural network hardware acceleration capabilities. This chip integrates a dedicated neural processing unit, specifically designed for optimized neural network operations. Through a parallel low-latency processing architecture and a dedicated instruction set, it efficiently handles the computation of multi-dimensional vehicle operating status time-series data. Compared to traditional microcontrollers that rely on software simulation, it significantly improves the inference efficiency of models such as LSTM, ensuring the real-time performance and accuracy of torque expectation demand prediction while meeting automotive-grade low power consumption and small size constraints.

[0055] The signal acquisition module 100 is used to acquire vehicle operating status signals;

[0056] In this embodiment, the signal acquisition module is used to acquire multi-dimensional vehicle operating status. It captures multi-dimensional signals during vehicle operation through a multi-interface fusion acquisition and preprocessing mechanism. On the one hand, the module directly acquires signals related to the driver's operating intentions through a hard-wired interface, including analog and digital signals such as accelerator pedal opening, brake pedal travel, steering wheel angle and speed, which are converted into calculable digital signals after analog-to-digital conversion. On the other hand, it achieves high-speed communication with vehicle controllers such as ADAS (Advanced Driver Assistance Systems), BMS (Battery Management System), and MCU (Motor Control Unit) through the HS CAN bus, receiving environmental signals such as slope angle, road friction coefficient, road curvature, and traffic congestion, as well as vehicle status data such as battery SOC value, motor status, and vehicle speed. Simultaneously, it acquires preset lookup table values ​​as supplementary references. For the collected multi-source signals, the module will also perform noise reduction filtering and data calibration processing. The road condition signals transmitted by ADAS will be fitted and optimized using the least squares method and matched with the VCU preset road condition model to ensure the accuracy and consistency of the signals. Finally, it will be integrated to form a complete set of vehicle operating status signals covering driver intention, vehicle status and external driving environment, providing a data foundation for the subsequent torque prediction process.

[0057] The torque prediction module 300 is used to predict the torque demand of the vehicle within a preset future time period based on the vehicle operating status signal and a preset neural network model.

[0058] The torque prediction module is the core computing unit within the VCU that enables intelligent torque control. Its execution logic relies on a pre-built LSTM (Long Short-Term Memory) neural network model to mine temporal correlation features in vehicle operating status signals to predict torque demand. This module receives multi-dimensional data integrated by the signal acquisition module, including driver-related signals such as accelerator pedal opening, brake pedal travel, and steering wheel angular velocity; vehicle-specific signals such as battery SOC and vehicle speed; and environmental signals transmitted by ADAS, such as slope angle, road friction coefficient, and road curvature. Pre-built lookup table values ​​are also incorporated as supplementary references. These multi-source signals are synchronously calibrated and then used to construct a temporal input vector with a set sampling step size (10ms in the example). This adapts to the LSTM model's ability to process sequential data. Through the synergistic effect of forget gates, input gates, and output gates within the model, it solves the problem of long-term data dependencies that traditional algorithms struggle to capture. Simultaneously, it dynamically incorporates new input data, avoiding prediction biases caused by gradient vanishing or exploding.

[0059] The LSTM model in this embodiment includes an input gate, a forget gate, and an output gate. Through a collaborative gating mechanism of these three gates, the LSTM model accurately processes the time-series input vector composed of vehicle operating status signals, achieving reliable prediction of torque demand. The forget gate, as the primary information filtering step, focuses on controlling the retention ratio of the previous time-series unit state. Through flexible adjustment, it filters out historical time-series information valuable for current torque prediction, discarding redundant data and establishing a data foundation for subsequent calculations, effectively avoiding the information overload problem that occurs in traditional algorithms when processing long sequences. The input gate focuses on the current input vector, controlling the retention ratio to filter out key real-time vehicle operating status information. This retention ratio is then used to dynamically update the unit state, ensuring that the unit state both continues effective historical information and incorporates new operating condition data, achieving a combination of historical and real-time information. The output gate takes the updated current unit state, first controlling its output ratio to filter out core information directly related to torque demand, and then optimizing and adjusting the output information through nonlinear transformation, ultimately generating a torque demand that accurately matches the vehicle's driving scenario. The three-gating mechanism works in concert to ensure the full capture of long-term dependencies in multi-dimensional time-series vehicle data, and through refined information filtering and updating, to ensure the real-time performance and accuracy of torque prediction results. Specifically, the interrelationships among the three can be calculated using the following formula:

[0060] ;

[0061] In the formula, f t i t o t These correspond to the forget gate, input gate, and output gate, respectively; W f W i W c W o b f , bi, bc, bo correspond to the weights and biases of each gate, respectively; x t c is the input at time t; t and These are the state variables and potential state variables at time t, respectively; h t-1 Let be the external state variable at time t-1; σ is the sigmoid function.

[0062] The energy coefficient correction module 400 is used to correct the energy recovery coefficient in the energy recovery strategy algorithm according to the expected torque demand, so as to obtain the corrected energy recovery coefficient.

[0063] The energy coefficient correction module is a functional unit of the VCU that optimizes energy recovery. Its role is to dynamically adjust the energy recovery coefficient in the energy recovery strategy algorithm based on the expected torque demand output by the torque prediction module, thereby achieving a balance between energy recovery efficiency and driving experience. In an exemplary application scenario, this module receives torque expectation demand data generated by the torque prediction module every 10ms within the next second (i.e., torque is collected according to a set signal acquisition cycle). Simultaneously, it collects auxiliary parameters such as the battery SOC value, current vehicle speed, and brake pedal signal transmitted by the battery management system (BMS), constructing a multi-dimensional basis for correction decisions. Based on the driving scenario reflected by the expected torque demand (such as acceleration, constant speed, braking, etc.) and the predicted braking intensity, the module analyzes the adaptability of the current energy recovery coefficient to the actual demand. When a forced braking demand is predicted, the energy recovery coefficient is appropriately increased to maximize the recovery of kinetic energy during braking. Conversely, when a slight deceleration or a SOC value close to full charge is predicted, the coefficient is appropriately reduced to avoid affecting driving smoothness or damaging the battery. In special scenarios such as driving on slopes, the coefficient is adjusted in conjunction with the slope angle signal to balance power assistance and energy recovery. During the correction process, the module uses a preset energy recovery strategy as a basis, quantifies the matching degree between the expected torque demand and the current operating conditions, and calculates the coefficient adjustment range to ensure that the corrected energy recovery coefficient not only matches the real-time driving intention but also meets the requirements of battery safety and energy utilization efficiency. The final corrected energy recovery coefficient is updated in real time to the energy recovery strategy algorithm of the VCU, guiding the energy recovery intensity when the motor controller switches to the power generation mode. This effectively reduces the number of times mechanical braking is used, improves the vehicle's driving range, and avoids driving jerks caused by sudden changes in energy recovery intensity, achieving synergistic optimization of energy recovery and driving comfort.

[0064] Next, with reference to the accompanying drawings of a specific process implementation, the process of the energy coefficient correction module correcting the energy recovery coefficient will be described in detail. See also... Figure 3 The figure is a flowchart illustrating an energy recovery coefficient correction method provided in an embodiment of this application, specifically including the following steps:

[0065] S1041: Obtain the initial energy recovery coefficient of the energy recovery strategy algorithm.

[0066] First, the initial energy recovery coefficient in the energy recovery strategy algorithm needs to be obtained. This initial coefficient is not a fixed value, but rather a baseline parameter pre-configured by the VCU based on a preset basic energy recovery strategy, combined with the vehicle's normal driving conditions, battery safety threshold, and overall vehicle energy consumption target. Simultaneously with obtaining the initial coefficient, the module can also acquire multi-source vehicle operating status signals collected by the signal acquisition module. These data, together with the initial energy recovery coefficient, form the basis for coefficient correction, ensuring that subsequent correction actions align with the vehicle's real-time operating status and avoiding adaptation deviations caused by decisions based on a single parameter.

[0067] S1042: Based on the braking intensity prediction result within the expected torque demand, determine the coefficient correction range for the initial energy recovery coefficient.

[0068] After acquiring the initial energy recovery coefficient and related auxiliary data, the correction range for the energy recovery coefficient is determined based on the braking intensity prediction result in the expected torque demand. The expected torque demand includes the torque change trend every 10ms within the next second. By analyzing this time-series data, the driver's braking intention and corresponding braking intensity can be predicted in advance. When the torque demand shows a large negative fluctuation, it is determined to be a forced braking scenario, and a larger correction range needs to be set to maximize the recovery of braking kinetic energy. Conversely, when the torque demand decreases slightly, it is determined to be a slight deceleration scenario, and the correction range will be appropriately reduced to balance energy recovery and driving smoothness. When special conditions such as driving on slopes are predicted, the correction logic will be adjusted in conjunction with the slope angle signal. When going uphill, the correction range will be appropriately reduced to prioritize power output, and when going downhill, the correction range will be increased to improve energy recovery efficiency. At the same time, the correction range can also be dynamically adjusted in conjunction with the battery SOC value. If the SOC value is close to the full charge threshold, the correction range will be further reduced or even energy recovery will be suspended to avoid the risk of battery overcharging, ensuring that the correction range meets the braking demand while taking into account battery safety and operating condition adaptability.

[0069] S1043: Adjust the initial energy recovery coefficient according to the coefficient correction range to obtain the corrected energy recovery coefficient.

[0070] Finally, based on the determined coefficient correction range, the initial energy recovery coefficient is precisely quantified and adjusted to obtain the corrected energy recovery coefficient. The adjustment process employs linear or non-linear quantification calculations, converting the correction range into specific coefficient adjustment values ​​according to the correction logic under different operating conditions. This ensures that the adjusted coefficient is within a reasonable range, neither exceeding the upper limit of the motor's power generation capacity nor falling below the basic recovery efficiency requirements. After correction, the energy coefficient correction module updates the new energy recovery coefficient in real time to the VCU's energy recovery strategy algorithm. This allows the algorithm to generate control commands based on the latest coefficient, transmitting them to the motor controller via the CAN bus. The commands guide the motor to switch to power generation mode during braking or deceleration and recover energy according to the corresponding recovery intensity. This reduces the frequency of mechanical braking and avoids the driving jerking caused by sudden changes in energy recovery intensity, achieving a dual optimization of energy utilization and driving experience.

[0071] The acceleration parameter correction module 500 is used to correct the acceleration parameters in the acceleration model according to the expected torque requirement, so as to obtain the corrected acceleration parameters.

[0072] The acceleration parameter correction module is a key functional unit of the VCU for optimizing motor power output and improving driving smoothness. This module needs to dynamically adjust the acceleration parameters in the motor controller's built-in acceleration model based on the expected torque demand, thereby improving driving smoothness. In the execution process, the module receives torque demand data for the next 1 second (10ms) generated by the torque prediction module in real time. Simultaneously, it establishes high-speed communication with the motor controller via the HS CAN bus to synchronously acquire the operating parameters of the current acceleration model (such as motor response gain, torque rise slope, speed adjustment coefficient, etc.), and combines this with the vehicle operating status signal provided by the signal acquisition module to correct the acceleration parameters in the acceleration model.

[0073] Next, the process of parameter correction in the acceleration parameter correction module will be described in detail with reference to the accompanying drawings of a specific implementation embodiment. See also Figure 4 The figure is a flowchart illustrating an acceleration parameter correction method provided in an embodiment of this application, specifically including the following steps:

[0074] S1051: Receive the current acceleration parameters fed back by the motor controller.

[0075] First, the acceleration parameter correction module needs to receive the current acceleration parameters from the motor controller. This can be achieved through real-time data transmission via the high-speed communication capabilities of the HSCAN bus, ensuring the timeliness and accuracy of parameter acquisition. The current acceleration parameters encompass core indicators such as motor response gain, torque rise slope, and speed regulation coefficient, directly determining the motor's power output characteristics.

[0076] S1052: Determine the adaptation deviation of the current acceleration parameters relative to the power output target based on the power output target within the expected torque demand.

[0077] After acquiring the current acceleration parameters, the acceleration parameter correction module quantifies and analyzes the adaptation deviation of the current acceleration parameters based on the power output target within the expected torque demand. For example, the expected torque demand includes the torque change trend every 10ms within the next second. By analyzing this time-series data, the module can clearly define the power output target under different scenarios, such as the rapid torque increase demand in rapid acceleration scenarios, the stable torque maintenance demand during constant speed driving, and the load compensation torque demand during uphill driving. Subsequently, the module compares the current acceleration parameters with the corresponding power output target. If the current motor response gain is insufficient, causing the torque increase to lag behind expectations, or if the torque rise slope is unreasonable and may cause sudden speed changes, the module quantifies and calculates the difference between the two to determine the specific value and direction of the adaptation deviation, ensuring that the deviation analysis aligns with actual driving intentions and operating conditions.

[0078] S1053: Determine a correction value for the current acceleration parameter based on the adaptation deviation, and adjust the current acceleration parameter using the correction value to obtain the corrected acceleration parameter.

[0079] Finally, based on the determined adaptation deviation, the module calculates the correction value for the current acceleration parameters using a preset algorithm model. The correction value follows the motor's operating safety threshold and driving comfort standards, ensuring power output efficiency while avoiding instability caused by excessive parameter adjustments. After obtaining the correction value, the module adjusts the current acceleration parameters. For example, it increases the motor response gain for deviations in rapid acceleration scenarios and optimizes the speed adjustment coefficient for deviations in smooth driving scenarios, ultimately obtaining the corrected acceleration parameters. The corrected acceleration parameters are fed back to the motor controller in real time via the HS CAN bus, updating the core parameters of the acceleration model. This allows the motor to adapt to torque change trends in advance, achieving smooth and precise power output, effectively avoiding the jerking sensation caused by sudden speed changes, while simultaneously optimizing power performance and energy consumption, further improving the vehicle's driving experience and handling stability.

[0080] As discussed earlier regarding the LSTM model, the neural network hardware acceleration function of the microcontroller chip in this embodiment primarily relies on the LSTM model. The following section, with reference to the accompanying drawings of a specific flowchart embodiment, will describe the deployment process of the LSTM model for neural network hardware acceleration.

[0081] See Figure 5 The figure is a schematic diagram of a process for deploying neural network hardware acceleration functionality using an LSTM model according to an embodiment of this application, specifically including the following steps:

[0082] S1061: Determine that the input sample is an input vector composed of the vehicle operating state signal, and configure the input layer, fully hidden layer and output layer of the model; the activation function of the fully hidden layer is the ReLU function, the activation function of the output layer is the sigmoid function, and the output of the output layer is the expected torque requirement.

[0083] Deploying hardware acceleration functionality for neural networks using LSTM first requires defining the input samples and building the core layers of the model. The input samples are input vectors composed of vehicle operating state signals, encompassing steering wheel angular velocity, accelerator pedal angle, brake pedal angle, battery SOC, vehicle speed, and road condition-related signals. These road condition-related signals include slope angle, road surface friction coefficient, road curvature, and traffic density. All data are collected synchronously with a sampling step of 10ms to fully preserve temporal characteristics and adapt to the processing characteristics of the LSTM model.

[0084] The model layer construction needs to be adapted to the hardware acceleration architecture of the microcontroller chip. First, the input layer is built, and the dimension and time step of the input vector are matched to ensure efficient data transmission to the hardware computing unit. Then, fully hidden layers are configured, and the ReLU activation function is used to alleviate the gradient vanishing problem, improving the model's computational efficiency on the hardware. Finally, the output layer is built, and the output result is normalized using the sigmoid activation function to ensure that the final output torque expectation is within a reasonable range, accurately aligning with subsequent energy recovery coefficient correction and acceleration parameter optimization scenarios.

[0085] S1062: Configure the dropout regularization coefficient of the model, configure the optimization algorithm of the model as the Adam algorithm, and configure the loss function of the model as the mean squared absolute percentage error and the corresponding learning rate.

[0086] After completing the model layer construction, further configuration of training parameters is required to ensure efficient operation after hardware deployment. First, dropout regularization coefficients are set to reasonably shield some neuron connections, preventing overfitting during model training and ensuring good generalization ability under complex road conditions. Then, the Adam algorithm is selected as the optimizer. Its adaptive learning rate allows the model to quickly converge to its optimal state, adapting to the real-time computing requirements of in-vehicle scenarios. Simultaneously, mean squared absolute percentage error is used as the loss function to more accurately reflect the relative error of torque prediction, meeting the stringent control precision requirements of the VCU. Furthermore, an appropriate initial learning rate is set based on hardware computing power and model complexity to balance training efficiency and prediction accuracy, ensuring that all parameters maximize the advantages of hardware-accelerated parallel computing.

[0087] S1063: According to the road condition types characterized by different vehicle operation state signals, calibrate the corresponding dropout regularization coefficient, learning rate, and the number of neurons in the fully hidden layer separately, and obtain optimized LSTM models separately corresponding to multiple different road condition types.

[0088] To make the model adapt to different driving scenarios, it is necessary to complete the targeted calibration of the core parameters in combination with the road condition types characterized by the vehicle operation state signals. Among the vehicle operation state signals, data such as the slope angle, road surface friction coefficient, road curvature, and congestion situation from the ADAS controller can directly divide typical road conditions such as congested sections, highway sections, and slope sections. For each road condition, collect a sufficient amount of samples through on-vehicle tests, compare the model performances under different parameter combinations, and calibrate the applicable dropout regularization coefficient, learning rate, and the number of neurons in the fully hidden layer one by one. For example, in congested sections, the model needs to be strengthened to capture frequent acceleration and deceleration signals, and in slope sections, the number of neurons needs to be optimized to adapt to the load change. Finally, generate exclusive optimized LSTM models for each road condition to ensure that the model can maintain high prediction accuracy in various scenarios.

[0089] S1064: Generate an executable file based on the optimized LSTM models separately corresponding to multiple different road condition types, and download the executable file to the microcontroller chip.

[0090] Finally, complete the generation of the executable file and the chip deployment based on the optimized LSTM models corresponding to multiple road conditions. With the development tools supporting the microcontroller chip, convert the structures and parameters of each optimized model into machine language that can be directly parsed by the hardware, generate an executable file adapted to the NPU architecture, and avoid the computing overhead caused by additional format conversion. Subsequently, download the file to the microcontroller chip through the development tools to complete the final deployment of the model. After deployment, supported by hardware acceleration, the model can quickly identify the current road condition based on the real-time collected vehicle operation state signals, automatically call the corresponding optimized model to predict the torque demand, and provide accurate and real-time data input for the subsequent functional modules, giving full play to the intelligent control efficiency of the VCU.

[0091] This application provides a vehicle control system based on a VCU. The system includes a signal acquisition module and a microcontroller chip with neural network hardware acceleration capabilities. The microcontroller chip's neural network hardware acceleration function provides dedicated computing power support for multi-dimensional time-series data processing, significantly improving data processing efficiency. Combined with the characteristics of a preset neural network model, it can fully capture the long-term dependencies between vehicle operating state signals, effectively overcoming the limitations of conventional algorithms and improving the prediction accuracy and real-time performance of torque expectation requirements. Furthermore, an energy coefficient correction module dynamically corrects the energy recovery coefficient based on torque expectation requirements, and an acceleration parameter correction module synchronously optimizes the relevant parameters of the motor controller acceleration model. This allows the VCU to adapt the control strategy in advance according to the driving scenario, not only solving the problem of poor intelligent control performance in traditional VCUs but also achieving synergy between energy recovery and power output, further ensuring the stability and intelligence level of vehicle control.

[0092] The following describes a vehicle control method based on a VCU provided in an embodiment of this application. The vehicle control method based on a VCU described below can be referred to in correspondence with the vehicle control system based on a VCU described above.

[0093] See Figure 6 The figure is a flowchart illustrating a vehicle control method based on a VCU provided in an embodiment of this application, specifically including the following steps:

[0094] S101: Acquire vehicle operating status signal;

[0095] S102: Based on the vehicle operating status signal and the preset neural network model, torque demand is predicted to determine the expected torque demand of the vehicle within a preset future time period.

[0096] S103: Based on the expected torque demand, the energy recovery coefficient in the energy recovery strategy algorithm is corrected to obtain the corrected energy recovery coefficient;

[0097] S104: Based on the expected torque requirement, the acceleration parameters in the acceleration model are corrected to obtain the corrected acceleration parameters.

[0098] In one possible implementation, the preset neural network model is an LSTM model, and the deployment steps for the hardware acceleration function of the neural network using the LSTM model include:

[0099] The input sample is determined to be an input vector composed of the vehicle operating state signal, and the input layer, fully hidden layer and output layer of the model are configured; the activation function of the fully hidden layer is the ReLU function, the activation function of the output layer is the sigmoid function, and the output of the output layer is the expected torque requirement;

[0100] Configure the model's dropout regularization coefficient, the model's optimization algorithm as the Adam algorithm, and the model's loss function as mean squared absolute percentage error and the corresponding learning rate;

[0101] Based on the road condition type represented by different vehicle operating state signals, the dropout regularization coefficient, the learning rate, and the number of neurons in the fully hidden layer are calibrated individually to obtain multiple optimized LSTM models corresponding to different road condition types.

[0102] An executable file is generated based on an optimized LSTM model corresponding to each of the multiple different road conditions, and the executable file is downloaded to the microcontroller chip.

[0103] This application also provides a VCU for implementing the VCU-based vehicle control method described in any of the above embodiments.

[0104] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the system, method, and VCU embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The system, method, and VCU embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0105] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A VCU-based vehicle control system, characterized by, The system applied to VCU comprises a signal acquisition module and a micro control chip with a neural network hardware acceleration function; the micro control chip comprises a torque prediction module, an energy coefficient correction module and an acceleration parameter correction module; the vehicle comprises a motor controller, and the acceleration controller controls the vehicle speed through the built-in acceleration model; The signal acquisition module is used to acquire vehicle operating state signals; The torque prediction module is used to predict torque demand according to the vehicle operating state signals and a preset neural network model, and determine the torque expected demand of the vehicle within a preset future time length; The energy coefficient correction module is used to correct the energy recovery coefficient in the energy recovery strategy algorithm according to the torque expected demand, to obtain a corrected energy recovery coefficient; The acceleration parameter correction module is used to correct the acceleration parameters in the acceleration model according to the torque expected demand, to obtain corrected acceleration parameters.

2. The system of claim 1, wherein, The energy coefficient correction module is specifically used for: acquiring an initial energy recovery coefficient of the energy recovery strategy algorithm; determining a coefficient correction amplitude for the initial energy recovery coefficient according to a braking intensity prediction result in the torque expected demand; adjusting the initial energy recovery coefficient according to the coefficient correction amplitude to obtain the corrected energy recovery coefficient.

3. The system of claim 1, wherein, The acceleration parameter correction module is specifically used for: receiving a current acceleration parameter fed back by the motor controller; determining an adaptation deviation of the current acceleration parameter relative to a power output target in the torque expected demand; determining a correction value for the current acceleration parameter according to the adaptation deviation, and adjusting the current acceleration parameter through the correction value to obtain the corrected acceleration parameter.

4. The system of claim 1, wherein, The preset neural network model is an LSTM model, and the deployment steps of the LSTM model for the neural network hardware acceleration function comprise: determining an input sample as an input vector constituted by the vehicle operating state signals, and configuring an input layer, a full hidden layer and an output layer of the model; the activation function of the full hidden layer is a Relu function, the activation function of the output layer is a sigmoid function, and the output of the output layer is the torque expected demand; configuring a dropout regularization coefficient of the model, configuring an optimization algorithm of the model as an adam algorithm, and configuring a loss function of the model as a mean square absolute percentage error and a corresponding learning rate; labeling separately corresponding dropout regularization coefficients, learning rates and neuron numbers of the full hidden layer according to different road types represented by the vehicle operating state signals, to obtain a plurality of separately corresponding optimization LSTM models under different road types; generating an executable file based on the plurality of separately corresponding optimization LSTM models under different road types, and downloading the executable file to the micro control chip.

5. The system of claim 4, wherein, The LSTM model comprises an input gate, a forget gate and an output gate. The input gate is configured to control a saving ratio of data in the input vector at the current time and update the cell state based on the saving ratio; The forget gate is configured to control a reservation ratio of the cell state at the previous time; The output gate is configured to control an output ratio of the current cell state and perform nonlinear conversion on the output information to obtain the torque expected demand.

6. The system of claim 1, wherein, The vehicle operating state signals include a steering wheel angular velocity signal, an accelerator pedal angle signal, a brake pedal angle signal, a battery SOC value signal, a vehicle speed signal, and a road condition related signal; the road condition related signal includes a slope angle signal, a road surface friction coefficient signal, a road curvature signal, and a road condition driving density signal.

7. The system of claim 1, wherein, The micro control chip is an S32K5 chip.

8. A VCU-based vehicle control method, characterized by, The VCU is applied to a vehicle, and the vehicle includes a motor controller and an acceleration controller configured to control vehicle speed by using an acceleration model. Vehicle operating state signals are acquired. Torque demand prediction is performed according to the vehicle operating state signals and a preset neural network model to determine a torque expected demand of the vehicle within a preset future time period. According to the torque expected demand, an energy recovery coefficient in the energy recovery strategy algorithm is modified to obtain a modified energy recovery coefficient. According to the torque expected demand, an acceleration parameter in the acceleration model is modified to obtain a modified acceleration parameter.

9. The method of claim 8, wherein, The preset neural network model is an LSTM model, and the LSTM model includes the following steps for deployment of the neural network hardware acceleration function: An input sample is determined to be an input vector constituted by the vehicle operating state signals, and an input layer, a full hidden layer, and an output layer of the model are configured; an activation function of the full hidden layer is a Relu function, an activation function of the output layer is a sigmoid function, and an output of the output layer is the torque expected demand; A dropout regularization coefficient of the model is configured, an optimization algorithm of the model is configured to be an adam algorithm, and a loss function of the model is configured to be a mean square absolute percentage error and a corresponding learning rate; According to road condition types represented by different vehicle operating state signals, individually corresponding dropout regularization coefficients, learning rates, and neuron numbers of the full hidden layer are calibrated to obtain individually corresponding optimization LSTM models under multiple different road condition types; Executable files are generated based on the individually corresponding optimization LSTM models under the multiple different road condition types, and the executable files are downloaded to the micro control chip.

10. A VCU, characterized in that, The VCU is configured to implement the vehicle control method based on the VCU according to any one of claims 8-9.

Images