Battery management method and vehicle

Abstract

The invention provides a battery management method and a vehicle, and relates to the technical field of electric energy management, and the method comprises the steps: obtaining an original signal of the driving data of the vehicle; extracting a feature representation of the driving data from the original signal based on a first sampling rate; the first sampling rate is smaller than the sampling rate of the original signal; inputting the feature representation into a binary neural network BNN to obtain probability values of multiple driving modes; a current driving mode of the vehicle is determined based on the probability value, and a rated output current of a battery of the vehicle is adjusted. According to the method, the features are extracted from the original signals in a dimensionality reduction mode, the BNN is adopted for reasoning to obtain the probability value of the driving mode, the driving mode can be rapidly recognized, the rated output current of the battery is adjusted, the real-time performance of battery management is guaranteed, and the service life of the battery is prolonged.

Description

Technical Field

[0001] The present application relates to the technical field of power management, and in particular, to a battery management method and a vehicle. Background Art

[0002] With the increasing penetration rate of the global new energy vehicle market, as a core component, the performance and lifespan of power batteries directly determine the competitiveness of vehicles and the user experience. As the "central nervous system" of power batteries, the Battery Management System (BMS) needs to achieve precise monitoring, dynamic regulation, and safety assurance under complex working conditions. However, the diversity of user driving behaviors (such as rapid acceleration, high-speed cruising, cold start at low temperatures, etc.) leads to non-linear fluctuations in battery load, posing challenges to the battery management system. Therefore, introducing data such as driving behavior and performing battery management through intelligent analysis is the current main development direction.

[0003] However, limited by in-vehicle computing power, the time required for intelligent analysis is relatively long, and the real-time performance of battery management is restricted. Summary of the Invention

[0004] The purpose of the present application is to provide a battery management method and a vehicle, which can quickly identify the driving mode and adjust the rated output current of the battery, ensure the real-time performance of battery management, and thus improve the battery lifespan.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] In a first aspect, a battery management method is provided. The method includes: obtaining the original signal of the driving data of the vehicle; extracting the feature representation of the driving data from the original signal based on a first sampling rate, where the first sampling rate is less than the sampling rate of the original signal; inputting the feature representation into a Binarized Neural Network (BNN) to obtain the probability values of multiple driving modes; determining the current driving mode of the vehicle based on the probability values, and adjusting the rated output current of the battery of the vehicle.

[0007] An embodiment of the present application provides a battery management method, which acquires the original signal of the driving data of a vehicle; extracts the feature representation of the driving data from the original signal based on a first sampling rate, where the first sampling rate is less than the sampling rate of the original signal; inputs the feature representation into a BNN to obtain the probability values of multiple driving modes; determines the current driving mode of the vehicle based on the probability values, and adjusts the rated output current of the battery of the vehicle. It can be seen that in the process of managing the battery in the embodiment of the present application, by reducing the dimension and extracting features from the original signal, and using a BNN for inference to obtain the probability values of the driving modes. On the one hand, reducing the dimension and extracting the feature representation can reduce the amount of data input to the model. On the other hand, the BNN is a binary lightweight model, with a small model size and a high inference rate. Based on the above two aspects, the driving mode can be quickly identified and the rated output current of the battery can be adjusted to ensure the real-time performance of battery management, thereby improving the battery life. In addition, reducing the amount of data to be processed and using a lightweight model can reduce the requirements for in-vehicle computing power and save costs.

[0008] Optionally, determining the current driving mode of the vehicle based on the probability values and adjusting the rated output current of the battery of the vehicle includes: when the probability value indicates that the vehicle is in an aggressive driving mode, reducing the rated output current of the battery according to the state of health (SOH) of the battery; when in an aggressive driving mode, the average torque of the vehicle is greater than a first threshold, and the standard deviation of the vehicle speed is greater than a second threshold.

[0009] Optionally, reducing the rated output current of the battery according to the state of health of the battery includes: reducing the rated output current of the battery by using the following expression:

[0010] A = B×(0.9 - 0.002×(100 - SOH))

[0011] A = 0.8A

[0012] Wherein, A represents the rated output current of the battery; B represents the maximum rated current of the battery.

[0013] Optionally, the vehicle includes an edge controller; extracting the feature representation of the driving data from the original signal based on the first sampling rate includes: the edge controller extracts the feature representation of the driving data from the original signal based on the first sampling rate.

[0014] Optionally, the vehicle further includes a domain controller; inputting the feature representation into the BNN to obtain the probability values of multiple driving modes includes: the domain controller obtains the feature representation from the edge controller; the domain controller inputs the feature representation into the BNN to obtain the probability values of multiple driving modes.

[0015] Optionally, extracting a feature representation of driving data from the original signal based on a first sampling rate, including: extracting the feature representation of driving data from the original signal by using sliding window statistics; wherein, the number of sampling points within the sliding window is determined according to the first sampling rate.

[0016] Optionally, the BNN is a neural network reconstructed based on the ARM NEON instruction set; the matrix operation of the BNN is converted from a convolution operation to an exclusive OR operation.

[0017] Optionally, obtaining the original signal of the driving data of the vehicle, including: obtaining the original signal of the driving data of the vehicle during the inference process of the BNN in the previous cycle.

[0018] Optionally, before obtaining the original signal of the driving data of the vehicle, the method further includes: obtaining the SOH of the battery; in the case where the SOH is less than a preset state threshold, setting the priority of driving behavior detection to be higher than the priority of battery temperature monitoring; the driving behavior detection is used to indicate the execution of the battery management method; otherwise, setting the priority of battery temperature monitoring to be higher than the priority of driving behavior detection.

[0019] In a second aspect, a battery management device is provided, including an acquisition module, a processing module, and an adjustment module; the acquisition module is configured to obtain the original signal of the driving data of the vehicle; the acquisition module is further configured to extract the feature representation of the driving data from the original signal based on a first sampling rate; the first sampling rate is less than the sampling rate of the original signal; the processing module is configured to input the feature representation into a binary neural network BNN to obtain probability values of multiple driving modes; the adjustment module is configured to determine the current driving mode of the vehicle based on the probability values and adjust the rated output current of the battery of the vehicle.

[0020] Optionally, the adjustment module is specifically configured to, in the case where the probability value indicates that the vehicle is in an aggressive driving mode, reduce the rated output current of the battery according to the battery health state of the battery; in the case of being in an aggressive driving mode, the torque mean value of the vehicle is greater than a first threshold, and the vehicle speed standard deviation is greater than a second threshold.

[0021] Optionally, the adjustment module is specifically configured to reduce the rated output current of the battery by using the following expression:

[0022] A = B × (0.9 - 0.002 × (100 - SOH))

[0023] A = 0.8A

[0024] Wherein, A represents the rated output current of the battery; B represents the maximum rated current of the battery.

[0025] Optionally, the vehicle includes an edge controller; the acquisition module is specifically configured to the edge controller extracts the feature representation of the driving data from the original signal based on the first sampling rate.

[0026] Optionally, the vehicle further includes a domain controller; the obtaining module is specifically configured to obtain, by the domain controller, a feature representation from the edge controller; and the domain controller inputs the feature representation into the BNN to obtain probability values of multiple driving modes.

[0027] Optionally, the obtaining module is specifically configured to extract, by using a sliding window statistics, a feature representation of the driving data from the original signal; wherein the number of sampling points within the sliding window is determined according to the first sampling rate.

[0028] Optionally, the BNN is a neural network reconstructed based on the ARM NEON instruction set; and the matrix operation of the BNN is converted from a convolution operation to an exclusive OR operation.

[0029] Optionally, the obtaining module is specifically configured to obtain, during the inference process of the BNN in the previous cycle, the original signal of the driving data of the vehicle.

[0030] Optionally, the obtaining module is further configured to obtain the battery health state of the battery; the processing module is further configured to, when the battery health state is less than a preset state threshold, set the priority of the driving behavior detection to be higher than the priority of the battery temperature monitoring; and the driving behavior detection is used to indicate the execution of the battery management method.

[0031] Otherwise, set the priority of the battery temperature monitoring to be higher than the priority of the driving behavior detection.

[0032] In a third aspect, a vehicle is provided, including a controller configured to execute the battery management method of the first aspect.

[0033] For the specific description of the second aspect and its implementation manners in this application, reference may be made to the detailed description of the first aspect and its various implementation manners, which will not be elaborated herein.

[0034] In a fourth aspect, this application provides a computer-readable storage medium, including: computer software instructions; when the computer software instructions run on an electronic device, the electronic device is caused to implement the method of the first aspect.

[0035] In a fifth aspect, this application provides a computer program product, which, when running on a computer, causes the computer to execute the steps of the related method described in the first aspect to implement the method of the first aspect.

[0036] These aspects or other aspects of this application will be more clearly understood in the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0038] Figure 1 A schematic diagram of the composition of a vehicle provided by an embodiment of the present application;

[0039] Figure 2 A schematic flowchart of a battery management method provided by an embodiment of the present application;

[0040] Figure 3 A schematic structural diagram of a BNN model provided by an embodiment of the present application;

[0041] Figure 4 A schematic flowchart of another battery management method provided by an embodiment of the present application;

[0042] Figure 5 A complete flowchart of a battery management method provided by an embodiment of the present application;

[0043] Figure 6 A schematic diagram of the composition of a battery management device provided by an embodiment of the present application. Detailed implementation manners

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0045] It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" aims to present relevant concepts in a specific manner.

[0046] To facilitate a clear description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, words such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and roles. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order.

[0047] At present, for a battery management system, it is very important to improve the performance and lifespan of the battery. Traditional BMS relies on fixed threshold control (such as current upper limit, temperature threshold) for battery management, but this approach has relatively low accuracy. However, with the increasing demand for intelligent driving, it is necessary to introduce driving behavior data (such as motor torque, vehicle speed, slope) to dynamically adjust the battery management strategy. By optimizing the battery health management strategy, not only can the battery lifespan be significantly extended and its high-efficiency performance be maintained, but also the user experience and device reliability can be improved simultaneously.

[0048] Under this background art, an embodiment of the present application provides a battery management method, which acquires the original signal of the driving data of the vehicle; extracts the feature representation of the driving data from the original signal based on a first sampling rate; the first sampling rate is less than the sampling rate of the original signal; inputs the feature representation into a binary neural network BNN to obtain the probability values of multiple driving modes; determines the current driving mode of the vehicle based on the probability values, and adjusts the rated output current of the battery of the vehicle. It can be seen that in the process of managing the battery in the embodiment of the present application, by reducing the dimension and extracting features from the original signal, and using BNN for inference to obtain the probability values of the driving mode. On the one hand, reducing the dimension and extracting the feature representation can reduce the amount of data input to the model. On the other hand, BNN is a binary lightweight model, with a small model size and a high inference rate. Based on the above two aspects, the driving mode can be quickly identified and the rated output current of the battery can be adjusted to ensure the real-time nature of battery management, thereby improving the battery lifespan. In addition, reducing the amount of data to be processed and using a lightweight model can reduce the requirements for in-vehicle computing power and save costs.

[0049] The battery management method provided by the present application can be applied to a vehicle such as Figure 1 shown.

[0050] Figure 1 The following is a schematic diagram of the composition of a vehicle provided by an embodiment of the present application. As Figure 1 shown, this vehicle includes a CAN bus 101, an edge controller 102, a domain controller 103, and a BMS 104.

[0051] Among them, the CAN bus 101, that is, the controller area network bus, is a serial communication protocol bus for real-time applications, which can use twisted pairs to transmit signals and is one of the most widely used field buses in the world. Its characteristics include complete serial data communication, providing real-time support, a transmission rate of up to 1 Mb / s, and having 11-bit addressing and error detection capabilities.

[0052] Among them, the edge controller can specifically be an MCU, a microcontroller, also known as a single-chip microcomputer. It appropriately reduces the frequency and specifications of the central processing unit and integrates peripheral interfaces such as memory, counters, and even diode drive circuits on a single chip to form a computer at the chip level, making different combined controls for different application scenarios.

[0053] Among them, the domain controller 103 is a centralized electronic control unit that integrates multiple subsystems with similar functions for unified management and control. By dividing the vehicle's electronic and electrical architecture into different domains according to functions, such as the power domain, chassis domain, body domain, infotainment domain, etc., each domain is managed by a domain controller, thereby improving the system's integration and scalability.

[0054] Among them, BMS104, i.e., the battery management system, is mainly used to monitor, manage, and protect the battery pack. It can real-time monitor parameters such as the battery's voltage, current, and temperature, calculate the remaining battery power, health status, etc., and at the same time protect the battery against overcharging, over-discharging, overcurrent, short circuit, etc., to ensure the safe and reliable operation of the battery.

[0055] In some embodiments, the CAN bus 101 belongs to the data transmission layer and can obtain the original signals during vehicle driving in real time, providing a data basis for the follow-up. Exemplarily, original signals such as motor torque (drive_motor_torque), vehicle speed (speed), slope (slope), etc. (sampling rate 100Hz) can be obtained in real time through the CAN bus.

[0056] In some embodiments, the MCU 102 belongs to the edge processing layer, edge controller, or edge terminal. The MCU 102 can extract the feature representation of the driving data from the original signals based on the first sampling rate. Exemplarily, the MCU 102 can complete the conversion from the original signals obtained in real time during vehicle driving through the CAN bus to low-dimensional statistical features within a certain time. For example, the lightweight feature extractor of the MCU102 can complete the conversion from the original data to low-dimensional statistical features within 5ms.

[0057] In some embodiments, the domain controller 103 belongs to the intelligent decision-making layer and can input the low-dimensional statistical features obtained from the MCU 102 into the BNN to obtain the probability values of multiple driving modes. Among them, the driving modes include: aggressive driving mode, gentle driving mode, high-speed cruise driving mode, low-speed congestion mode.

[0058] In some embodiments, the BMS104 belongs to the execution layer. The BMS104 can determine the vehicle's current driving mode based on the probability values obtained by the domain controller 103 and adjust the rated output current of the vehicle's battery to extend the battery life and maintain its high-efficiency performance.

[0059] In the embodiment of the present application, the CAN bus 101 can collect the original signal and send it to the MCU 102. The MCU 102 performs dimensionality reduction on the basis of the original data to extract the feature representation of the driving data, and sends it to the domain controller 103 and inputs it into the BNN model to obtain the probability values of driving modes such as aggressive driving mode, gentle driving mode, high-speed cruise driving mode, and low-speed congestion mode, and determines the current driving mode of the vehicle according to the probability values. The BMS 104 can adjust the rated output current of the vehicle's battery according to the current driving mode of the vehicle and the obtained state of health (SOH) of the current vehicle, so as to extend the battery life and maintain its high efficiency performance.

[0060] Figure 2 It is a schematic flow chart of a battery management method provided by an embodiment of the present application. As Figure 2 shown, the battery management method provided by the present application can be implemented by the above-mentioned vehicle, specifically by the controller in the vehicle, and specifically may include S201-S204.

[0061] S201. Obtain the original signal of the driving data of the vehicle.

[0062] Among them, the original signal is an analog or digital signal, which is obtained in real time through the CAN bus 101.

[0063] In a possible implementation manner, the original signal specifically includes at least: motor torque signal (drive_motor_torque), vehicle speed signal (speed), slope signal (slope). The motor torque is a signal transmitted from the motor controller through the CAN bus 101; the vehicle speed is a signal transmitted from the wheel speed sensor or the in-vehicle navigation system through the CAN bus 101; the slope is a signal output from the inertial measurement unit (IMU) or the in-vehicle slope sensor through the CAN bus 101.

[0064] In another possible implementation manner, if more sensors are connected to the system, for example, an accelerometer and a battery voltage sensor, the CAN bus can also obtain signals such as motor speed, battery current, and cell temperature.

[0065] In some embodiments, during vehicle driving, the CAN bus 101 in the vehicle can obtain the original signals of vehicle driving data such as motor torque, vehicle speed, and slope in real time, providing a data basis for subsequent operations.

[0066] S202. Extract the feature representation of the driving data from the original signal based on the first sampling rate.

[0067] Among them, the first sampling rate is less than the sampling rate of the original signal. For example, the sampling rate of the original signal is 200HZ, and the first sampling rate is 100HZ.

[0068] Among them, the characteristics of the driving data include but are not limited to torque mean, torque variance, torque peak-to-peak value, torque zero-crossing rate, vehicle speed mean, vehicle speed maximum value, slope mean, slope change rate, and slope variance.

[0069] In some embodiments, the above S202 can be specifically implemented as: the edge controller extracts the feature representation of the driving data from the original signal based on the first sampling rate. It can be understood that the edge controller can complete the real-time acquisition of the original signal during vehicle driving through the CAN bus within a certain time and complete the conversion of the original signal to low-dimensional statistical features.

[0070] In a possible implementation manner, a sliding window statistical method is adopted to extract the feature representation of the driving data from the original signal. Among them, the number of sampling points in the sliding window is determined according to the first sampling rate. Exemplarily, the number of points N in the sampling window is jointly determined by the first sampling rate fs and the window time length Tw, and the formula is N = fs * Tw.

[0071] For example, if the first sampling rate is 100HZ and the window length is 0.5s, then the number of sampling points is 50, which is specifically reflected in the following pseudo-code implementation:

[0072] Def edge_feature_extraction(torque_series,speed_series,current_slope,prev_slope):

[0073] # Torque series (torque_series), speed series (speed_series), current slope (current_slope) and previous slope (prev_slope)

[0074] # Sliding window statistic (window length 0.5s, i.e., 50 sampling points)

[0075] features={

[0076] 'torque_mean':np.mean(torque_series[-50:]),

[0077] 'speed_std':np.std(speed_series[-50:]),

[0078] 'slope_diff':current_slope-prev_slope,

[0079] 'current_peak': np.max(torque_series[-50:]) - np.min(torque_series[-50:]),

[0080] 'speed_diff_mean': np.mean(np.diff(speed_series[-50:])),

[0081] # Other features...

[0082] }

[0083] return features

[0084] Among them, edge_feature_extraction is edge feature extraction, torque_mean represents the mean torque, speed_std represents the standard deviation of speed, slope_diff represents the slope difference, current_peak represents the current peak, and speed_diff_mean represents the mean speed difference.

[0085] It can be understood that the specific implementation of the edge controller using the sliding window statistical method is as follows: under the conditions of a 100HZ sampling rate, a window length of 0.5s is set, and 50 sampling points are used. The mean torque is obtained from the torque sequence, the standard deviation data of speed is obtained from the speed sequence, the slope difference is obtained from the difference between the current slope and the previous slope, the current peak is obtained from the corresponding difference between the maximum torque sequence and the minimum difference sequence, and the mean speed difference is obtained by taking the difference and averaging in the speed sequence, etc.

[0086] It should be noted that the related technology directly processes the original data, which will result in data redundancy. In the embodiments of the present application, by performing dimensionality reduction feature extraction on the original data, the data dimension is reduced, and the subsequent model calculation burden is reduced.

[0087] S203. Input the feature representation into the binary neural network BNN to obtain the probability values of multiple driving modes.

[0088] In the embodiments of the present application, by inputting the driving data features obtained by the edge controller into the BNN model in the domain controller 103, the probability values of multiple driving modes can be obtained.

[0089] Among them, BNN is a neural network that binarizes weights and activation values (usually +1 / -1 or 0 / 1). It uses logical operations (such as XNOR) to replace floating-point operations, is suitable for deployment on edge devices, and has low computing power requirements for devices.

[0090] Among them, the driving modes at least include: aggressive driving mode, gentle driving mode, high-speed cruise driving mode, and low-speed congestion mode.

[0091] Exemplarily, the classification and definition of the above four driving behavior modes are described below. For the aggressive driving mode, its characteristic thresholds are: average torque > 70 N·m, standard deviation of vehicle speed > 12 km / h, slope change rate ∈ [-2°, 0°] (downhill acceleration). Typical scenarios include overtaking, rapid acceleration, and mountain downhill sections. For battery management, the current is limited to 60%-70% of the rated value to reduce the discharge rate; for the gentle driving mode, its characteristic thresholds are: average torque < 30 N·m, standard deviation of vehicle speed < 5 km / h, average slope ∈ [-0.5°, 0.5°] (flat road). Typical scenarios include smooth driving on urban roads and uniform driving in the suburbs. For battery management, the current limit is relaxed to 85%-90% of the rated value; for the high-speed cruise mode, its characteristic thresholds are: average vehicle speed > 80 km / h, average torque ∈ [40, 60] N·m, gentle slope (within ±1°). Typical scenarios include stable driving on highways or expressways. For battery management, the rated current is maintained and thermal management is started as needed; for the low-speed congestion mode, its characteristic thresholds are: average vehicle speed < 20 km / h, zero-crossing rate of torque > 15 times / 0.5 s, frequent start and stop (such as at traffic light intersections). Typical scenarios include urban congestion sections and construction sections. For battery management, the current is restricted to prevent short-distance high-load discharge and protect the battery life.

[0092] Figure 3 The figure is a schematic structural diagram of a BNN model provided by an embodiment of the present application. The BNN model structure includes an input layer 301, a hidden layer 302, and an output layer 303.

[0093] In some embodiments, the input layer 301 receives the feature representation of the driving data acquired by the MCU 102.

[0094] Exemplarily, the input layer inputs 10 neurons, which can be understood as directly corresponding to 10-dimensional driving behavior statistical features (such as average motor torque, standard deviation of vehicle speed, slope change rate, etc.) extracted at the edge side.

[0095] In some embodiments, the hidden layer 302 extracts the most critical driving mode features with the least amount of computation by gradually reducing the number of neurons layer by layer. It can be divided into three binary fully connected layers. The first binary connection layer performs a preliminary analysis of the input data to extract basic features; the second binary connection layer further compresses the information to focus on the key features; the third binary connection layer refines the core mode to prepare for the final classification.

[0096] Exemplarily, the first layer contains 32 neurons, equivalent to the preliminary input data volume, where basic features are extracted, such as high or low power demand, fast or slow speed change, etc.; the second layer has 16 neurons, equivalent to the data volume after further compressing the data, where key features are focused on, such as acceleration trend, uphill and downhill patterns, etc.; the third layer has 8 neurons, equivalent to the refined core pattern data volume, where the extracted core patterns are, for example, aggressive driving tendency, smooth driving tendency, etc.

[0097] In some embodiments, the output layer 303 classifies driving behaviors. Exemplarily, the output layer outputs 4 neurons, equivalent to four types of driving behaviors, including: aggressive driving, such as rapid acceleration, frequent overtaking, etc.; gentle driving, such as smooth acceleration, constant speed driving, etc.; high-speed cruising, such as stable driving on the highway; low-speed congestion, such as stop-and-go driving on urban roads. In one possible implementation, the output layer calculates the probabilities of each pattern through the Softmax function.

[0098] Exemplarily, Softmax is a function that maps any real number vector to a probability distribution, representing the probability value that the current driving behavior belongs to a certain pattern. For example, the probability of the aggressive mode is 65%, the probability of the gentle mode is 20%, the probability of high-speed cruising is 10%, and the probability of low-speed congestion is 5%. For the driving mode probability, the mode with the highest probability value is taken as the final classification result. For example, if the probability of the aggressive mode is the highest, it is determined as aggressive driving. When the probabilities of multiple patterns are close, for example, the probability of the aggressive mode is 45% and the probability of low-speed congestion is 40%, the system will comprehensively adjust the strategy according to the battery health status and temperature.

[0099] It should be noted that the BNN model uses binary calculation and a small number of neurons, enabling the BNN model to run quickly on the vehicle with extremely low latency and extremely low power consumption. At the same time, the BNN model is small and does not require additional hardware, saving costs.

[0100] Among them, the weights and activation values are binary: only represented by ±1, the model is small, reducing memory occupancy.

[0101] In some embodiments, BNN is a neural network reconstructed based on the ARM NEON instruction set; the matrix operation of BNN is converted from convolution operation to exclusive OR operation.

[0102] Exemplarily, the BNN matrix operation converted from convolution operation to exclusive OR operation is as follows:

[0103] / / Example: Binary matrix multiplication optimization

[0104] uint8x16_t weights = vld1q_u8(weight_ptr);

[0105] uint8x16_t inputs = vld1q_u8(input_ptr);

[0106] uint8x16_t product = vmullq_u8(weights, inputs);

[0107] It can be interpreted that uint8x16_t is a data type representing a 128-bit vector register of NEON, which can store 16 unsigned 8-bit integers (uint8_t); vld1q_u8 is a NEON instruction that continuously loads 16 bytes (i.e., 16 uint8_t) from the memory address weight_ptr into the register; weight_ptr usually points to the weight array after 8-bit quantization; the first line of code means that 16 bytes (i.e., 16 uint8_t) are continuously loaded from the weight array pointed to by the memory address into the register. The second line of code means that 16 8-bit input data are loaded from input_ptr into another register to prepare for calculation. The third line of code means that the corresponding elements of the two uint8x16_t vectors are multiplied, and the result is extended to 16 bits (to prevent overflow), and finally a uint16x8_t vector (8 16-bit results) is output. Finally, instructions such as vaddq_u16 are used to accumulate the results to complete the multiply-accumulate operation of the convolutional or fully connected layer.

[0108] It should be noted that reconstructing the BNN matrix operation through the NEON instruction set can not only significantly improve the parallel efficiency of the binarization operation, but also greatly reduce the memory requirements and optimize the energy consumption performance at the same time.

[0109] In some embodiments, during the process of the BNN performing inference in the previous cycle, the original signal of the driving data of the vehicle is acquired.

[0110] It should be understood that the embodiments of the present application adopt a double-buffer mechanism. When the BNN is performing inference on the current data, the acquisition of the second set of data and the BNN inference can be carried out at the same time. For example, when the MCU acquires data A and transmits the data to the domain controller for processing data A, the MCU can acquire data B and transmit the data to the domain controller, and then process data B, where A is the current set of data and B is the next set of data.

[0111] It should be noted that when calculating the BNN model, the data required for the next inference is taken in parallel to implement the double-buffer mechanism, reducing the end-to-end latency.

[0112] In the battery management method provided by this application, the "ARM NEON instruction set optimization + double buffering mechanism" is adopted to solve the problems that the direct input of the original signal with high acquisition volume into the model leads to an explosion in the amount of calculation, and the mismatch between the recognition of driving behaviors relying on complex models and the low vehicle computing power. At the same time, it further reduces the inference time of the BNN model and the latency of battery management.

[0113] In the battery management method provided by this application, in steps S202 - S203, a hierarchical feature extraction architecture is adopted. This architecture includes: MCU 102 and domain control 103. Among them, MCU 102 belongs to the edge processing layer. MCU 102 can extract the feature representation of driving data from the original signal based on the first sampling rate, realizing the conversion from the original signal to low-dimensional statistical features. The domain controller 103 belongs to the intelligent decision-making layer and can input the low-dimensional statistical features obtained through MCU 102 into the BNN to obtain the probability values of various driving modes. This hierarchical feature extraction architecture not only solves the problems of high-dimensional original signals and data redundancy, but also solves the problems of complex existing models and high computing power requirements. In addition, through task division by the hierarchical architecture, the effect of battery management is improved to meet the real-time requirements.

[0114] S204. Determine the current driving mode of the vehicle based on the probability value, and adjust the rated output current of the vehicle's battery.

[0115] In some embodiments, according to step S203, if the probability value of the obtained driving behavior is used to classify the current driving mode, the mode with the highest probability is taken as the final classification result. Further, the execution layer BMS104 adjusts the rated output power of the vehicle's battery according to the finally classified driving mode.

[0116] In a possible implementation manner, the above S204 can be specifically implemented as: when the probability value indicates that the vehicle is in an aggressive driving mode, the rated output current of the battery is reduced according to the battery health state of the battery; where, when in the aggressive driving mode, the average torque of the vehicle is greater than the first threshold, and the standard deviation of the vehicle speed is greater than the second threshold.

[0117] Among them, SOH represents the degree of attenuation of the current performance of the battery relative to the brand-new state, and the value range is usually 0% - 100%.

[0118] Exemplarily, when in the aggressive driving mode, its characteristic thresholds are: average torque > 70 N·m, standard deviation of vehicle speed > 12 km / h, slope change rate ∈ [-2°, 0°] (downhill acceleration). Typical scenarios include overtaking, rapid acceleration, and mountain downhill sections.

[0119] Exemplarily, when the probability value indicates that the vehicle is in an aggressive driving mode, according to the battery health state of the battery, the rated output current of the battery is reduced. The rated output current of the battery can be reduced by using the following expression:

[0120] A = B × (0.9 - 0.002 × (100 - SOH))

[0121] A = 0.8A

[0122] Wherein, A represents the rated output current of the battery; B represents the maximum rated current of the battery.

[0123] Exemplarily, when the probability value indicates that the vehicle is in an aggressive driving mode, the dynamic current limit formula and code implementation are as follows:

[0124] / / Dynamic current limit formula

[0125] float max_current = rated_current * (0.9 - 0.002 * (100 - soh));

[0126] / / Example: Limit the current to 80% of the rated output current in the aggressive driving mode

[0127] if (model_output == "aggressive driving mode") {

[0128] max_current *= 0.8;

[0129] }

[0130] It can be understood that in the case of the final result being the aggressive driving mode, according to the battery health condition, the dynamic current limit formula is used, and combined with the current driving behavior mode being the aggressive driving mode, the current is further limited to 80% of the rated output current, which can effectively reduce the number of high-rate discharges in the aggressive driving mode, thereby extending the battery life.

[0131] In another possible implementation manner, the battery management method of the embodiment of the present application further includes: adjusting the thermal management parameters when the probability value indicates that the vehicle is in an aggressive driving mode. Wherein, the thermal management parameters at least include the cooling pump speed and the radiator fan power.

[0132] Exemplarily, when it is determined that the vehicle is in an aggressive driving mode, the cooling pump speed is increased, and / or the radiator fan power is increased to avoid the battery being in an overheated state for a long time, thereby delaying the battery life.

[0133] For example, the pump speed in the highest temperature area is increased by 15%.

[0134] It should be noted that in the battery management method of the present application, based on the strategy of dynamic current limit for driving behavior and SOH, by optimizing the battery output strategy, the battery life can be effectively extended. In addition, by collecting driving data to analyze the driving mode of the vehicle and performing precise current regulation, the battery cycle life is increased by 18%-25%, which is better than the traditional fixed threshold scheme. In addition, the battery management method provided by the present application can also support value-added services such as UBI insurance (based on driving behavior scoring) and after-sales battery health reports. Among them, UBI usually refers to Universal Basic Income, which is a social security policy that advocates the government to regularly distribute a fixed amount of cash unconditionally to all citizens (or specific groups) to meet basic living needs regardless of their employment or income status.

[0135] In some embodiments, as Figure 4 shown, before obtaining the original signal of the driving data of the vehicle, a dynamic priority scheduling strategy is carried out.

[0136] Figure 4 It is a schematic flow chart of another battery management method provided by an embodiment of the present application. As Figure 4 shown, the specific steps may include S401-S403.

[0137] S401. Obtain the SOH of the battery.

[0138] Among them, SOH represents the attenuation degree of the current performance of the battery relative to the brand-new state, and the value range is usually 0%-100%.

[0139] In a possible implementation manner, the SOH of the battery is based on the battery parameters real-time monitored by the sensors inside the battery pack, and the SOH can be output through the built-in algorithm formula of the BMS104.

[0140] Exemplarily, SOH≥80%: The battery is in a "healthy state", the performance attenuation is within an acceptable range, and normal charge and discharge cycles are supported.

[0141] Exemplarily, SOH<80%: The battery enters the "aging state", the internal resistance increases, the capacity decreases, and permanent damage is likely to occur under high-load working conditions, such as an aggressive driving mode.

[0142] S402. When the SOH is less than the preset state threshold, set the priority of driving behavior detection higher than the priority of battery temperature monitoring; the driving behavior detection is used to indicate the execution of the foregoing battery management method.

[0143] In some embodiments, when the SOH is less than a preset state threshold, the battery is in a vulnerable period, and high-risk driving behaviors are preferentially identified, such as an aggressive driving mode. In this aggressive driving mode, the circuit management method of the embodiments of the present application can be used to adjust the battery to avoid over-discharging of the battery.

[0144] In some embodiments, the BMS 104 monitors the SOH in real time. When the continuously sampled values are less than the preset state threshold for multiple times, the "high-risk mode" is triggered, and the priority scheduling of the driving behavior detection task is performed, and the battery management method of the present application is executed simultaneously.

[0145] Exemplarily, the task scheduling logic (C language pseudo-code):

[0146]

[0147] Illustratively, the BMS 104 monitors the SOH in real time. When the continuously sampled values are less than 80% for 3 consecutive times, the "high-risk mode" is triggered. The driving behavior detection tasks at least include: raising the driving behavior priority from "medium" to "highest", increasing the computing power allocation ratio, and ensuring that the edge feature extraction and BNN inference are completed strictly on time. The steps S201-204 of the battery management method of the present application are executed.

[0148] For example, the BMS 104 monitors the SOH in real time. When the continuously sampled values are less than 80% for 3 consecutive times, the driving behavior priority is raised from "medium" to "highest", and the computing power allocation ratio is raised from 30% to 60%, ensuring that the edge feature extraction (5 ms) and BNN inference (3 ms) are completed strictly on time. When it is detected that the average torque is greater than 70% of the rated torque and the average first-order difference of the vehicle speed is greater than 5 km / h / s, it is determined as "aggressive driving", and the execution layer immediately limits the maximum current to 60% of the rated value (80% in the traditional solution), and increases the cooling power of the thermal management system by 20%.

[0149] It should be noted that the related technology effectively extends the remaining battery life by monitoring the SOH in real time and performing early intervention in a timely manner.

[0150] S403. When the SOH is greater than or equal to the preset state threshold, set the priority of the battery temperature monitoring higher than the priority of the driving behavior detection.

[0151] In some embodiments, when the SOH is greater than or equal to the preset state threshold, the battery is in the battery healthy period, and temperature abnormalities (overheating / cooling) are preferentially prevented to avoid accelerated aging.

[0152] In some embodiments, the BMS 104 monitors the SOH in real time. When the sampled values are greater than or equal to the preset status threshold for multiple consecutive times, the system is in the "conventional monitoring mode", and the priority scheduling of the temperature monitoring task and the driving behavior detection task is performed.

[0153] Exemplarily, the BMS 104 monitors the SOH in real time. When the sampled values are greater than or equal to 80% for 3 consecutive times, the system is in the "conventional monitoring mode". Performing the temperature monitoring task at least includes setting the priority of the battery temperature monitoring to "highest", increasing the computing power allocation ratio, regularly collecting the temperatures of the battery cells. When the temperature is greater than the predetermined first threshold, triggering the thermal balance control, such as starting the liquid cooling pump, and displaying that the temperature drops to the second threshold. Performing the driving behavior detection task, maintaining the priority as "medium", reducing the computing power ratio, and triggering in-depth analysis when abnormal features are detected.

[0154] For example, the BMS 104 monitors the SOH in real time. When the sampled values are greater than or equal to 80% for 3 consecutive times, the priority of the battery temperature monitoring is set to "highest", the computing power allocation ratio is 50%, the temperatures of each battery cell are collected every 10 ms (sampling rate 100 Hz). When the temperature difference > 5 °C, trigger the thermal balance control (such as starting the liquid cooling pump), and the target is to reduce the temperature difference to within 2 °C within 300 ms. The driving behavior priority is maintained at "medium", the computing power allocation ratio is 30%, and in-depth analysis is only triggered when abnormal features are detected (such as the peak-to-peak torque > 80 N·m), avoiding over-allocation of computing power to the driving behavior detection during the healthy period and reducing the overall power consumption by about 15%.

[0155] It should be noted that the related technology not only realizes the intelligent allocation of computing power resources to ensure the real-time performance of key tasks (such as battery protection), but also extends the battery life by preferentially processing high-risk scenarios.

[0156] Figure 5 This is the complete flowchart of a battery management method provided by the embodiments of the present application. As Figure 5 shown, the specific steps include S501 - S505.

[0157] S501. Data acquisition: Obtain the original signal (100 Hz) through the CAN bus. (Corresponding to the above step S201)

[0158] S502. Edge feature extraction: The MCU calculates 10-dimensional statistical features (completed within 5 ms). (Corresponding to the above step S202)

[0159] S503. BNN inference: The domain controller outputs the driving mode probability and determines the driving mode classification (completed within 3 ms). (Corresponding to the above step S203)

[0160] S504. Dynamic current limit: Adjust the maximum current according to the SOH and driving mode. (Corresponding to step S204 above)

[0161] Exemplarily, the complete process of the battery management method of the present application is as follows: Obtain the 100HZ original signal through the CAN bus to achieve data acquisition, then complete the MCU calculation of 10-dimensional statistical features within 5 ms to achieve edge feature extraction, then complete the domain controller to adopt BNN inference to output the driving mode probability and obtain the driving model classification within 3 ms, and finally adjust the maximum current according to the SOH and driving mode to achieve dynamic current limit.

[0162] The embodiment of the present application provides a battery management method, which obtains the original signal of the driving data of the vehicle; extracts the feature representation of the driving data from the original signal based on the first sampling rate; the first sampling rate is less than the sampling rate of the original signal; inputs the feature representation into the binary neural network BNN to obtain the probability values of various driving modes; determines the current driving mode of the vehicle based on the probability values, and adjusts the rated output current of the battery of the vehicle. It can be seen that in the process of managing the battery in the embodiment of the present application, the feature is extracted by dimension reduction from the original signal, and the BNN is used for inference to obtain the probability value of the driving mode. On the one hand, extracting the feature representation by dimension reduction can reduce the amount of data input to the model. On the other hand, BNN is a binary lightweight model, with a small model size and a high inference rate. Based on the above two aspects, the driving mode can be quickly recognized and the rated output current of the battery can be adjusted to ensure the real-time performance of battery management, thereby improving the battery life. In addition, reducing the amount of data to be processed and adopting a lightweight model can reduce the requirements for in-vehicle computing power and save costs.

[0163] The following content is another optional implementation manner in the technology of the present application.

[0164] In some embodiments, compared with the solution of step S201 that relies on the CAN bus to collect the original signal, another optional implementation manner is to adopt the fusion of a camera and a radar, assist in classifying driving behaviors through visual recognition (such as emergency brake light detection) or millimeter-wave radar data, use an inertial measurement unit (IMU), detect sudden acceleration or sudden braking by using an accelerometer to supplement the CAN data, use a driver biometric strategy, and monitor the driver's fatigue state through a steering wheel pressure sensor or a camera to indirectly infer driving behaviors.

[0165] Among them, visual recognition is a technology that captures environmental information through optical sensors (such as cameras, event cameras) and uses deep learning models (such as ViT-4.0, YOLOv9) to parse pixel-level semantics in real time to achieve object detection, classification, tracking, and scene understanding.

[0166] Among them, millimeter-wave radar data is 4D environmental perception data (distance, azimuth, elevation angle, speed) generated by the reflected signal of 77 / 79 GHz electromagnetic waves, with the ability to work all-weather, and is mainly used for dynamic target tracking and ranging.

[0167] Among them, the inertial measurement unit is a microelectronic system that captures the acceleration, angular velocity and attitude changes of an object in real time through multi-axis motion sensors.

[0168] Among them, the driver biometric strategy is an intelligent solution that monitors and verifies the driver's identity and status in real time through multi-modal biometric technologies and dynamically adjusts the vehicle system.

[0169] It should be noted that related technologies improve the robustness of behavior recognition through multi-source sensors.

[0170] In some embodiments, compared with the solution of using the sliding window statistic method for feature extraction in step S202, another alternative embodiment is to perform frequency domain feature extraction, extract the frequency domain features of motor torque and vehicle speed through the Fast Fourier Transform (FFT), such as the dominant frequency, energy distribution, etc., extract event trigger features, calculate features only when significant changes (such as torque mutation) are detected, reduce the calculation amount, perform machine learning feature selection, and dynamically select key features using PCA or LDA algorithms to reduce the dimension.

[0171] Among them, the Fast Fourier Transform is an optimized algorithm of the Discrete Fourier Transform (DFT), which reduces the computational complexity of O(N 2 ) to O(NlogN) through a divide-and-conquer strategy.

[0172] It should be noted that related technologies achieve data dimensionality reduction through different feature engineering methods to meet the real-time requirements.

[0173] In some embodiments, compared with the BNN model in step S203, another alternative embodiment is to adopt the Quantized Convolutional Neural Networks (QCNN), quantize the weights and activation values to 8 bits or 16 bits, reduce the calculation amount while maintaining the accuracy, perform knowledge distillation, transfer the knowledge of complex models (such as LSTM) to lightweight models (such as MobileNetV3), and adopt the Spiking Neural Network (SNN) to utilize the event-driven sparse calculation characteristics to reduce power consumption.

[0174] Among them, the quantized convolutional neural network is a model compression technology that reduces the numerical precision of the neural network parameters (weights / activation values), sacrificing a minimal model performance to significantly improve the computational efficiency and energy efficiency ratio.

[0175] Among them, the spiking neural network is the third-generation neural network model, which performs calculations by simulating the pulse timing coding and event-driven mechanism of biological neurons.

[0176] It should be noted that the related technology realizes the lightweight of the model without using binarization.

[0177] In some embodiments, compared with the "ARM NEON instruction set optimization + double buffering mechanism" adopted in step S203, another alternative embodiment is to use FPGA hardware acceleration, map the BNN model to the logic unit of the FPGA, accelerate the inference through parallel computing, perform RISC-V instruction set extension, customize RISC-V instructions for specific operations (such as binary multiplication), improve the energy efficiency ratio, adopt an asynchronous processing architecture, decouple data acquisition and model inference, and reduce latency through an asynchronous FIFO queue.

[0178] Among them, FPGA is a field-programmable gate array that realizes user-customized digital circuits through the flexible combination of hardware logic units, interconnection resources, and configurable I / Os.

[0179] It should be noted that the related technology achieves equivalent or higher acceleration effects on different hardware platforms.

[0180] In some embodiments, compared with the hierarchical feature extraction architecture of "edge processor + domain controller" adopted in steps S202 - S203, another alternative embodiment is to adopt an edge-cloud collaboration architecture, complete feature extraction at the edge, run a complex model in the cloud to generate a global optimization strategy, perform heterogeneous multi-core processing, integrate multiple processor cores in a single SoC (such as NXP S32G), parallel process different tasks, adopt a distributed architecture, and distribute feature extraction and model inference to multiple edge nodes (such as the microcontrollers built into each sensor).

[0181] It should be noted that the related technology balances real-time performance, cost, and scalability through different architecture designs.

[0182] In some embodiments, compared with the linear formula based on SOH in step S204 (max_current = rated_current * (0.9 - 0.002 * (100 - soh))), another alternative embodiment is to adopt fuzzy logic control, establish a fuzzy rule base, dynamically adjust the current limit by integrating parameters such as SOH, temperature, and driving mode, use a reinforcement learning model, optimize the current strategy through online learning to adapt to different driving scenarios, use a look-up table method, pre-define the current limit thresholds for different combinations of SOH and driving modes, and execute by looking up the table.

[0183] It should be noted that in the related art, more accurate current regulation is achieved through non-linear or adaptive methods.

[0184] In some embodiments, compared with the method of adjusting the task priority based on the battery SOH in step S204, another alternative embodiment is to schedule based on temperature. When the battery temperature exceeds the threshold, the thermal management task is preferentially executed, risk level scheduling is performed, the priority is dynamically allocated by combining the driving mode (such as an aggressive mode) and environmental factors (such as slope), and cloud collaborative scheduling is adopted. The cloud road condition data is obtained through LTE / V2X to adjust the local task priority.

[0185] Among them, LTE-V2X (Long-Term Evolution Vehicle-to-Everything) is a vehicle networking communication protocol optimized based on 4G LTE, which realizes low-latency direct communication between vehicles and everything in the environment (V2X).

[0186] It should be noted that in the related art, more refined resource allocation is achieved through multi-dimensional metrics.

[0187] In an exemplary embodiment, the present application further provides a battery management device. The battery management device may include one or more functional modules for implementing the battery management method described in the above method embodiments.

[0188] For example, Figure 6 is a schematic diagram of the composition of a battery management device provided by an embodiment of the present application. As Figure 6 shown, the battery management device includes: an acquisition module 601, a processing module 602, and an adjustment module 603.

[0189] The acquisition module 601 is used to acquire the original signal of the driving data of the vehicle; the acquisition module 601 is further used to extract the feature representation of the driving data from the original signal based on the first sampling rate; the first sampling rate is less than the sampling rate of the original signal; the processing module 602 is used to input the feature representation into the binary neural network BNN to obtain the probability values of multiple driving modes; the adjustment module 603 is used to determine the current driving mode of the vehicle based on the probability values and adjust the rated output current of the vehicle's battery.

[0190] Optionally, the adjustment module 603 is specifically used to reduce the rated output current of the battery according to the battery health state of the battery when the probability value indicates that the vehicle is in an aggressive driving mode; when in the aggressive driving mode, the average torque of the vehicle is greater than the first threshold, and the standard deviation of the vehicle speed is greater than the second threshold.

[0191] Optionally, the adjustment module 603 is specifically used to reduce the rated output current of the battery by using the following expression:

[0192] A = B×(0.9 - 0.002×(100 - SOH))

[0193] A = 0.8A

[0194] Wherein, A represents the rated output current of the battery; B represents the maximum rated current of the battery.

[0195] Optionally, the vehicle includes an edge controller; the acquisition module 601 is specifically used to extract the feature representation of the driving data from the original signal by the edge controller based on the first sampling rate.

[0196] Optionally, the vehicle further includes a domain controller; the acquisition module 601 is specifically used to obtain the feature representation from the edge controller by the domain controller; the domain controller inputs the feature representation into the BNN to obtain the probability values of multiple driving modes.

[0197] Optionally, the acquisition module 601 is specifically used to extract the feature representation of the driving data from the original signal by using sliding window statistics; wherein, the number of sampling points in the sliding window is determined according to the first sampling rate.

[0198] Optionally, the BNN is a neural network reconstructed based on the ARM NEON instruction set; the matrix operation of the BNN is converted from a convolution operation to an exclusive OR operation.

[0199] Optionally, the acquisition module 601 is specifically used to acquire the original signal of the driving data of the vehicle during the inference process of the BNN in the previous cycle.

[0200] Optionally, the obtaining module 601 is further configured to obtain the battery health status of the battery; the processing module 602 is further configured to set the priority of the driving behavior detection to be higher than the priority of the battery temperature monitoring when the battery health status is lower than a preset status threshold; the driving behavior detection is used to indicate the execution of the battery management method;

[0201] Otherwise, set the priority of the battery temperature monitoring to be higher than the priority of the driving behavior detection.

[0202] In an exemplary embodiment, the embodiment of the present application further provides a vehicle, including a controller, and the controller is configured to execute the foregoing battery management method.

[0203] In an exemplary embodiment, the embodiment of the present application further provides a computer-readable storage medium, on which computer program instructions are stored; when the computer program instructions are executed by an electronic device, the electronic device is caused to implement the method described in the foregoing embodiments. The computer-readable storage medium may be a non-transitory computer-readable storage medium. For example, the non-transitory computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0204] In an exemplary embodiment, the embodiment of the present application further provides a computer program product, when the computer program product runs on a computer, the computer is caused to execute the above-related method steps to implement the battery management method in the above embodiments.

[0205] In the description of the embodiments of the present application, specific features, structures, materials, or characteristics may be combined in a suitable manner in any one or more embodiments or examples.

[0206] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all of them should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A battery management method, characterized in that, Applied to a vehicle, the method includes: Obtain the original signal of the driving data of the vehicle; Extract the feature representation of the driving data from the original signal based on a first sampling rate; the first sampling rate is less than the sampling rate of the original signal; Input the feature representation into a binary neural network (BNN) to obtain probability values of multiple driving modes; Determine the current driving mode of the vehicle based on the probability values, and adjust the rated output current of the vehicle's battery.

2. The method according to claim 1, wherein The determining the current driving mode of the vehicle based on the probability values and adjusting the rated output current of the vehicle's battery includes: In the case where the probability values indicate that the vehicle is in an aggressive driving mode, reduce the rated output current of the battery according to the state of health (SOH) of the battery; in the case of an aggressive driving mode, the average torque of the vehicle is greater than a first threshold, and the standard deviation of the vehicle speed is greater than a second threshold.

3. The method according to claim 2, wherein The reducing the rated output current of the battery according to the state of health (SOH) of the battery includes: Reduce the rated output current of the battery using the following expression: A = B × (0.9 - 0.002 × (100 - SOH)) A = 0.8A Wherein, A represents the rated output current of the battery; B represents the maximum rated current of the battery.

4. The method according to claim 1, wherein The vehicle includes an edge controller; The extracting the feature representation of the driving data from the original signal based on a first sampling rate includes: The edge controller extracts the feature representation of the driving data from the original signal based on a first sampling rate.

5. The method according to claim 4, characterized in that The vehicle further includes a domain controller; The inputting the feature representation into the BNN to obtain probability values of multiple driving modes includes: The domain controller obtains the feature representation from the edge controller; The domain controller inputs the feature representation into the BNN to obtain probability values of multiple driving modes.

6. The method according to claim 4, wherein The extracting the feature representation of the driving data from the original signal based on a first sampling rate includes: Adopt sliding window statistics to extract the feature representation of the driving data from the original signal; wherein, the number of sampling points within the sliding window is determined according to the first sampling rate.

7. The method according to claim 1, characterized in that The BNN is a neural network reconstructed based on the ARM NEON instruction set; the matrix operations of the BNN are converted from convolution operations to exclusive OR operations.

8. The method according to claim 1, wherein Obtaining the original signal of the driving data of the vehicle includes: During the inference process of the BNN in the previous cycle, obtain the original signal of the driving data of the vehicle.

9. The method according to claim 1, characterized in that, Before obtaining the original signal of the driving data of the vehicle, the method further includes: Obtain the state of health (SOH) of the battery; In the case where the state of health (SOH) of the battery is less than a preset state threshold, set the priority of driving behavior detection to be higher than the priority of battery temperature monitoring; the driving behavior detection is used to indicate the execution of the battery management method; Otherwise, set the priority of battery temperature monitoring to be higher than the priority of driving behavior detection.

10. A vehicle, characterized in that, Including a controller, the controller is configured to execute the battery management method according to any one of claims 1 - 9.

Images