Lithium iron phosphate battery BMS parameter calibration method and device based on state estimation

Abstract

The invention relates to a lithium iron phosphate battery BMS parameter calibration method and device based on state estimation, and the monitoring method comprises the steps: obtaining real-time working data and historical working data, and carrying out the estimation based on the real-time working data and the historical working data, and obtaining an estimated electric quantity state and an estimated health state; training a prediction model I and a prediction model II for predicting the electric quantity state and the health state based on the historical working data; performing dynamic correction based on the estimated electric quantity state and the estimated health state to obtain a corrected electric quantity state and a corrected health state; and respectively calculating a deviation value I of the predicted electric quantity state and the corrected electric quantity state and a deviation value II of the predicted health state and the corrected health state, respectively comparing the deviation value I and the deviation value II with a preset threshold value of the deviation value I and a preset threshold value of the deviation value II, and judging whether a calibration instruction is generated or not. By dynamically calibrating the BMS parameters, the calibration accuracy of the electric quantity state and the health state during riding is improved.

Description

Technical Field

[0001] The present invention relates to the field of battery management systems, and particularly to a method and device for calibrating BMS parameters of lithium iron phosphate batteries based on state estimation. Background Art

[0002] The statements herein only provide background information related to the present invention and do not necessarily constitute prior art.

[0003] As a common type of lithium battery, lithium iron phosphate batteries are widely used in electric vehicles, energy storage systems, and other electronic devices. Due to their high safety, long life, and good temperature stability, lithium iron phosphate batteries have become a core component in modern battery management systems (BMS). The BMS is used to monitor the working state of the battery to ensure its safe and effective operation, mainly including the estimation and management of the state of charge and state of health of the battery.

[0004] For example, a BMS voltage sampling self-calibration method and system disclosed in the publication number: CN118362893B predicts correction coefficients through model training. However, in actual applications of lithium iron phosphate batteries, especially in two-wheeled or three-wheeled electric vehicles, due to the actual working environment of two-wheeled or three-wheeled electric vehicles, there are many uncertainties, such as load, temperature, etc., which affect the data. Using a single estimation method will cause the error of the state of charge and state of health to become more and more serious, affecting the performance and safety of lithium iron phosphate batteries.

[0005] Therefore, the present invention proposes a method and device for calibrating BMS parameters of lithium iron phosphate batteries based on state estimation. Summary of the Invention

[0006] The purpose of the present invention is to provide a new method and device for calibrating BMS parameters of lithium iron phosphate batteries based on state estimation, so as to achieve dynamic calibration of BMS parameters and improve the calibration accuracy of the state of charge and state of health during riding.

[0007] The purpose of the present invention is achieved by the following technical solutions. According to the method for calibrating BMS parameters of lithium iron phosphate batteries based on state estimation proposed in the present disclosure, the monitoring method includes:

[0008] Obtain real-time working data and historical working data;

[0009] Estimate the estimated state of charge and estimated state of health based on the real-time working data and historical working data;

[0010] Train prediction models one and two for predicting the state of charge and state of health based on the historical working data;

[0011] Obtain the corrected state of charge and the corrected state of health through dynamic correction based on the estimated state of charge and the estimated state of health;

[0012] Calculate the deviation value one between the predicted state of charge and the corrected state of charge and the deviation value two between the predicted state of health and the corrected state of health respectively, and compare the deviation value one and the deviation value two with the preset deviation value one threshold and the preset deviation value two threshold respectively to determine whether to generate a calibration instruction.

[0013] Preferably, obtain the working state change nodes of the lithium iron phosphate battery. The working state change nodes include node one and node two. Node one is from the moving state to the stationary state, and node two is from the stationary state to the moving state.

[0014] Preferably, record the node data after monitoring node one. The node data is the state of charge, the total driving time, the cumulative discharge capacity, and the riding load of the lithium iron phosphate battery before the node change each time the node changes.

[0015] Preferably, the estimation method of the estimated state of charge is as follows:

[0016] Initialize the state of charge, collect the charge and discharge current values of the lithium iron phosphate battery at each moment, and obtain the basic state of charge through integral formula calculation based on the current value and the collection time;

[0017] Preset the temperature correction coefficient and the load correction coefficient, and obtain the temperature-corrected state of charge and the load-corrected state of charge through integral formula calculation based on the temperature correction coefficient and the load correction coefficient;

[0018] Obtain the estimated state of charge through integral formula calculation based on the temperature-corrected state of charge and the load-corrected state of charge.

[0019] Preferably, the estimation method of the estimated state of health is as follows:

[0020] Correct the internal resistance based on the preset temperature correction coefficient;

[0021] Obtain the estimated state of health through integral formula calculation based on the corrected internal resistance.

[0022] Preferably, the dynamic correction method of the corrected state of charge is as follows:

[0023] Based on the estimated state of charge, randomly generate n initial particles, where n is an integer greater than 1, and assign corresponding weight correction factors to the n initial particles based on the node data;

[0024] Pre-construct the charge and discharge model of the battery, predict the n initial particles, and obtain the state of charge at the next moment of n';

[0025] Based on the difference between the estimated state of charge and the state of charge at the next moment, update the weights of n initial particles and resample, that is, retain the initial particles with larger weights until the updated weights reach the preset weight size and then stop the update, obtain n″ updated particles, and take the average of the obtained n″ updated particles as the corrected state of charge.

[0026] Preferably, the dynamic correction method of the corrected health state is the same as the dynamic correction method of the corrected state of charge.

[0027] Preferably, the method for determining whether to generate a calibration instruction includes:

[0028] The calibration instruction includes calibration instruction one and calibration instruction two;

[0029] Compare the first deviation value and the second deviation value with the preset first deviation value threshold and the preset second deviation value threshold respectively;

[0030] If the first deviation value is greater than the preset first deviation value threshold, generate calibration instruction one and calibrate the state of charge of the BMS based on calibration instruction one;

[0031] If the second deviation value is greater than the preset second deviation value threshold, generate calibration instruction two and calibrate the health state of the BMS based on calibration instruction two;

[0032] If the first deviation value is less than or equal to the preset first deviation value threshold, no calibration instruction is generated;

[0033] If the second deviation value is less than or equal to the preset second deviation value threshold, no calibration instruction is generated.

[0034] Preferably, the training method of the first prediction model is:

[0035] Collect multiple sets of historical working data and the corresponding true state of charge of the historical working data, and convert the corresponding true state of charge of the historical working data into a first feature vector;

[0036] Divide multiple sets of historical working data and the corresponding true state of charge of the historical working data into a training set, a validation set and a test set, and perform normalization processing on the data. Use the first feature vector as the input of the first machine learning model, use the training set to train the first machine learning model, and at the same time monitor the performance of the first machine learning model through the validation set, adjust the hyperparameters to optimize the prediction effect. During the training process, with the goal of minimizing the prediction error, when the error on the validation set converges and the performance of the first machine learning model meets the requirements, stop the training, and obtain the first machine learning model that can accurately predict the state of charge as the first prediction model;

[0037] The first machine learning model is a decision tree or a neural network, etc.

[0038] Preferably, the training method of the second prediction model is as follows:

[0039] Collect multiple sets of historical working data and the corresponding true health status of the historical working data, and convert the corresponding true health status of the historical working data into a second feature vector;

[0040] Divide multiple sets of historical working data and the corresponding true power status of the historical working data into a training set, a validation set, and a test set, and perform normalization processing on the data. Use the second feature vector as the input of the second machine learning model, and use the training set to train the second machine learning model. At the same time, monitor the performance of the second machine learning model through the validation set, adjust the hyperparameters to optimize the prediction effect. During the training process, with the goal of minimizing the prediction error, when the error on the validation set converges and the performance of the second machine learning model meets the requirements, stop training, and obtain the second machine learning model that can accurately predict the health status as the second prediction model;

[0041] The second machine learning model is a decision tree or a neural network, etc.

[0042] A lithium iron phosphate battery BMS parameter calibration device based on state estimation is applied to the above-mentioned lithium iron phosphate battery BMS parameter calibration method based on state estimation. The device includes:

[0043] An acquisition module, configured to acquire real-time working data and continuously record it as historical working data;

[0044] A state estimation unit, configured to estimate and obtain an estimated power status and an estimated health status based on the real-time working data and the historical working data;

[0045] A model training module, configured to train a first prediction model and a second prediction model for predicting the power status and the health status based on the historical working data, and input the real-time working data into the first prediction model and the second prediction model respectively to obtain the predicted power status and the predicted health status;

[0046] A data fusion unit, configured to dynamically correct and obtain a corrected power status and a corrected health status based on the estimated power status and the estimated health status;

[0047] A calibration module, configured to calculate a first deviation value between the predicted power status and the corrected power status and a second deviation value between the predicted health status and the corrected health status, and compare the first deviation value and the second deviation value with a preset first deviation value threshold and a preset second deviation value threshold respectively to determine whether to generate a calibration instruction.

[0048] From the above technical solutions, it can be seen that the present application has the following beneficial effects:

[0049] 1: By combining real-time working data and historical data, using machine learning prediction models and dynamic correction mechanisms, the estimation accuracy of the state of charge and state of health is optimized. Through temperature correction and load correction factors, the estimation errors caused by temperature, load, and battery degradation are effectively addressed. The deviation value comparison mechanism avoids frequent calibration, ensuring the accuracy and efficiency of calibration.

[0050] 2: Record key data when node changes. Through the dynamic correction mechanism, reduce the fluctuations in the power display, improve the user's perception of the power display, adapt to complex working conditions and long-term battery aging, enhance the long-term stability of the battery management system, while extending the battery life and strengthening the system's prediction ability.

[0051] The above description is only an overview of the technical solution of the present invention. In order to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given, and in conjunction with the drawings, the details are described as follows. Brief Description of the Drawings

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0053] Figure 1 It is a schematic flowchart of a method for calibrating parameters of a lithium iron phosphate battery BMS based on state estimation according to an embodiment of the present invention;

[0054] Figure 2 It is a schematic flowchart of a method for calibrating parameters of a lithium iron phosphate battery BMS based on state estimation according to another embodiment of the present invention;

[0055] Figure 3 It is a schematic flowchart of a device for calibrating parameters of a lithium iron phosphate battery BMS based on state estimation according to an embodiment of the present invention. Detailed Embodiments

[0056] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following, in conjunction with the drawings and preferred embodiments, details the specific embodiments, structures, features, and effects of the third-party system monitoring system, method, device, equipment, and storage medium proposed according to the present invention.

[0057] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Additionally, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device that includes a series of elements includes not only those elements but also other elements not expressly listed, or elements that are inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including..." does not exclude the presence of additional identical elements in the process, method, article, or device that includes the said element.

[0058] I. Refer to Figure 1 As shown, a parameter calibration method for a lithium iron phosphate battery BMS based on state estimation, the calibration method includes:

[0059] Initialize the system and activate the BMS, monitor the working state of the lithium iron phosphate battery in real time, obtain real-time working data, and collect and store the real-time working data to obtain historical working data. The working data includes the voltage, current, temperature, internal resistance, etc. of the lithium iron phosphate battery; Exemplarily, the lithium iron phosphate battery is applied to, for example, two-wheeled or three-wheeled electric vehicles. The working state of the lithium iron phosphate battery includes riding or charging. The data monitored during riding is mainly the discharge state of the lithium iron phosphate battery, and the data monitored during charging is mainly the lithium iron phosphate battery;

[0060] The voltage, current, temperature, and internal resistance of the lithium iron phosphate battery are obtained by sensors integrated on the lithium iron phosphate battery. For example, a voltage sensor is used to measure the voltage of the lithium iron phosphate battery. The voltage directly reflects the charging state of the lithium iron phosphate battery. The changing voltage can indicate the battery's power, load, and whether it is overcharged or over-discharged; The current is monitored in real time by a current sensor, such as a Hall sensor, to measure the magnitude of the current flowing through the lithium iron phosphate battery. The change in current reflects the charge and discharge process of the lithium iron phosphate battery. At different loads and charge and discharge rates, the magnitude of the current directly affects the power state and health state of the lithium iron phosphate battery; The temperature is monitored by a temperature sensor, such as an NTC thermistor, to detect the temperature changes inside and outside the battery. The temperature change of the battery has a significant impact on the charge and discharge performance of the battery. Excessive or too low temperature will cause the attenuation of the battery performance, affecting the accurate estimation of the power state and health state; The internal resistance is indirectly estimated by measuring the voltage change of the lithium iron phosphate battery during the charge and discharge process, or by calculating the internal resistance through periodic charge and discharge experiments. The internal resistance of the battery is closely related to the health state of the lithium iron phosphate battery. Abnormal internal resistance indicates battery degradation or a fault in the battery, resulting in low efficiency or accelerated loss;

[0061] Working data also includes charging rate, load and charging and discharging time. The charging and discharging rate refers to the charging and discharging speed of the battery, which directly affects the battery's temperature, life and voltage changes. Excessive charging and discharging rates can cause the battery to heat up and increase losses. Load refers to the load that the battery bears in actual use. By adding a load sensor to the battery load end, the battery load changes can be monitored in real time. The size of the load will affect the battery's current and voltage, and indirectly affect the estimation of the state of charge. The charging and discharging time records the time spent on the battery's charging and discharging process through a precise timer. The charging and discharging time, charging and discharging rate, load and battery capacity are related, and are key parameters for estimating the state of charge, health status and battery performance degradation.

[0062] The collection of historical working data includes the cumulative storage of real-time collected working data or the simulation of working data of lithium iron phosphate batteries under different loads, charge and discharge rates, temperatures and other working conditions in an experimental environment.

[0063] Specifically, the charge state and health state of the lithium iron phosphate battery are estimated based on the real-time working data, marked as SOC and SOH respectively, and a preliminary lithium iron phosphate battery performance model is constructed to estimate the SOC and SOH of the lithium iron phosphate battery through the existing working condition data.

[0064] The estimation methods of the state of charge SOC include:

[0065] SOC indicates the ratio of the current amount of electricity stored in the lithium iron phosphate battery to the rated capacity of the lithium iron phosphate battery. Initialize SOC, that is, when the lithium iron phosphate battery starts to be used, set an initial SOC = 100%. Collect the charge and discharge current value I(t) of the lithium iron phosphate battery at every moment, and use the integral calculation to calculate the change of SOC using the integral of current and time. The calculation formula of SOC is:

[0066]

[0067] Among them, SOC(t) is the SOC at the current moment, sec(t-1) is the SOC at the previous moment, C rated is the rated capacity of the lithium iron phosphate battery, I(t) is the current value at time t, and dt is the time interval. The SOC is calculated by integrating the current. The positive or negative current determines whether the lithium iron phosphate battery is charging or discharging. The current data needs to be accurately collected to ensure the accuracy of the SOC estimation;

[0068] The methods for estimating the state of health SOH include:

[0069] SOH is related to the capacity and internal resistance of the lithium iron phosphate battery. The internal resistance increases with the use time. Therefore, SOH can be estimated by monitoring the change of the internal resistance of the lithium iron phosphate battery. The estimation formula of SOH is:

[0070]

[0071] Wherein, SOH(t) is the state of health at time t, and R initial is the initial internal resistance of the lithium iron phosphate battery, which is measured when the lithium iron phosphate battery starts to be used. R(t) is the internal resistance at time t. The internal resistance changes with the aging and usage degree of the lithium iron phosphate battery. A higher internal resistance means a poorer health condition of the lithium iron phosphate battery and a lower SOH. Therefore, by monitoring the change of the internal resistance of the lithium iron phosphate battery, the state of health of the lithium iron phosphate battery can be estimated.

[0072] In some embodiments, considering the influence of factors such as temperature, load, and charge-discharge rate on SOC and SOC, for example, low temperature will increase the internal resistance of the lithium iron phosphate battery and reduce the discharge efficiency; while high temperature will accelerate the decline of the lithium iron phosphate battery. Therefore, it is necessary to correct SOC(t) and SOH(t). Specifically, for the temperature influence correction, temperature has a significant influence on SOC and SOH, and the estimation of SOC and SOH is adjusted. SOC c = SOC(t) × f c , where SOC c is the SOC after temperature influence correction, and f c is the temperature correction factor, and f c is fitted by those skilled in the art according to experimental data;

[0073] For the correction of the influence of load and charge-discharge rate, the load and charge-discharge rate will affect the estimation accuracy of SOC. The greater the load, the higher the current, and the faster the change speed of SOC. Therefore, it is necessary to consider the influence of load and charge-discharge rate on SOC. The load influence formula:

[0074] SOC FZ = SOC(t) × (1 + k load × L)

[0075] wherein, SOC FZ is the corrected SOC, k load is the load correction coefficient, and k load is fitted by those skilled in the art according to experimental data, and L is the load magnitude;

[0076] The formula for the relationship between internal resistance and temperature: R adj = R(T) × g(T), where R adj is the internal resistance after temperature correction, and g(T) is the temperature correction factor. g(T) is obtained according to the experimental data of the temperature characteristics of the lithium iron phosphate battery. The estimation formula for the corrected SOH is: where SOH adj is the corrected SOH.

[0077] The aim is to collect data of lithium iron phosphate batteries under different conditions such as different loads, charge and discharge rates, and temperatures, and combine the estimation formulas of SOC and SOH to construct a preliminary performance model of lithium iron phosphate batteries. The accuracy of this model depends on the precision of data collection and how these data are correlated with the electrochemical characteristics of lithium iron phosphate batteries, and it can more accurately predict the state of lithium iron phosphate batteries in actual use, providing effective support for BMS correction.

[0078] It is worth mentioning that in practical applications, the SOC and SOH of lithium iron phosphate batteries will continuously change with the usage time. Therefore, it is necessary to continuously update and optimize the model according to real-time data. It is possible to conduct deep charge and discharge tests on lithium iron phosphate batteries regularly, collect more historical data, and continuously adjust and optimize the parameters of the model (such as load correction coefficients, temperature correction factors, etc.).

[0079] Based on the dynamic correction of the estimated state of charge and the estimated state of health, the corrected state of charge and the corrected state of health are obtained. The dynamic correction method is based on particle filtering or Kalman filtering. In some embodiments, Kalman filtering is a recursive algorithm used to estimate the state of a dynamic system. Kalman filtering includes two main steps: the prediction step and the update step. In SOC estimation, multiple corrected SOC data will be incorporated into the filtering process as part of the measurement update, helping the system to more accurately estimate SOC or SOH at each time step;

[0080] Step 1: In the prediction step, Kalman filtering uses the system model (state transition matrix A and control input matrix B) to predict the state at the current moment. Assume that the predicted value of SOC at the current moment is obtained based on the SOC estimation value of the previous step and system dynamics such as current and load. The formula for the prediction step is as follows:

[0081]

[0082] is the predicted SOC at time k, A is the state transition matrix, is the known state (SOC estimation) at the previous time k - 1, u k is the control input, such as current or load. The state transition matrix A is a matrix that describes how the system state changes over time, and it defines the change relationship of the state from time k - 1 to time k;

[0083] The state transition matrix A is obtained through a battery model. For example, the state transition matrix is derived based on the electrochemical model, thermal model, etc. of the battery. For instance, SOC (state of charge of the battery) and SOH (state of health of the battery) are usually affected by factors such as current, temperature, and charge and discharge rate. The state transition matrix A is a matrix that describes how these effects change over time.

[0084] Step 2: Update. In the update step, the Kalman filter uses actual measurement data to adjust the predicted SOC value. Multiple corrected SOC values (such as SOC corrections based on temperature and load) are used as measurement inputs to provide corrections to the SOC, that is, the estimated state of charge and the estimated state of health are used as measurement inputs, and weighted averaging is performed in combination with the Kalman gain. The formula for the update step:

[0085]

[0086] x k|k is the updated SOC estimate at time k, that is, the corrected state of charge. is the predicted SOC at time k, K k is the Kalman gain, representing the system's confidence in the measurement result, z k is the current measurement value, that is, the weighted average of multiple corrected SOC values;

[0087] Fusing multiple SOC correction data: In the update step 1, when fusing multiple corrected SOC data, the weighted average method can be used, or the weights of each correction result can be dynamically adjusted according to the credibility of the measurement. Specifically, different corrections (temperature, load, etc.) can be given weights according to their influence on the battery state, so as to obtain a comprehensive measurement value z k .

[0088] z k = w C ·SOC C + w FZ ·SOC FZ +…

[0089] where w C and w FZ are weight coefficients, which are set based on factors such as the credibility and influence degree of each correction data. The Kalman gain K k will be automatically adjusted according to the noise level of each correction data.

[0090] It is worth mentioning that the calculation method of correcting the state of health is the same as that of correcting the state of charge, and will not be elaborated here.

[0091] In some embodiments, the corrected state of charge or the corrected state of health can also be modeled based on the particle filter method. Specifically, the dynamic correction method of the corrected state of charge is:

[0092] Based on the estimated state of charge, n initial particles are randomly generated, where n is an integer greater than 1, and corresponding weight correction factors are assigned to the n initial particles based on node data;

[0093] Pre-build the charge and discharge model of the battery, predict n initial particles, and obtain the power states of n' particles at the next moment. The prediction formula for each particle is as follows: Among them, is the predicted state of the nth particle, that is, the power state at the next moment.

[0094] Based on the difference between the estimated power state and the power state at the next moment, update the weights of the n initial particles. For example, for multiple estimated power states, such as SOC C and SOC FZ , the weight update formula for the particles is: Among them, is the probability density function of the estimated power state z k , which reflects the matching degree between the particle and the actual corrected SOC. If the state of a certain particle is closer to the actual corrected power state, then the weight of this particle will increase, indicating that it is more likely to be the true value of the power state at the current moment;

[0095] After updating the weights of the particles, the particle filter will resample according to the weights of each particle, that is, retain the particles with larger weights and discard the particles with smaller weights. Stop updating until the updated weight reaches the preset weight size, and take the average of the obtained n″ updated particles as the corrected power state.

[0096] It is worth mentioning that the method for obtaining the corrected health state is the same as that for the corrected power state, so it will not be elaborated here.

[0097] In some embodiments, in order to further improve the accuracy of dynamic estimation, obtain the working state change nodes of the lithium iron phosphate battery. The working state change nodes include Node 1 and Node 2. Node 1 is the transition from the moving state to the stationary state, and Node 2 is the transition from the stationary state to the moving state. Record the node data after detecting Node 1. The node data is the power state, total driving time, cumulative discharge power, and riding load of the lithium iron phosphate battery before the node transition each time the node changes; the node data is used as the weight correction factor for initial particle distribution; the purpose is to accurately calibrate the power state at the beginning of each initial ride, without the need to correct the power state according to the load during the ride;

[0098] Exemplary; at the end of each ride (in a stationary state), record information such as the SOC of the battery, total driving time, cumulative discharge power, riding load, etc., and record environmental parameters such as temperature to avoid errors in the next calibration caused by environmental changes; at the start of the next ride, combine the previous ride data and the current open-circuit voltage of the battery, and adjust the initial SOC estimation through a correction model; use the final SOC (the battery level at the end) of the previous ride as the weight correction factor for the initial value of the next SOC. For example, if the current static SOC estimation is 100%, but the SOC at the end of the previous ride was only 80%, the initial SOC can be dynamically adjusted to a weighted value (such as 90%) to avoid errors directly relying on OCV estimation; based on the load history and riding mode, predict the SOC change trend of the next ride by analyzing the load and battery discharge behavior of the previous ride. For example, if a high-load ride last time caused the SOC to decrease rapidly, the discharge curve of the SOC estimation model can be dynamically adjusted in advance;

[0099] Specifically, for the calibration optimization method, optimize by combining particle filtering and a dynamic adjustment mechanism:

[0100] Initial particle correction, use the end SOC of the previous ride as the weight correction factor for the initial particles.

[0101] Adjust the particle distribution according to the stationary state and historical data to make the prediction closer to the actual value.

[0102] Load prediction and dynamic correction, during the ride, adjust the SOC estimation according to the real-time load and current change trend; for example: if a sudden increase in load is detected, dynamically increase the load correction coefficient in the model to avoid large fluctuations in the SOC display;

[0103] Dynamic discharge curve correction, use historical data (such as the current-voltage relationship curve during the previous ride) to correct the current discharge model. In some embodiments, introduce an adaptive mechanism, and gradually optimize the discharge curve with the accumulation of historical data from multiple rides to adapt to the user's riding mode and the actual performance of the battery.

[0104] Train prediction model one and prediction model two for predicting the state of charge and the state of health based on historical working data. Specifically, the training method for prediction model one is:

[0105] Collect multiple sets of historical working data and the corresponding real state of charge of the historical working data, and convert the corresponding real state of charge of the historical working data into feature vector one;

[0106] Divide multiple sets of historical working data and the corresponding real state of charge of the historical working data into a training set, a validation set, and a test set, and perform normalization processing on the data. For example, normalize the voltage, current, and temperature in the historical working data to the range of [0, 1];

[0107] To improve the model convergence speed, take the feature vector one as the input of the machine learning model one, and use the training set to train the machine learning model one. At the same time, monitor the performance of the machine learning model one through the validation set, adjust the hyperparameters to optimize the prediction effect. During the training process, aim to minimize the prediction error. When the error on the validation set converges and the performance of the machine learning model one meets the requirements, stop the training to obtain the machine learning model one that can accurately predict the power state as the prediction model one.

[0108] The training method of the prediction model two is as follows:

[0109] Collect multiple groups of historical working data and the corresponding true health states of the historical working data, and convert the corresponding true health states of the historical working data into feature vector two;

[0110] Divide multiple groups of historical working data and the corresponding true power states of the historical working data into a training set, a validation set, and a test set, and perform normalization processing on the data. Take the feature vector two as the input of the machine learning model two, use the training set to train the machine learning model two. At the same time, monitor the performance of the machine learning model two through the validation set, adjust the hyperparameters to optimize the prediction effect. During the training process, aim to minimize the prediction error. When the error on the validation set converges and the performance of the machine learning model two meets the requirements, stop the training to obtain the machine learning model two that can accurately predict the health state as the prediction model two.

[0111] Specifically, use the mean square error (MSE) as the loss function to measure the prediction error of the model:

[0112]

[0113] where, y i is the actual SOC or SOH value, is the SOC or SOH value predicted by the model, and n is the number of samples. The goal of the model is to minimize this loss function.

[0114] During the training process, input the historical working data into the machine learning model one or the machine learning model two, and adjust the parameters of the model through optimization algorithms such as the gradient descent method to minimize the loss function, thereby optimizing the prediction results of SOC and SOH; for model verification and evaluation, use the validation set data to evaluate the generalization ability of the model to ensure that the model can make accurate predictions on new data;

[0115] The machine learning model one is linear regression, decision tree, neural network, etc.

[0116] More specifically, the method for determining whether to generate a calibration instruction includes:

[0117] The calibration instructions include calibration instruction one and calibration instruction two;

[0118] Compare deviation value one and deviation value two with the preset deviation value one threshold and the preset deviation value two threshold respectively;

[0119] If deviation value one is greater than the preset deviation value one threshold, generate calibration instruction one and calibrate the state of charge of the BMS based on calibration instruction one;

[0120] If deviation value two is greater than the preset deviation value two threshold, generate calibration instruction two and calibrate the health state of the BMS based on calibration instruction two;

[0121] If deviation value one is less than or equal to the preset deviation value one threshold, no calibration instruction is generated;

[0122] If deviation value two is less than or equal to the preset deviation value two threshold, no calibration instruction is generated.

[0123] The purpose is that the predicted state of charge and the predicted health state predicted by prediction model one and prediction model two are predicted based on historical working data and have a high long-term prediction accuracy for lithium iron phosphate batteries. The corrected state of charge and the corrected health state estimated and corrected based on real-time working data are applicable to dynamic estimation. By calculating deviation value one and deviation value two, the accuracy of the prediction and the corrected estimation can be reflected. If the accuracy is high, calibration is not required; otherwise, calibration is required.

[0124] II. Refer to Figure 3 As shown, a lithium iron phosphate battery BMS parameter calibration device based on state estimation is applied to the above-mentioned lithium iron phosphate battery BMS parameter calibration method based on state estimation. The device includes an acquisition module, a state estimation unit, a model training module, a data fusion unit, and a calibration module, where each module is connected by wire and / or wirelessly:

[0125] The acquisition module is used to acquire real-time working data and continuously record it as historical working data;

[0126] The state estimation unit is used to estimate and obtain the estimated state of charge and the estimated health state based on real-time working data and historical working data;

[0127] The model training module is used to train prediction model one and prediction model two for predicting the state of charge and the health state based on historical working data, and input the real-time working data into prediction model one and prediction model two to obtain the predicted state of charge and the predicted health state respectively;

[0128] The data fusion unit is used to dynamically correct and obtain the corrected state of charge and the corrected health state based on the estimated state of charge and the estimated health state;

[0129] A calibration module, which is used to calculate the deviation value one between the predicted state of charge and the corrected state of charge and the deviation value two between the predicted health state and the corrected health state, and compare the deviation value one and the deviation value two with a preset deviation value one threshold and a preset deviation value two threshold respectively to determine whether to generate a calibration instruction for a lithium iron phosphate battery BMS parameter calibration device based on state estimation.

[0130] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments by using the above-disclosed technical content within the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A lithium iron phosphate battery BMS parameter calibration method based on state estimation, characterized in that: Monitoring methods include: Obtain real-time and historical work data; Obtain estimated power status and estimated health status based on real-time working data and historical working data; Based on historical working data, prediction models 1 and 2 are trained to predict power status and health status; Dynamically correct the estimated state of charge and the estimated state of health to obtain a corrected state of charge and a corrected state of health; The deviation value 1 between the predicted power state and the corrected power state and the deviation value 2 between the predicted healthy state and the corrected healthy state are calculated respectively, and the deviation value 1 and the deviation value 2 are compared with the preset deviation value 1 threshold and the preset deviation value 2 threshold respectively to determine whether to generate a calibration instruction.

2. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 1, characterized in that: Obtain the working state change node of the lithium iron phosphate battery, the working state change node includes node 1 and node 2, node 1 is from the moving state to the stationary state, and node 2 is from the stationary state to the moving state.

3. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 1, characterized in that: After node one is monitored, the node data is recorded. The node data includes the power status of the lithium iron phosphate battery before the node changes, the total driving time, the accumulated discharge power, and the riding load each time the node changes.

4. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 1, characterized in that: The method for estimating the state of charge is as follows: Initialize the power state, collect the charge and discharge current value of the lithium iron phosphate battery at each moment, and calculate the basic power state based on the integral formula of the current value and the collection time; A temperature correction coefficient and a load correction coefficient are preset, and a temperature-corrected power state and a load-corrected power state are obtained by formulating and calculating the temperature correction coefficient and the load correction coefficient; The estimated state of charge is calculated based on the temperature-corrected state of charge and the load-corrected state of charge.

5. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 4, characterized in that: The method for estimating the health status is: Correct internal resistance based on a preset temperature correction coefficient; The estimated health status is calculated based on the modified internal resistance formula.

6. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 5, characterized in that: The dynamic correction method for correcting the health status is the same as the dynamic correction method for correcting the power status. The dynamic correction method for correcting the power status is: Based on the estimated power state, n initial particles are randomly generated, where n is an integer greater than 1, and corresponding weight correction factors are assigned to the n initial particles based on the node data; Pre-build the battery charging and discharging model, predict n initial particles, and obtain n′ next-moment power states; Based on the difference between the estimated power state and the power state at the next moment, the weights of the n initial particles are updated and resampled, that is, the initial particles with larger weights are retained until the updated weight reaches the preset weight size and the updating is stopped to obtain n″ updated particles, and the average value of the obtained n″ updated particles is taken as the corrected power state.

7. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 1, characterized in that: The method of determining whether to generate a calibration instruction includes: The calibration instruction includes calibration instruction one and calibration instruction two; Compare the deviation value 1 and the deviation value 2 with the preset deviation value 1 threshold and the preset deviation value 2 threshold respectively; If the deviation value 1 is greater than the preset deviation value 1 threshold, a calibration instruction 1 is generated, and the power state of the BMS is calibrated based on the calibration instruction 1; If the deviation value 2 is greater than the preset deviation value 2 threshold, a calibration instruction 2 is generated, and the health status of the BMS is calibrated based on the calibration instruction 2; If the deviation value is less than or equal to the preset deviation value threshold, no calibration instruction is generated; If the deviation value 2 is less than or equal to the preset deviation value 2 threshold, no calibration instruction is generated.

8. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 7, characterized in that: The training method of the prediction model 1 is: Collect multiple sets of historical working data and real power states corresponding to the historical working data, and convert the real power states corresponding to the historical working data into a feature vector one; Divide multiple groups of historical working data and the actual power status corresponding to the historical working data into a training set, a validation set, and a test set, and normalize the data. Use feature vector one as the input of machine learning model one, use the training set to train machine learning model one, and monitor the performance of machine learning model one through the validation set. Adjust hyperparameters to optimize the prediction effect. During the training process, minimize the prediction error as the goal. When the error on the validation set converges and the performance of machine learning model one meets the requirements, stop training, and obtain machine learning model one that can accurately predict the power status as prediction model one; One machine learning model is a decision tree or a neural network.

9. The method for calibrating BMS parameters of a lithium iron phosphate battery based on state estimation according to claim 7, characterized in that: The training method of the prediction model 2 is: Collect multiple groups of historical work data and real health status corresponding to the historical work data, and convert the real health status corresponding to the historical work data into feature vector 2; Divide multiple groups of historical working data and the actual power status corresponding to the historical working data into a training set, a validation set, and a test set, and perform a normalization process on the data. Use the feature vector 2 as the input of the machine learning model 2, use the training set to train the machine learning model 2, and monitor the performance of the machine learning model 2 through the validation set. Adjust the hyperparameters to optimize the prediction effect. During the training process, minimize the prediction error as the goal. When the error on the validation set converges and the performance of the machine learning model 2 meets the requirements, stop the training, and obtain the machine learning model 2 that can accurately predict the health status as the prediction model 2. The second machine learning model is a decision tree or a neural network.

10. A lithium iron phosphate battery BMS parameter calibration device based on state estimation, applied to the lithium iron phosphate battery BMS parameter calibration method based on state estimation according to any one of claims 1 to 9, characterized in that: The device includes: The collection module is used to collect real-time working data and continuously record it as historical working data; A state estimation unit, used to estimate the power state and health state based on real-time working data and historical working data; A model training module is used to train prediction model 1 and prediction model 2 for predicting power state and health state based on historical working data, and input real-time working data into prediction model 1 and prediction model 2 to obtain predicted power state and predicted health state respectively; A data fusion unit, used for dynamically correcting the estimated state of charge and the estimated state of health to obtain a corrected state of charge and a corrected state of health; The calibration module is used to calculate the deviation value 1 between the predicted power state and the corrected power state and the deviation value 2 between the predicted health state and the corrected health state, and compare the deviation value 1 and the deviation value 2 with the preset deviation value 1 threshold and the preset deviation value 2 threshold respectively to determine whether to generate a calibration instruction.

Images