Battery BMS Parameter Calibration Method and Device

Abstract

The present invention belongs to the technical field of BMS management, and discloses a battery BMS parameter calibration method and device. The method includes: collecting internal and external data of the battery and user data, extracting voltage and current signals from the internal data, and filtering them through an acceleration-ripple coupling model; dividing external data into vehicle working condition segments and forming working condition segment labels; building a dynamic equivalent model, setting trigger rules to determine the trigger mode; inputting multiple types of data into a bidirectional feature fusion network to obtain fusion features, and then extracting relevant features. Using an error prediction model to obtain the SOC estimation error, obtaining a population label through a population classification model, and finally adjusting the BMS parameters based on reinforcement learning combined with multiple factors. The present invention significantly improves the SOC estimation accuracy, effectively delays the battery aging rate, and enhances the safety and energy utilization rate under complex working conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of BMS management, and more specifically, to a method and device for calibrating battery BMS parameters. Background Art

[0002] Chinese Patent Application No. CN117767462A discloses a method and device for automatically generating a general BMS control model. The method includes: S1. Obtaining battery data in multiple battery array units, attaching battery number information, and uploading it to the information processing center module in real time; S2. In the information processing center module, storing, sorting, statistically analyzing the obtained battery data in real time through a learning mode; S3. Based on the analysis of the battery data, establishing an intelligent battery management system control neural network model through deep learning to obtain an initial BMS control model; S4. Using the obtained initial BMS control model to generate a core working program and a call interface; and completing data verification during charge and discharge, comparing with the theoretical simulation effect, and simultaneously recording the deviation values of parameter indicators; S5. Judging the deviation values and adjusting the parameters of the BMS control model; S6. Solidifying the adjusted BMS control model parameters to generate an overall model adapted to the usage conditions of different batteries, that is, a general BMS control model. This invention selects the best working strategy according to different scenarios and requirements to achieve the optimal operation of the battery system.

[0003] Although the above method can meet most scenarios, through research and practical application of the above method and the existing technology, it is found that the above method and the existing technology have at least the following partial defects:

[0004] No dynamic coupling relationship between battery model parameters (such as internal resistance) and states (such as SOC) is established, resulting in the separation of parameter calibration and state estimation; the vehicle hardware characteristics (such as heat dissipation limitations caused by the compact space of two-wheel vehicles) or user behaviors (such as high-frequency charge and discharge of express delivery vehicles) are not considered during the model generation process, and the generated control strategy may deviate significantly from the actual requirements.

[0005] In view of this, the present invention proposes a method and device for calibrating battery BMS parameters to solve the above problems. Summary of the Invention

[0006] To overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: A method for calibrating battery BMS parameters, including the following steps:

[0007] Collect the internal data of a lithium iron phosphate battery, extract voltage signals and current signals from the internal data, and set an acceleration-ripple coupling model for adaptive filtering to obtain voltage and current;

[0008] Collect external data of the battery, divide the vehicle running state into different working condition segments based on the external data, and form working condition segment labels;

[0009] Build a dynamic equivalent model, divide the dynamic equivalent model into a basic model and an extended model, set model trigger rules, and determine the detected current and battery temperature based on the model trigger rules to determine whether to trigger the extended mode or the basic mode;

[0010] Collect user data, extract the battery temperature from the internal data, and use the voltage, current, battery temperature, external data, and user data as the input of the bidirectional feature fusion network to obtain fusion features;

[0011] Extract the vibration frequency from the external data, perform feature extraction on the voltage, current, and vibration frequency to obtain the voltage change rate, current pulse characteristics, and vibration intensity;

[0012] Use the fusion features, voltage change rate, and current pulse characteristics as the input of the error prediction model to obtain the SOC estimation error;

[0013] Use the external data, user data, and continuous working condition segment labels during the vehicle operation as the input of the population classification model to obtain population labels;

[0014] Adjust the BMS parameters based on reinforcement learning combined with the dynamic equivalent model, vibration intensity, SOC estimation error, and population labels.

[0015] Furthermore, the method for adjusting the BMS parameters includes:

[0016] Define the state space : including the SOC estimation error, battery temperature, instantaneous current overlimit flag, and vibration intensity; among them, the instantaneous current overlimit flag is obtained by comparing the current with the preset current overlimit threshold. When the current exceeds the preset current overlimit threshold, the value is 1, otherwise it is 0;

[0017] Define the action space : including the adjustment amount of the BMS parameters and the fan start / stop action;

[0018] Set the reward function based on the safety reward, accuracy penalty, parameter voltage stability penalty, overheat penalty, polarization suppression penalty, and energy efficiency penalty ;

[0019] Use the Q-learning algorithm to learn the optimal policy, initialize the Q-value function as a table, where the rows of the table correspond to different environmental states and the columns correspond to different actions , set the learning rate , discount factor and exploration rate ; At each time step , detect the current environmental state of the lithium iron phosphate battery , according to the exploration rate , with probability, randomly select an action , with probability, select the action with the maximum Q value in the current environmental state ; Execute the selected action to obtain a new environmental state , meanwhile, calculate the reward function value according to the reward function; Update the Q value function according to the update formula;

[0020] Repeat updating the Q value function until the Q value converges to obtain the corresponding optimal Q value function. According to the environmental state, use the optimal Q value function to select and execute the optimal action from the action space.

[0021] Further, the working condition segments include starting, stopping, accelerating, decelerating, constant speed, climbing, descending, turning and bumping; The methods for obtaining the working condition segment labels include:

[0022] Extract the throttle opening, steering angle, acceleration, pitch angle, roll angle and vibration frequency from the external data of the battery, and detect whether the throttle opening increases from 0 to greater than the preset opening threshold value, and whether the acceleration increases from 0 to greater than the preset acceleration threshold value; If so, it is classified as starting and marked with a starting label; Detect whether the throttle opening drops to 0 and whether the acceleration remains less than 0 until the speed drops to 0; If so, it is classified as stopping and marked with a stopping label; Calculate the derivative of the throttle opening and detect whether the derivative of the throttle opening is greater than the preset acceleration derivative threshold value and whether the acceleration is greater than the preset acceleration threshold value; If so, it is classified as accelerating and marked with an accelerating label; Detect whether the derivative of the throttle opening is less than the preset deceleration derivative threshold value and whether the acceleration is less than 0; If so, it is classified as decelerating and marked with a decelerating label;

[0023] Detect whether the change value of the throttle opening is less than the preset fluctuation threshold value, whether the change value of the acceleration is less than the preset acceleration change threshold value, and whether the duration is greater than the preset constant speed time threshold value; If so, it is classified as constant speed and marked with a constant speed label;

[0024] Detect whether the pitch angle is greater than the preset pitch angle climbing threshold value and whether the throttle opening is greater than the preset climbing throttle opening threshold value; If so, it is classified as climbing and marked with a climbing label; Detect whether the pitch angle is less than the preset pitch angle descending threshold value and whether the throttle opening is less than the preset descending throttle opening threshold value; If so, it is classified as descending and marked with a descending label;

[0025] Detect whether the steering angle is greater than the preset angle threshold, whether the roll angle is greater than the preset roll angle threshold, and whether the duration is greater than the preset turning time threshold; if so, classify it as a turn and mark it with a turn label; detect whether the vibration frequency is greater than the preset vibration frequency threshold, whether the acceleration standard deviation is greater than the preset standard deviation threshold, and whether the duration is greater than the preset bump time threshold; if so, classify it as a bump and mark it with a bump label.

[0026] Further, the method for obtaining voltage and current includes:

[0027] Perform adaptive filtering on the voltage signal and current signal respectively by combining the compensation coefficient, the acceleration of the vehicle operation, and the ripple frequency to obtain the corresponding voltage and current.

[0028] Further, the method for building the basic mode of the dynamic equivalent model includes:

[0029] When the lithium iron phosphate battery is in a stable state, apply a current pulse, measure the instantaneous change in the voltage at the battery terminal, and calculate the value of the battery internal resistance according to Ohm's law;

[0030] Conduct charge and discharge experiments on the lithium iron phosphate battery by applying rectangular current pulses until a stable voltage is reached and the protection function of the lithium iron phosphate battery is not triggered, and synchronously collect the voltage of the lithium iron phosphate battery;

[0031] Use the first-order RC equivalent circuit model to build the terminal voltage equation and the first-order dynamic equation of the polarization voltage;

[0032] Derive and fit the curves of the terminal voltage equations for the charging pulse stage and the discharging pulse stage respectively;

[0033] Aiming at minimizing the mean square error between the terminal voltage obtained from the charge and discharge experiments and the theoretical terminal voltage, use the nonlinear least squares method to iteratively update to obtain the first-order polarization resistance and the first-order polarization capacitance values;

[0034] Connect in parallel with and connect them in series to obtain the basic mode of the dynamic equivalent model; where is the internal resistance of the lithium iron phosphate battery.

[0035] Further, the method for building the extended mode of the dynamic equivalent model includes:

[0036] Apply two sets of rectangular current pulses with different durations to the lithium iron phosphate battery; synchronously collect the voltage and pulse current waveforms of the lithium iron phosphate battery;

[0037] Use the second-order RC equivalent circuit model to build the terminal voltage equation and the second-order dynamic equation of the polarization voltage;

[0038] With the goal of minimizing the mean square error between the terminal voltage obtained from the charge-discharge experiment and the theoretical terminal voltage, a natural inspiration optimization algorithm is used to optimize and obtain the second-order polarization resistance , the second-order polarization capacitance , the second-order polarization resistance , the second-order polarization capacitance .

[0039] Furthermore, the method for determining the detection current and the battery temperature based on the model trigger rule to determine whether to trigger the extended mode or the basic mode includes:

[0040] If the detection current and the battery temperature meet the preset extended trigger conditions, the extended mode of the dynamic equivalent model is triggered; otherwise, the basic mode of the dynamic equivalent model is triggered.

[0041] Furthermore, the training method of the bidirectional feature fusion network includes:

[0042] Pre-collect K groups of training data. The training data includes input data and fusion features. The input data includes voltage, current, battery temperature, external data, and user data;

[0043] Taking the input data as the input of the bidirectional feature fusion network and the fusion feature as the output of the bidirectional feature fusion network, with the goal of minimizing the error between the output fusion feature and the actual fusion feature, the network parameters of the bidirectional feature fusion network are optimized through a natural inspiration optimization algorithm to obtain the network parameters corresponding to the minimum error between the fusion feature output by the bidirectional feature fusion network and the actual fusion feature, and the bidirectional feature fusion network constructed with the corresponding network parameters is used as the trained bidirectional feature fusion network.

[0044] Furthermore, the method for obtaining the voltage change rate, the current pulse feature, and the vibration intensity includes:

[0045] Calculate the ratio of the voltage change to the corresponding time interval, and take the ratio as the voltage change rate;

[0046] If the detected current is greater than the preset minimum amplitude threshold, the current at the previous moment is not greater than the preset minimum amplitude threshold, and the current duration is not less than the preset minimum duration threshold, it is determined that the current pulse starts; if the detected current is not greater than the preset minimum amplitude threshold, the current at the previous moment is greater than the preset minimum amplitude threshold, and the current duration is not less than the preset minimum duration threshold, it is determined that the current pulse ends, and the start time, end time, and maximum amplitude of the current pulse are recorded;

[0047] Calculate the ratio of the number of pulses per unit time to the difference between the start time and the end time of the corresponding current pulse to obtain the pulse density;

[0048] The integral of the calculated current within the start time and end time of the current pulse is obtained to acquire the pulse energy;

[0049] The ratio of the current difference in the current rising stage to the difference between the start time and end time of the current pulse is calculated to obtain the rising edge slope;

[0050] The pulse density, pulse energy, and rising edge slope are concatenated to obtain the current pulse characteristics;

[0051] The root mean square of the calculated acceleration is obtained to acquire the vibration intensity.

[0052] Further, the method for obtaining the throttle opening includes:

[0053] Obtain the throttle rotation angle range of the vehicle throttle grip ; where is the maximum throttle rotation angle; based on the quantization function, the throttle rotation angle range is quantized into a preset quantization range to obtain the quantized value of the throttle opening; the quantized value of the throttle opening is used as the throttle opening.

[0054] The battery BMS parameter calibration device for implementing the battery BMS parameter calibration method includes:

[0055] The first analysis module: Collect the internal data of the lithium iron phosphate battery, extract the voltage signal and current signal from the internal data, and set up an acceleration-ripple coupling model for adaptive filtering to obtain the voltage and current;

[0056] The second analysis module: Collect the external data of the battery, divide the vehicle running state into different working condition segments based on the external data, and form working condition segment labels;

[0057] The model building module: Build a dynamic equivalent model, divide the dynamic equivalent model into a basic model and an extended model, set model trigger rules, and determine the detection current and battery temperature based on the model trigger rules to determine whether to trigger the extended mode or the basic mode;

[0058] The data fusion module: Collect user data, extract the battery temperature from the internal data, and use the voltage, current, battery temperature, external data, and user data as the input of the bidirectional feature fusion network to obtain the fusion features;

[0059] The feature extraction module: Extract the vibration frequency from the external data, perform feature extraction on the voltage, current, and vibration frequency to obtain the voltage change rate, current pulse characteristics, and vibration intensity;

[0060] The error estimation module: Use the fusion features, voltage change rate, and current pulse characteristics as the input of the error prediction model to obtain the SOC estimation error;

[0061] Group Classification Module: Using external data, user data, and continuous working condition segment labels during the vehicle operation as the input of the group classification model to obtain group labels;

[0062] Parameter Calibration Module: Adjust the BMS parameters based on reinforcement learning combined with a dynamic equivalent model, vibration intensity, SOC estimation error, and group labels.

[0063] Technical Effects and Advantages of the Battery BMS Parameter Calibration Method and Device of the Present Invention:

[0064] Through the fusion of multi-source data and the coupling of dynamic models, the present invention breaks through the limitation of the separation of traditional BMS parameter calibration and state estimation; realizes high-precision filtering of voltage and current signals through an acceleration-ripple coupling model to eliminate noise interference; based on working condition segment labels and a group classification model, transforms vehicle hardware characteristics and user behaviors into quantifiable control parameters, enabling the model parameters to deeply match the actual operation scenario; the dynamic equivalent model realizes the two-way dynamic adjustment of parameters and states by real-time coupling of internal resistance and SOC state through an extended mode; the two-way feature fusion network integrates internal and external data, combines vibration intensity, SOC error feedback, and a reinforcement learning algorithm to form a closed-loop optimization mechanism, continuously optimizing the charge and discharge strategy, thermal management threshold, and capacity correction coefficient, significantly improving the SOC estimation accuracy, effectively delaying the battery aging rate, and enhancing the safety and energy utilization rate under complex working conditions. Description of the Drawings

[0065] Figure 1 It is a schematic flowchart of the battery BMS parameter calibration method according to Embodiment 1 of the present invention;

[0066] Figure 2 It is a schematic flowchart of the method for adjusting BMS parameters according to Embodiment 1 of the present invention;

[0067] Figure 3 It is a schematic flowchart of the method for dynamically adjusting the charge and discharge priorities of adjacent battery cells according to Embodiment 2 of the present invention;

[0068] Figure 4 It is a schematic structural diagram of the battery BMS parameter calibration device of the present invention. Detailed Embodiments

[0069] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0070] Embodiment 1

[0071] Please refer to Figure 1 as shown, this embodiment provides a method for calibrating battery BMS parameters, including the following steps:

[0072] Collect the internal data of the lithium iron phosphate battery, extract the voltage signal and current signal from the internal data, and set an acceleration-ripple coupling model for adaptive filtering to obtain voltage and current; the internal data includes voltage signal, current signal and battery temperature, and the internal data is the core basis for calibrating the BMS parameters of the lithium iron phosphate battery, which can be obtained by sensor collection. The voltage signal directly reflects the real-time state of charge and health state of the battery, providing a basis for SOC estimation and overvoltage protection threshold setting; the current signal is used to accurately calculate the battery charge and discharge rate and energy loss, ensuring the dynamic adjustment of the BMS to the battery power output; the battery temperature data calibrates the SOC correction algorithm and optimizes the thermal management strategy by affecting the internal resistance and chemical activity, avoiding capacity misjudgment and safety risks in high or low temperature environments. The combination of the three can achieve the precise perception and control of the battery state by the BMS.

[0073] The methods for obtaining voltage and current include:

[0074] Perform adaptive filtering on the voltage signal and current signal respectively in combination with the compensation coefficient, the acceleration and ripple frequency of the vehicle operation to obtain the corresponding voltage and current, such as ; where is voltage or current; is voltage signal or current signal; is acceleration; is ripple frequency, which can be obtained by time-domain measurement with an oscilloscope; is compensation coefficient, which can be obtained by optimizing through nature-inspired optimization algorithms.

[0075] Collect the external data of the battery, divide the vehicle operation state into different working condition segments based on the external data, and form working condition segment labels; the external data includes vehicle type (such as marked as 1 for two-wheeled vehicles and marked as 2 for three-wheeled vehicles), speed, throttle opening, steering angle, acceleration, pitch angle, roll angle and vibration frequency. The external data provides a scenario-based basis for calibrating the BMS parameters by describing the actual operation conditions of the battery, and can be obtained by sensor collection. The vehicle type determines the basic load characteristics of the battery (such as the power demand difference between two-wheeled vehicles and three-wheeled vehicles); speed, throttle opening and acceleration reflect the dynamic behavior of the vehicle, helping the BMS adjust the discharge rate limit and energy recovery strategy; steering angle, pitch angle and roll angle are related to the change of the vehicle center of gravity, assisting in calibrating the mechanical stress compensation parameters of the battery pack; the vibration frequency data is used to evaluate the seismic performance of the battery pack, optimize the mechanical structure design or health state warning threshold. These data together support the stable operation of the BMS under complex working conditions.

[0076] The method for obtaining the throttle opening includes:

[0077] Obtain the throttle rotation angle range of the vehicle throttle grip ; where is the maximum throttle rotation angle; Quantize the throttle rotation angle range to a preset quantization range (such as 0 - 100) based on the quantization function ; where is the throttle opening quantization value; is the collected throttle rotation angle; Use the throttle opening quantization value as the throttle opening.

[0078] The working condition segments include start, stop, acceleration, deceleration, constant speed, climbing, downhill, turning, and bumping. The method for obtaining the working condition segment labels includes:

[0079] Detect whether the throttle opening increases from 0 to greater than the preset opening threshold, and whether the acceleration increases from 0 to greater than the preset acceleration threshold; If so, classify it as a start and mark the start label;

[0080] Detect whether the throttle opening drops to 0, and whether the acceleration remains less than 0 until the speed drops to 0; If so, classify it as a stop and mark it with the stop label;

[0081] Calculate the derivative of the throttle opening, and detect whether the derivative of the throttle opening is greater than the preset acceleration derivative threshold and whether the acceleration is greater than the preset acceleration threshold; If so, classify it as acceleration and mark it with the acceleration label;

[0082] Detect whether the derivative of the throttle opening is less than the preset deceleration derivative threshold and whether the acceleration is less than 0; If so, classify it as deceleration and mark it with the deceleration label;

[0083] Detect whether the change value of the throttle opening is less than the preset fluctuation threshold, whether the change value of the acceleration is less than the preset acceleration change threshold, and whether the duration is greater than the preset constant speed time threshold; If so, classify it as constant speed and mark it with the constant speed label;

[0084] Detect whether the pitch angle is greater than the preset pitch angle climbing threshold and whether the throttle opening is greater than the preset climbing throttle opening threshold; If so, classify it as climbing and mark it with the climbing label;

[0085] Detect whether the pitch angle is less than the preset pitch angle downhill threshold and whether the throttle opening is less than the preset downhill throttle opening threshold; If so, classify it as downhill and mark it with the downhill label;

[0086] Detect whether the steering angle is greater than the preset angle threshold, whether the roll angle is greater than the preset roll angle threshold, and whether the duration is greater than the preset turning time threshold; if so, classify it as a turn and mark it with a turn label.

[0087] Detect whether the vibration frequency is greater than the preset vibration frequency threshold, whether the acceleration standard deviation is greater than the preset standard deviation threshold, and whether the duration is greater than the preset bump time threshold; if so, classify it as a bump and mark it with a bump label.

[0088] The working condition segment labels obtained through the above rules can accurately identify the load characteristics of the battery in different dynamic scenarios. For example, the high current discharge demand corresponding to the climbing working condition can calibrate the correction coefficient of the SOC estimation model to avoid misjudgment of the capacity caused by the increase in internal resistance; the lateral acceleration in the turning working condition can trigger the mechanical stress compensation algorithm of the battery pack to optimize the fixed structure parameters. At the same time, the throttle opening and acceleration combination under each working condition can dynamically adjust the discharge rate limit. For example, short-term high-rate discharge is allowed during the acceleration stage, while continuous high-power output is restricted during the climbing stage to protect the battery life. These labels also provide scenario-based inputs for the thermal management strategy. For example, the cooling system is started in advance during continuous acceleration or climbing to ensure that the battery operates in the optimal temperature range.

[0089] Build a dynamic equivalent model, divide the dynamic equivalent model into a basic model and an extended model, set the model trigger rules, and determine the detected current and battery temperature based on the model trigger rules to determine whether to trigger the extended mode or the basic mode.

[0090] The method for building the basic mode of the dynamic equivalent model includes:

[0091] When the lithium iron phosphate battery is in a stable state (such as after being open circuit for a period of time), apply a current pulse , measure the instantaneous change in the voltage at the battery terminal , and calculate the internal resistance of the battery according to Ohm's law ;

[0092] Conduct charge and discharge experiments on the lithium iron phosphate battery by applying rectangular current pulses until a stable voltage is reached without triggering the protection function of the lithium iron phosphate battery, and synchronously collect the voltage of the lithium iron phosphate battery.

[0093] Use a first-order RC equivalent circuit model to build the terminal voltage equation and the first-order dynamic equation of the polarization voltage.

[0094] Derive and fit the curves of the terminal voltage equations for the charging pulse stage and the discharging pulse stage respectively.

[0095] Aiming at minimizing the mean square error between the terminal voltage obtained from charge-discharge experiments and the theoretical terminal voltage, the first-order polarization resistance is iteratively updated using the non-linear least squares method and the first-order polarization capacitance values;

[0096] Connect in parallel with in series to obtain the basic mode of the dynamic equivalent model; where is the internal resistance of the lithium iron phosphate battery.

[0097] The method for building the extended mode of the dynamic equivalent model includes:

[0098] Apply two sets of rectangular pulses with different durations (such as 1s fast charge + 10s rest + 1s slow charge) to the lithium iron phosphate battery; synchronously collect the voltage and pulse current waveforms of the lithium iron phosphate battery;

[0099] Use a second-order RC equivalent circuit model to build the terminal voltage equation and the second-order dynamic equation of the polarization voltage;

[0100] Aiming at minimizing the mean square error between the terminal voltage obtained from charge-discharge experiments and the theoretical terminal voltage, use a nature-inspired optimization algorithm to optimize and obtain the second-order polarization resistance and the second-order polarization capacitance and the second-order polarization resistance and the second-order polarization capacitance values.

[0101] The method for determining the detection current and battery temperature based on the model trigger rule to determine whether to trigger the extended mode or the basic mode includes:

[0102] If the detection current and battery temperature meet the preset extended trigger conditions, trigger the extended mode of the dynamic equivalent model; otherwise, trigger the basic mode of the dynamic equivalent model.

[0103] The basic mode of the dynamic equivalent model constructs an equivalent circuit model under standard conditions by integrating real-time battery voltage, current, and temperature data, providing basic parameters for SOC estimation, internal resistance correction, and charge and discharge protection threshold setting. When the detection current (such as a short-term pulsed current exceeding the rated value) or battery temperature (such as a high-temperature warning threshold) triggers the expansion condition, the model switches to the expansion mode. By introducing vehicle condition segment tags (such as climbing, accelerating), vibration frequency compensation coefficients, and user load data, it dynamically adjusts parameters such as polarization internal resistance and diffusion impedance in the equivalent circuit. The expansion mode can accurately match the battery's dynamic response under extreme conditions. For example, it corrects the SOC estimation error during high-rate discharge and optimizes the thermal management trigger threshold in a high-temperature environment, thereby improving the BMS's prediction accuracy of the battery state, avoiding overcharge / overdischarge risks caused by lagging model parameters, and providing input variables closer to the actual usage scenario for the cycle life prediction algorithm, extending the overall service life of the battery pack.

[0104] Collect user data, extract the battery temperature from the internal data, and use voltage, current, battery temperature, external data, and user data as inputs to a bidirectional feature fusion network to obtain fused features. By inputting internal data such as voltage, current, and battery temperature and external / user data such as vehicle type, speed, and user load into the bidirectional feature fusion network, multi-dimensional information cross-correlation and deep integration can be achieved. Through a bidirectional information flow mechanism (such as the self-attention mechanism), this network can not only capture the dynamic mapping relationship between the internal state of the battery and external conditions (such as climbing, accelerating), but also reversely use user behavior characteristics (such as high-frequency charge and discharge) to optimize internal model parameters.

[0105] User data includes the daily driving mileage, daily typical operation time period, brake signal (0 / 1), and vehicle load. By quantifying user usage habits and requirements, user data promotes personalized calibration of BMS parameters and can be obtained through statistics. The daily driving mileage and load directly affect the battery capacity configuration and the low SOC limit setting to ensure a balance between endurance and safety. The daily typical operation time period data (such as morning and evening rush hours) can be combined with temperature changes to optimize the thermal management strategy. The brake signal (0 / 1) helps the BMS identify braking energy recovery opportunities and improve energy utilization efficiency. The load information adjusts the correction coefficient of the SOC estimation model by affecting the equivalent internal resistance of the battery pack. The introduction of user data makes the BMS parameters more suitable for the actual usage scenario, extending the battery life and enhancing the user experience.

[0106] The methods for obtaining the daily driving mileage and the daily typical operation time period include:

[0107] Statistically calculate the total driving mileage for each day within a preset period, and calculate the ratio of the sum of the total driving mileage for each day to the preset period to obtain the daily driving mileage;

[0108] Statistically count all the operation time periods with operation frequencies exceeding a preset frequency (such as 5 times) as the daily typical operation time periods.

[0109] The training method of the bidirectional feature fusion network includes:

[0110] Pre-collect K groups of training data. The training data includes input data and fusion features. The input data includes voltage, current, battery temperature, external data, and user data.

[0111] Take the input data as the input of the bidirectional feature fusion network, and take the fusion feature as the output of the bidirectional feature fusion network. With the goal of minimizing the error between the output fusion feature and the actual fusion feature, optimize the network parameters of the bidirectional feature fusion network through a nature-inspired optimization algorithm, obtain the network parameters corresponding to minimizing the error between the fusion feature output by the bidirectional feature fusion network and the actual fusion feature, and construct the bidirectional feature fusion network with the corresponding network parameters as the trained bidirectional feature fusion network; among them, the bidirectional feature fusion network includes a battery state branch network, a scenario demand branch network, and an attention fusion layer. Take voltage, current, and battery temperature as the input of the battery state branch network to obtain local time-domain features. Take external data and user data as the input of the scenario demand branch network to obtain scenario semantic features. Take the local time-domain features and scenario semantic features as the input of the attention fusion layer to obtain fusion features; the training methods of the battery state branch network and the scenario demand branch network are similar to the training method of the bidirectional feature fusion network; the battery state branch network is a CNN network, and the scenario demand branch network is a GRU network.

[0112] Extract the vibration frequency from the external data, and perform feature extraction on the voltage, current, and vibration frequency to obtain the voltage change rate, current pulse characteristics, and vibration intensity; by extracting the voltage change rate, current pulse characteristics, and vibration intensity, the internal dynamic response of the battery and external environmental interference can be accurately captured.

[0113] The methods for obtaining the voltage change rate, current pulse characteristics, and vibration intensity include:

[0114] Calculate the ratio of the voltage change to the corresponding time interval, and take the ratio as the voltage change rate;

[0115] If it is detected that the current at present is greater than the preset minimum amplitude threshold, the current at the previous moment is not greater than the preset minimum amplitude threshold, and the current duration is not less than the preset minimum duration threshold, then it is judged that the current pulse starts; if it is detected that the current is not greater than the preset minimum amplitude threshold, the current at the previous moment is greater than the preset minimum amplitude threshold, and the current duration is not less than the preset minimum duration threshold, then it is judged that the current pulse ends, and record the start time, end time, and maximum amplitude of the current pulse.

[0116] The pulse density is obtained by calculating the ratio of the number of pulses per unit time to the difference between the start time and the end time of the corresponding current pulse; if the pulse density ; where is the number of pulses; is the start time of the current pulse; is the end time of the current pulse;

[0117] The pulse energy is obtained by calculating the integral of the current during the start time and the end time of the current pulse; if the pulse energy ; where is the current pulse at time

[0118] The rising edge slope is obtained by calculating the ratio of the current difference in the current rising stage to the difference between the start time and the end time of the current pulse; if the rising edge slope ; where is the current corresponding to the start time of the current pulse; is the current corresponding to the end time of the current pulse;

[0119] The pulse density, pulse energy and rising edge slope are concatenated to obtain the current pulse feature;

[0120] The vibration intensity is obtained by calculating the root mean square of the acceleration.

[0121] The fusion feature, voltage change rate and current pulse feature are used as the input of the error prediction model to obtain the SOC estimation error; by inputting the fusion feature, voltage change rate and current pulse feature into the error prediction model, the deviation between the SOC estimated value and the actual value can be accurately quantified. This error feedback signal dynamically adjusts key parameters such as ohmic internal resistance and polarization internal resistance in the equivalent circuit model through the backpropagation mechanism. For example, when detecting an underestimation of SOC caused by high-frequency current pulses, the discharge rate compensation coefficient is optimized; at the same time, combining features such as vibration intensity, the aging factor in the temperature compensation algorithm is corrected to reduce capacity misjudgment in high-temperature or vibration environments. This error-driven parameter calibration strategy enables the BMS to compensate for model errors in real time, improve the SOC estimation accuracy, and provide a closed-loop optimization signal for the reinforcement learning framework to continuously iterate control parameters such as charge and discharge cut-off voltages and thermal management trigger thresholds, ultimately achieving the dual goals of extending battery life and improving safety.

[0122] The training method of the error prediction model includes:

[0123] Pre-collect D groups of error training data, where the error training data includes fusion features, voltage change rates, current pulse features, and SOC estimation errors.

[0124] Taking the fusion feature, voltage change rate, and current pulse feature as the inputs of the error prediction model, and the SOC estimation error as the output of the error prediction model, with the goal of minimizing the error between the output SOC estimation error and the actual SOC estimation error, the network parameters of the error prediction model are optimized through a nature-inspired optimization algorithm to obtain the network parameters corresponding to the minimum error between the SOC estimation error output by the error prediction model and the actual SOC estimation error, and the error prediction model constructed with the corresponding network parameters is used as the trained error prediction model.

[0125] Taking external data, user data, and consecutive working condition segment labels during the vehicle operation as the inputs of the population classification model to obtain population labels (for example, for two-wheel vehicles, they can be divided into urban commuters, express / delivery riders, casual cyclists, and mountain biking enthusiasts; for three-wheel vehicles, they can be divided into freight drivers, rural transporters, scenic shuttle drivers, and logistics handlers, etc.); by inputting external data such as vehicle type, speed, throttle opening, user data such as daily driving mileage, load, and consecutive working condition segment labels (such as start, climb, bump) into the population classification model, user group characteristics (such as high-frequency charging and discharging of express vehicles, smooth driving of household vehicles) and vehicle hardware characteristics (such as heat dissipation limitations of two-wheel vehicles) can be identified. The population label is the core output of the user portrait, providing a differentiated basis for BMS parameter calibration: for example, optimizing parameters related to cycle life (such as the charging cut-off voltage threshold) for the high-frequency charging and discharging population, adjusting the thermal management trigger temperature for vehicles operating at high temperatures, and correcting the mechanical stress compensation coefficient for vibration-sensitive vehicles (such as three-wheel vehicles). This model also reduces redundant features through clustering analysis, improves the algorithm efficiency, enables the reinforcement learning framework to more accurately match the population needs, dynamically adjusts the SOC estimation error compensation strategy and equivalent circuit model parameters, and finally realizes the scenario-based and personalized optimization of the battery full life cycle management.

[0126] The training method of the population classification model includes:

[0127] Pre-collecting E groups of classification training data, where the classification training data includes external data, user data, consecutive working condition segment labels, and the corresponding population labels.

[0128] Taking external data, user data, and consecutive working condition segment labels during the vehicle operation as the inputs of the population classification model, and the population label as the output of the population classification model, with the goal of minimizing the error between the output population label and the actual population label, the network parameters of the population classification model are optimized through a nature-inspired optimization algorithm to obtain the network parameters corresponding to the minimum error between the population label output by the population classification model and the actual population label, and the population classification model constructed with the corresponding network parameters is used as the trained population classification model.

[0129] Adjust the BMS parameters based on reinforcement learning combined with a dynamic equivalent model, SOC estimation error, and population labels;

[0130] Refer to Figure 2 , the method for adjusting the BMS parameters includes:

[0131] Define the state space : including SOC estimation error, battery temperature, instantaneous current overlimit flag, and vibration intensity; among them, the instantaneous current overlimit flag is obtained by comparing the current with a preset current overlimit threshold, and when the current exceeds the preset current overlimit threshold, the value is 1, otherwise it is 0;

[0132] Define the action space : including the adjustment amount of BMS parameters and the fan start / stop action, such as the adjustment amount of the battery internal resistance and polarization resistance in the dynamic equivalent model; action constraint: the change range of the battery internal resistance does not exceed the preset nominal value; the change step of the polarization resistance is greater than the change step of the battery internal resistance;

[0133] Set the reward function based on safety rewards, accuracy penalties, parameter voltage stability penalties, overheat penalties, polarization suppression penalties, and energy efficiency penalties , such as the reward function is:

[0134] ;

[0135] Among them:

[0136] ;

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] Among them, is the accuracy penalty weight; is the SOC estimation error; is the safety reward weight; is the current safety flag; is the current signal value; is the preset current safety threshold; is the parameter voltage stability penalty weight; is the voltage change amount; is the overheat penalty weight; is the battery temperature overheat flag; is the battery temperature; is the polarization suppression penalty weight; is the polarization voltage at time is the base mode current of the dynamic equivalent model; is the current with respect to time derivative; is the base mode of the dynamic equivalent model; is the second-order polarization resistance of the extended mode of the dynamic equivalent model in the branch where it is located; is the current with respect to time derivative; is the second-order polarization resistance of the extended mode of the dynamic equivalent model in the branch where it is located; is the current with respect to time derivative; is the extended mode of the dynamic equivalent model; is the energy efficiency penalty weight; is the energy efficiency; is the start time of the lithium iron phosphate battery operation; is the battery terminal voltage at time is the open circuit voltage at time is the charge and discharge current at time is the operation time of the lithium iron phosphate battery; The weights in the reward function are obtained by optimizing through a nature-inspired optimization algorithm.

[0142] Use the Q-learning algorithm combined with the exploration rate to learn the optimal policy and update the Q-value function: Initialize the Q-value function as a table, where the rows of the table correspond to different environmental states and the columns correspond to different actions , set the learning rate , discount factor and exploration rate ; At each time step , detect the current environmental state of the lithium iron phosphate battery , according to the exploration rate , with probability randomly select an action , with probability select the action with the maximum Q-value in the current environmental state ; Execute the selected action to obtain a new environmental state , meanwhile, the reward function value is calculated according to the reward function; the Q-value function is updated according to the update formula:

[0143] ;

[0144] where is the updated Q-value; is the environmental state when taking the action of the Q-value; is the learning rate; is the reward function value; is the discount factor; is the next environmental state the optimal adjustment action among all possible adjustment actions; is the next environmental state all possible adjustment actions the maximum value of the corresponding Q-value among them, that is, after transferring from the current environmental state to the next environmental state, the maximum expected cumulative reward that can be obtained among all optional adjustment actions.

[0145] Repeat updating the Q-value function until the Q-value converges, obtain the corresponding optimal Q-value function, and select and execute the optimal action from the action space according to the environmental state.

[0146] Through the reinforcement learning framework, the real-time operating condition parameters of the dynamic equivalent model (such as the dynamic correction of the polarization internal resistance during high-rate discharge), the SOC estimation error feedback (such as the deviation between the model prediction value and the measured value), and the population label (such as the clustering characteristics of different vehicle types and user driving habits) are combined to realize the adaptive optimization of the BMS parameters. The reinforcement learning algorithm aims to minimize the battery health degradation rate and maximize the energy utilization rate, etc., and dynamically adjusts parameters such as the charge and discharge cut-off voltage threshold, the heat management trigger temperature, and the SOC correction coefficient. For example, for user groups with frequent rapid acceleration, the algorithm can optimize the equivalent circuit model parameters during high-rate discharge and reduce the SOC estimation error; for vehicles operating in high-temperature environments, the heat dissipation strategy can be triggered in advance and the internal resistance compensation coefficient can be adjusted. This method breaks through the limitations of traditional fixed-parameter models, enabling the BMS to continuously optimize the battery operation strategy under complex operating conditions and different user requirements, extend the cycle life, and improve safety.

[0147] Embodiment 2

[0148] Please refer to Figure 3 as shown, this embodiment provides a method for embedding a virtual thermal topology barrier algorithm in the BMS firmware to dynamically adjust the charge and discharge priorities of adjacent battery cells when a thermal diffusion path is detected, including:

[0149] Monitor the change of the cell temperature field in real time through a distributed temperature sensor;

[0150] Combine the finite element method to predict the direction of the thermal diffusion path;

[0151] When the local temperature gradient exceeds the threshold, construct a virtual barrier between adjacent cells on the thermal runaway propagation path.

[0152] The method for constructing the virtual barrier includes:

[0153] Reduce the charging current weight of the cells downstream of the thermal diffusion and increase the discharging current weight of the cells upstream. Achieve precise thermal management through the weighted current distribution model, and at the same time trigger active cooling in combination with the three-level early warning mechanism.

[0154] Embodiment 3

[0155] Please refer to Figure 4 As shown, this embodiment provides a battery BMS parameter calibration device, including:

[0156] The first analysis module: Collect the internal data of the lithium iron phosphate battery, extract the voltage signal and current signal from the internal data, and set an acceleration-ripple coupling model for adaptive filtering to obtain the voltage and current;

[0157] The second analysis module: Collect the external data of the battery, divide the vehicle running state into different working condition segments based on the external data, and form working condition segment labels;

[0158] The model building module: Build a dynamic equivalent model, divide the dynamic equivalent model into a basic mode and an extended mode, set the model trigger conditions, and determine the detected current and battery temperature based on the model trigger rules to determine whether to trigger the extended mode or the basic mode;

[0159] The data fusion module: Collect user data, extract the battery temperature from the internal data, and use the voltage, current, battery temperature, external data, and user data as the input of the bidirectional feature fusion network to obtain fusion features;

[0160] The feature extraction module: Extract the vibration frequency from the external data, and perform feature extraction on the voltage, current, and vibration frequency to obtain the voltage change rate, current pulse characteristics, and vibration intensity;

[0161] The error estimation module: Use the fusion features, voltage change rate, and current pulse characteristics as the input of the error prediction model to obtain the SOC estimation error;

[0162] The population classification module: Use the external data, user data, and continuous working condition segment labels during the vehicle operation as the input of the population classification model to obtain population labels;

[0163] Parameter calibration module: Adjust the BMS parameters based on reinforcement learning combined with a dynamic equivalent model, vibration intensity, SOC estimation error, and population labels.

[0164] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the said claims.

[0165] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall all be included within the protection scope of the present invention.

Claims

1. Battery BMS parameter calibration method, characterized in that It includes the following steps: Collect the internal data of the lithium iron phosphate battery, extract the voltage signal and current signal from the internal data, and set an acceleration-ripple coupling model for adaptive filtering to obtain the voltage and current; Collect the external data of the battery, divide the operating state of the vehicle into different working condition segments based on the external data, and form working condition segment labels; Build a dynamic equivalent model, divide the dynamic equivalent model into a basic model and an extended model, set model trigger rules, and determine the detected current and battery temperature based on the model trigger rules to determine whether to trigger the extended mode or the basic mode; Collect user data, extract the battery temperature from the internal data, and use the voltage, current, battery temperature, external data, and user data as the input of a bidirectional feature fusion network to obtain fusion features; Extract the vibration frequency from the external data, perform feature extraction on the voltage, current, and vibration frequency to obtain the voltage change rate, current pulse characteristics, and vibration intensity; Use the fusion features, voltage change rate, and current pulse characteristics as the input of an error prediction model to obtain the SOC estimation error; Use external data, user data, and continuous working condition segment labels during the vehicle operation as the input of the population classification model to obtain population labels; Based on reinforcement learning, combine the dynamic equivalent model, vibration intensity, SOC estimation error, and population label to adjust the BMS parameters.

2. The battery BMS parameter calibration method according to claim 1, characterized in that The method for adjusting the BMS parameters includes: Define the state space : including the SOC estimation error, battery temperature, instantaneous current over-limit flag, and vibration intensity; among them, the instantaneous current over-limit flag is obtained by comparing the current with a preset current over-limit threshold. When the current exceeds the preset current over-limit threshold, the value is 1; otherwise, it is 0. Define the action space : including the adjustment amount of BMS parameters and the fan start / stop action; Set the reward function based on safety rewards, precision penalties, parameter voltage stability penalties, overheating penalties, polarization suppression penalties, and energy efficiency penalties ; Learn the optimal policy using the Q-learning algorithm, initialize the Q-value function as a table, where the rows of the table correspond to different environmental states , and the columns correspond to different actions , set the learning rate , discount factor and exploration rate ; at each time step , detect the current environmental state of the lithium iron phosphate battery , according to the exploration rate , with probability randomly select an action , with probability select the action with the maximum Q-value in the current environmental state ; execute the selected action to obtain a new environmental state , meanwhile, calculate the reward function value according to the reward function; update the Q-value function according to the update formula; Repeatedly update the Q-value function until the Q-value converges to obtain the corresponding optimal Q-value function. According to the environmental state, use the optimal Q-value function to select and execute the optimal action from the action space.

3. The battery BMS parameter calibration method according to claim 1, wherein The working condition segments include start, stop, acceleration, deceleration, constant speed, climbing, downhill, turning, and bumping; the methods for obtaining the working condition segment labels include: Extract the throttle opening, steering angle, acceleration, pitch angle, roll angle, and vibration frequency from the external data of the battery, and detect whether the throttle opening increases from 0 to greater than the preset opening threshold and whether the acceleration increases from 0 to greater than the preset acceleration threshold; if so, it is classified as a start and marked with a start label; detect whether the throttle opening drops to 0 and whether the acceleration remains less than 0 until the speed drops to 0; if so, it is classified as a stop and marked with a stop label; calculate the derivative of the throttle opening and detect whether the derivative of the throttle opening is greater than the preset acceleration derivative threshold and whether the acceleration is greater than the preset acceleration threshold; if so, it is classified as acceleration and marked with an acceleration label; detect whether the derivative of the throttle opening is less than the preset deceleration derivative threshold and whether the acceleration is less than 0; if so, it is classified as deceleration and marked with a deceleration label; Detect whether the change value of the throttle opening is less than the preset fluctuation threshold, whether the change value of the acceleration is less than the preset acceleration change threshold, and whether the duration is greater than the preset constant speed time threshold; if so, it is classified as constant speed and marked with a constant speed label; Detect whether the pitch angle is greater than the preset pitch angle climbing threshold and whether the throttle opening is greater than the preset climbing throttle opening threshold; if so, it is classified as climbing and marked with a climbing label; detect whether the pitch angle is less than the preset pitch angle downhill threshold and whether the throttle opening is less than the preset downhill throttle opening threshold; if so, it is classified as downhill and marked with a downhill label; Detect whether the steering angle is greater than the preset angle threshold, whether the roll angle is greater than the preset roll angle threshold, and whether the duration is greater than the preset turning time threshold; if so, classify it as a turn and mark it with a turn label; detect whether the vibration frequency is greater than the preset vibration frequency threshold, whether the acceleration standard deviation is greater than the preset standard deviation threshold, and whether the duration is greater than the preset bump time threshold; if so, classify it as a bump and mark it with a bump label.

4. The battery BMS parameter calibration method according to claim 1, wherein, The methods for obtaining voltage and current include: Adaptive filtering is respectively performed on the voltage signal and the current signal by combining the compensation coefficient, the acceleration of the vehicle operation, and the ripple frequency to obtain the corresponding voltage and current.

5. The battery BMS parameter calibration method according to claim 1, wherein The methods for building the basic mode of the dynamic equivalent model include: When the lithium iron phosphate battery is in a stable state, apply a current pulse, measure the instantaneous change in the voltage at the battery terminal, and calculate the value of the battery internal resistance according to Ohm's law; Perform charge and discharge experiments on the lithium iron phosphate battery by applying rectangular current pulses until a stable voltage is reached and the protection function of the lithium iron phosphate battery is not triggered, and synchronously collect the voltage of the lithium iron phosphate battery; Use a first-order RC equivalent circuit model to build the terminal voltage equation and the first-order dynamic equation of the polarization voltage; Derive and fit the curves for the terminal voltage equations in the charging pulse stage and the discharging pulse stage respectively; Aiming at minimizing the mean square error between the terminal voltage obtained from the charge-discharge experiment and the theoretical terminal voltage, the first-order polarization resistance is iteratively updated using the nonlinear least squares method and the first-order polarization capacitance values; Connect in parallel with a series circuit to obtain the basic mode of the dynamic equivalent model; among them, is the internal resistance of the lithium iron phosphate battery.

6. The battery BMS parameter calibration method according to claim 1, wherein The methods for building the extended mode of the dynamic equivalent model include: Apply two sets of rectangular current pulses with different durations to the lithium iron phosphate battery; synchronously collect the voltage and the pulse current waveform of the lithium iron phosphate battery; Use a second-order RC equivalent circuit model to build the terminal voltage equation and the second-order dynamic equation of the polarization voltage; Aiming at minimizing the mean square error between the terminal voltage obtained from charge-discharge experiments and the theoretical terminal voltage, the natural inspiration optimization algorithm is used to optimize and obtain the second-order polarization resistance , the second-order polarization capacitance , the second-order polarization resistance , the second-order polarization capacitance values.

7. The battery BMS parameter calibration method according to claim 1, wherein The method for determining to trigger the extended mode or the basic mode by judging the detected current and the battery temperature based on the model trigger rule includes: If the detected current and the battery temperature meet the preset extended trigger conditions, trigger the extended mode of the dynamic equivalent model; otherwise, trigger the basic mode of the dynamic equivalent model.

8. The battery BMS parameter calibration method according to claim 1, characterized in that, The training method of the two-way feature fusion network includes: Pre-collect K sets of training data, where the training data includes input data and fusion features, and the input data includes voltage, current, battery temperature, external data, and user data; Use the input data as the input of the two-way feature fusion network and the fusion feature as the output of the two-way feature fusion network. With the goal of minimizing the error between the output fusion feature and the actual fusion feature, optimize the network parameters of the two-way feature fusion network through a natural inspiration optimization algorithm, obtain the network parameters corresponding to minimizing the error between the fusion feature output by the two-way feature fusion network and the actual fusion feature, and use the two-way feature fusion network constructed with the corresponding network parameters as the trained two-way feature fusion network.

9. The battery BMS parameter calibration method according to claim 1, wherein The methods for obtaining the voltage change rate, the current pulse feature, and the vibration intensity include: Calculate the ratio of the voltage change to the corresponding time interval, and use the ratio as the voltage change rate; If it is detected that the current is greater than the preset minimum amplitude threshold, the current at the previous moment is not greater than the preset minimum amplitude threshold, and the current duration is not less than the preset minimum duration threshold, then it is determined that the current pulse starts; if it is detected that the current is not greater than the preset minimum amplitude threshold, the current at the previous moment is greater than the preset minimum amplitude threshold, and the current duration is not less than the preset minimum duration threshold, then it is determined that the current pulse ends, and the start time, end time, and maximum amplitude of the current pulse are recorded; Calculate the ratio of the number of pulses per unit time to the difference between the start time and end time of the corresponding current pulse to obtain the pulse density; Calculate the integral of the current during the start time and end time of the current pulse to obtain the pulse energy; Calculate the ratio of the current difference in the current rising stage to the difference between the start time and end time of the current pulse to obtain the rising edge slope; Concatenate the pulse density, pulse energy, and rising edge slope to obtain the current pulse characteristics; Calculate the root mean square of the acceleration to obtain the vibration intensity.

10. The battery BMS parameter calibration method according to claim 3, characterized in that, The method for obtaining the throttle opening includes: Obtain the throttle rotation angle range of the vehicle throttle grip ; where is the maximum throttle rotation angle; Quantize the throttle rotation angle range to a preset quantization range to obtain the quantized throttle opening value; Use the quantized throttle opening value as the throttle opening 11. A battery BMS parameter calibration device, which implements the battery BMS parameter calibration method according to any one of claims 1-10, characterized in that, including: The first analysis module: Collect the internal data of the lithium iron phosphate battery, extract the voltage signal and current signal from the internal data, and set up an acceleration-ripple coupling model for adaptive filtering to obtain the voltage and current; The second analysis module: Collect the external data of the battery, divide the vehicle operating state into different working condition segments based on the external data, and form working condition segment labels; The model building module: Build a dynamic equivalent model, divide the dynamic equivalent model into a basic model and an extended model, set model trigger rules, and determine the detected current and battery temperature based on the model trigger rules to determine whether to trigger the extended mode or the basic mode; The data fusion module: Collect user data, extract the battery temperature from the internal data, and use the voltage, current, battery temperature, external data, and user data as the input of the bidirectional feature fusion network to obtain the fusion features; The feature extraction module: Extract the vibration frequency from the external data, and perform feature extraction on the voltage, current, and vibration frequency to obtain the voltage change rate, current pulse characteristics, and vibration intensity; The error estimation module: Use the fusion features, voltage change rate, and current pulse characteristics as the input of the error prediction model to obtain the SOC estimation error; Group classification module: Using external data, user data, and continuous working condition segment labels during vehicle operation as the input of the group classification model to obtain group labels; The parameter calibration module: Adjust the BMS parameters based on reinforcement learning combined with the dynamic equivalent model, vibration intensity, SOC estimation error, and population labels.

Images