Battery life prediction method and system based on BMS (Battery Management System)

Abstract

The invention discloses a battery life prediction method and system based on a BMS system, and relates to the technical field of battery management, and the method comprises the steps: preliminarily screening out a key variable set based on an internal electrochemical characteristic variable and an external environment variable according to the influence degree of each variable on battery life prediction; constructing a state matrix group based on the key variable set; obtaining a dominating feature set based on the matrix correlation model and the state matrix group, and obtaining a common matrix based on the dominating feature set; obtaining actual observation data of the to-be-predicted battery, and obtaining a personalized state matrix based on the actual observation data and the common matrix; inputting the personalized state matrix into a preset life cycle prediction model to obtain life prediction data of the to-be-predicted battery; and key variables influencing battery life prediction are screened out to form a key variable set, so that the prediction accuracy and comprehensiveness are further improved.

Description

Technical Field

[0001] The present invention relates to the field of battery management, and in particular to a battery life prediction method and system based on a BMS system. Background Art

[0002] With the development of application fields such as electric vehicles and energy storage batteries, higher requirements are placed on battery life prediction methods. Battery life prediction methods require the use of reliable battery management systems. At present, there is an urgent need to develop a more intelligent BMS battery management system. The accuracy of parameter classification of battery status data in the prior art has a great impact on the prediction accuracy, and the battery life will be affected by factors such as ambient temperature and humidity, which leads to low accuracy of the prior art.

[0003] The Chinese invention patent with application number 202310934815.6 provides a battery life prediction method and device based on the BMS battery management system, which constructs a first state parameter distribution map according to the target voltage data and the target current data, and constructs a second state parameter distribution map according to the target temperature data and the target power data; extracts the first characteristic point and the second characteristic point respectively; constructs a battery state matrix; and inputs the battery state matrix into the life cycle prediction model to predict the charge and discharge life cycle.

[0004] However, in the prior art, when faced with dealing with complex and changeable battery operating conditions and uncertain external environmental factors, the prior art does not analyze the interaction between the internal electrochemical characteristics of the battery and the external environmental factors, and ignores the relationship between the generated state matrices, which affects the comprehensiveness and accuracy of the prediction and causes large deviations in the prediction results. Summary of the invention

[0005] In view of the technical problems existing in the prior art, the present invention provides a battery life prediction method and system based on a BMS system, which solves the problem in the prior art that the data dimension is single and affects the prediction accuracy.

[0006] The technical solution of the present invention to solve the above technical problem is as follows: a battery life prediction method based on a BMS system, the method comprising: S100: Obtain observation data of multiple target batteries, extract internal electrochemical characteristic variables and external environmental variables based on the observation data, preliminarily screen out a key variable set according to the influence of each variable on battery life prediction; and construct a state matrix group based on the key variable set; S200: obtaining a dominant feature set based on the matrix association model and the state matrix group, and obtaining a commonality matrix based on the dominant feature set; S300: Acquire actual observation data of the battery to be predicted, and obtain a personalized state matrix based on the actual observation data and the commonality matrix; S400: Inputting the personalized state matrix into a preset life cycle prediction model to obtain life prediction data of the battery to be predicted.

[0007] Furthermore, step S200 also includes: S210: Obtain variable parameters in the state matrix group, and divide the variable parameters into levels; establish a hierarchical nesting model of the variable parameters, analyze the nesting relationship between variable parameters at different levels, and obtain a nesting relationship analysis result; S220: Based on the nested relationship analysis results, a coupling model between parameters at different levels is established; based on the coupling model, multi-scale features are extracted respectively, and cross-scale collaborative analysis between multi-level parameters is performed based on the multi-scale features; S230: Filter out dominant features based on the cross-scale collaborative analysis results, and combine the filtered dominant features to form a dominant feature set; The hierarchical division includes micro-level parameters, meso-level parameters and macro-level parameters.

[0008] Furthermore, the method further comprises: S500: constructing an uncertainty quantification model, determining the probability distribution of each dominant feature based on the dominant feature set and the uncertainty quantification model, and generating a number of random samples according to the probability distribution of each dominant feature; S600: inputting the generated random samples into a preset life cycle prediction model to obtain a battery life prediction value corresponding to each group of samples; according to the statistical characteristics of all battery life prediction values, obtaining the contribution degree of the dominant characteristics to the uncertainty of the prediction result, and then calculating the confidence interval of the prediction result; S700: Input the confidence intervals of different dominant features and the corresponding personalized state matrix into a preset life cycle prediction model, and execute step S400.

[0009] Furthermore, the method further comprises: S800: Establish a battery group to be predicted, wherein the battery group to be predicted includes a plurality of batteries to be predicted; obtain a personalized state matrix of each battery to be predicted to form a cross-group matrix; based on the cross-group matrix, establish a cross-group life collaborative prediction mechanism, determine the collaborative rules between different batteries, and obtain a cross-group collaborative rule base.

[0010] A battery life prediction system based on a BMS system, the system comprising: The data acquisition and processing module is used to obtain the observation data of multiple target batteries, extract the internal electrochemical characteristic variables and external environmental variables based on the observation data, preliminarily screen out the key variable set according to the influence of each variable on the battery life prediction; and construct a state matrix group based on the key variable set; A commonality matrix generation module is used to obtain a dominant feature set according to a matrix association model and a state matrix group, and obtain a commonality matrix based on the dominant feature set; A personalized state matrix generation module is used to obtain actual observation data of the battery to be predicted, and obtain a personalized state matrix based on the actual observation data and the commonality matrix; The uncertainty analysis module is used to determine the probability distribution of each dominant feature according to the dominant feature set and the uncertainty quantification model, and generate a number of random samples according to the probability distribution of each dominant feature; input the generated random samples into the preset life cycle prediction model to obtain the battery life prediction value corresponding to each group of samples; according to the statistical characteristics of all battery life prediction values, obtain the contribution of the dominant feature to the uncertainty of the prediction result, and then calculate the confidence interval of the prediction result; The cross-group collaboration module is used to establish the battery group to be predicted, obtain the personalized state matrix of each battery to be predicted, and form a cross-group matrix; based on the cross-group matrix, a cross-group life collaborative prediction mechanism is established, the collaborative rules between different batteries are determined, and a cross-group collaborative rule base is obtained; The life prediction module is used to input the personalized state matrix and confidence interval into the preset life cycle prediction model to obtain the life prediction data of the battery to be predicted.

[0011] The beneficial effects of the present invention are as follows: by calculating the correlation coefficients between internal electrochemical characteristic variables and external environmental variables and battery life, key variables that affect battery life prediction are screened out to form a key variable set; and the correlation matrix, sample data matrix and time series matrix constructed further improve the accuracy and comprehensiveness of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic diagram of a battery life prediction method based on a BMS system of the present invention Figure 1 ; Figure 2 A schematic diagram of a battery life prediction method based on a BMS system of the present invention Figure 2 ; Figure 3 This is a battery life prediction system architecture diagram based on a BMS system of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0014] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0015] In the description of the present application, the term "for example" is used to mean "used as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid unnecessary details to obscure the description of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.

[0016] Example 1 like Figure 1 As shown, a battery life prediction method based on a BMS system, the method comprising: S100: Obtain observation data of multiple target batteries, extract internal electrochemical characteristic variables and external environmental variables based on the observation data, preliminarily screen out a key variable set according to the influence of each variable on battery life prediction; and construct a state matrix group based on the key variable set; In this embodiment, the internal electrochemical characteristic variables refer to quantifiable physical or chemical parameters that characterize the electrochemical state and reaction process of the battery, and are monitored in real time through dedicated sensors or equivalent circuit models. The internal electrochemical characteristic variables include but are not limited to: electrical characteristic parameters, chemical state parameters, thermal characteristic parameters and kinetic parameters; the electrical characteristic parameters include terminal voltage, charge and discharge current, internal resistance and open circuit voltage; the chemical state parameters include charge state, health state, lithium ion concentration gradient and electrolyte decomposition product concentration; the thermal characteristic parameters include internal temperature distribution, tab temperature rise rate and thermal runaway critical threshold; the kinetic parameters include lithium ion diffusion coefficient and solid electrolyte interface membrane impedance.

[0017] The external environmental variables include mode variable data and extreme variable data, and the mode variable data is a high-frequency environmental variable that characterizes the statistics of the battery in the normal operation cycle; specifically, the environmental data of all the operation cycles of the target battery are obtained, the environmental variables in the environmental data are counted, and the frequency of occurrence of each environmental variable is determined according to the frequency. Whether the environmental variable belongs to high frequency is determined by presetting the frequency threshold or judging according to the actual statistical situation. For example, the statistical environmental variables are classified according to different attributes (such as temperature, mechanical, electrical stress, etc.), and the frequency of occurrence of specific variable values ​​in different categories is arranged in descending order, and the selection ratio is set. The environmental variables are selected in turn according to the selection ratio and defined as high-frequency environmental variables; the selection ratio needs to be set according to the actual statistical situation to adjust the conditions for selecting environmental variables. For example, climate category: median of ambient temperature, mode of relative humidity; mechanical category: root mean square value of vibration acceleration, frequency distribution of impact load; electrical stress category: mode of charge and discharge rate, median of cycle depth; time series category: typical value of daily temperature amplitude, time distribution of charge and discharge cycle, etc.

[0018] The extreme variable data refers to the extreme environmental variables counted during the entire operating cycle of the battery, which are defined as the physical limits or statistical boundary values ​​of each variable, such as climate limits: storage temperature extremes, humidity thresholds; mechanical limits: maximum vibration acceleration, extreme impact loads; electrical stress limits: maximum charge and discharge rates, overcharge / overdischarge cut-off voltages; timing limits: continuous charge and discharge duration extremes, standstill time thresholds; specifically, during the monitoring process, high-precision sensors are used for detection, and the detection limits of the sensors are also defined as extreme variable data.

[0019] According to the influence of each variable on the battery life prediction, a key variable set is obtained, including: calculating the correlation coefficient between each variable and the battery life, adding the variables corresponding to the correlation coefficient greater than a preset relationship threshold into the set to form a key variable set.

[0020] In some embodiments, an empty set is pre-set, and the correlation coefficient between each variable and the battery life is calculated one by one, and compared with the pre-set relationship threshold respectively, and the variables corresponding to the correlation coefficient greater than the relationship threshold are added to the set, and finally a key variable set is formed. The relationship threshold needs to be dynamically set according to historical data or historical experimental data, and is used to represent the critical value of the corresponding variable affecting the battery life. Preferably, experimental monitoring is performed under different environmental conditions to determine the impact of different variables on the battery life.

[0021] Calculate the correlation coefficient between each variable and battery life, including: establish a multiple linear regression model, with battery life as the dependent variable and each variable as the independent variable; determine the influence of each variable on battery life through the size and significance test of the regression coefficient.

[0022] In some embodiments, a multivariate linear regression model is established, specifically including: determining the dependent variable (battery life) and independent variables (various factors affecting battery life, i.e., internal electrochemical characteristic variables and external environmental variables); collecting a large amount of observational data containing dependent variables and independent variables to ensure the quality and integrity of the data, for example, the data should have no obvious missing values, outliers, etc. To handle missing values, mean filling, median filling, or model-based methods can be used; outliers can be identified and processed, and corrected or eliminated; independent variables are normalized (data is scaled to the interval [0, 1]) so that data of different variables can be in the same dimension for comprehensive comparison or calculation. The data set is divided into a training set and a test set. The commonly used division ratio is 7:3 or 8:2. The training set is used to build the model, and the test set is used to evaluate the performance of the model.

[0023] Model construction, setting the specific form of the multiple linear regression model, for example:

[0024] in, is the dependent variable, is the independent variable, is the intercept term, is the regression coefficient, is a random error. The least squares method is used to estimate the regression coefficient using the training set data. When evaluating the model, a goodness of fit test is required to calculate the coefficient of determination. The range of the coefficient of determination is [0, 1]. The closer the coefficient of determination is to 1, the better the model fits the data. When performing a significance test, the overall significance test (F test) and the significance test of a single regression coefficient (t test) are used. The test method is an existing related technology, and this application will not go into details here. The absolute value of the regression coefficient reflects the degree of influence of the independent variable on the dependent variable.

[0025] For example, collect battery data and study charging times , Use temperature and depth of discharge Impact on battery life (y). First, data preprocessing is performed, namely data cleaning and data standardization. The data set is divided into a training set (70 samples) and a test set (30 samples) in a ratio of 7:3. Set the multivariate linear regression model: Using the training set data, the regression coefficients are estimated by the least squares method, and we get The calculated determination coefficient is 0.85, indicating that the model fits the data well.

[0026] The F test was performed and the calculated F statistic was 30.5. At the significance level of 0.05, the null hypothesis was rejected and the model was considered significant as a whole. The t test was performed and the p values ​​corresponding to the regression coefficients were 0.01, 0.03, and 0.001, respectively, all less than 0.05, indicating that the three independent variables had a significant impact on the dependent variable.

[0027] Judging from the size of the regression coefficient, the absolute value of the regression coefficient of the discharge depth is the largest, indicating that the discharge depth has the greatest impact on the battery life; followed by the number of charges, and the impact of the operating temperature is relatively small.

[0028] Judging from the results of the significance test, all three independent variables are significant, further confirming that they have a significant impact on battery life.

[0029] The state matrix group includes a correlation matrix, a sample data matrix and a time series matrix. The correlation matrix is ​​a matrix formed according to the correlation coefficients between different key variables in the key variable set, which is used to identify the linear relationship between the key variables and identify the highly correlated variable group. At the same time, for highly correlated variables, in order to avoid multicollinearity problems in the model, it can be considered to retain one of them; the sample data matrix is ​​a matrix formed according to the observed data of the key variables, with the observed data as rows and the key variables as columns, and different observed data represent different samples. The sample data matrix is ​​the basis for training and prediction of the multivariate linear regression model. By inputting the sample data into the model, the regression coefficient can be estimated and a relationship model between the battery life and the key variables can be established. The time series matrix is ​​to sort the observed data of the key variables in chronological order, with the time points as rows and the key variables as columns, to construct a time series matrix; the time series matrix is ​​used to analyze the changing trend of the key variables over time, for example, by drawing a time series graph to observe the fluctuation of the variables. Preferably, the time series analysis method can also be used to predict the battery life.

[0030] S200: Obtain a dominant feature set based on the matrix association model and the state matrix group, and obtain a commonality matrix based on the dominant feature set.

[0031] In some embodiments, the core goal of the matrix association model is to fuse the information of the correlation matrix, sample data matrix, and time series matrix, extract the dominant features across the matrix through principal component analysis (PCA), and determine the core features (such as the common change trend of temperature and internal resistance). The dominant features include dynamic features: trend items in the time series (such as the internal resistance increases by 0.3mΩ every 100 cycles); static features: mean and standard deviation in sample data (such as the average charge and discharge current is ±1.5A); association features: correlation coefficient threshold (such as retaining variable pairs with |ρ|≥0.5). According to the set of dominant features, the structure and format of the commonality matrix are designed. The commonality matrix contains placeholders for key variables to form the final commonality matrix.

[0032] Specifically, PCA is an unsupervised dimensionality reduction method that projects high-dimensional data into a low-dimensional space through linear transformation, retaining information in the direction of maximum variance. The purpose is to find the orthogonal direction (principal component) with the largest variance in the data. The features of the three matrices are concatenated by column to form a new data set, the integrated data set is standardized, the covariance matrix of the standardized data is calculated, the covariance matrix is ​​eigenvalue decomposed, and the eigenvalues ​​are sorted from large to small. The principal components with a cumulative variance contribution rate of ≥80% are selected, and the principal component loads are analyzed to determine the original variables with the main associations. For example, the integrated data set contains 5 features, including voltage, current, temperature, internal resistance, and SOC, and 1525 samples.

[0033] PCA results: The first principal component (PC1) explains 65% of the variance, with loadings of [0.45, -0.32, 0.68, -0.72, -0.41], corresponding to temperature, internal resistance, and SOC.

[0034] The second principal component (PC2) explained 20% of the variance, with loadings of [-0.82, 0.58, 0.12, 0.05, 0.03], corresponding to voltage and current.

[0035] Then, the dominant features are: the coordinated change of temperature-internal resistance-SOC (PC1) and the correlation between voltage and current (PC2).

[0036] S300: Acquire actual observation data of the battery to be predicted, and obtain a personalized state matrix based on the actual observation data and the commonality matrix.

[0037] In some embodiments, the placeholders containing key variables in the commonality matrix are replaced with key data in the actual observation data to form a personalized state matrix, and the commonality matrix is ​​dynamically adjusted according to the specific characteristics of the battery to be predicted (such as battery type, usage environment, etc.). For example, for a specific battery, the placeholders are replaced with actual measured values, such as the terminal voltage is 3.75V, the charge and discharge current is 2.1A, the median ambient temperature is 26°C, and the charge and discharge rate mode is 1.6. According to the use of the battery in a high temperature environment, the weight of the median ambient temperature variable is increased.

[0038] S400: Inputting the personalized state matrix into a preset life cycle prediction model to obtain life prediction data of the battery to be predicted.

[0039] In some embodiments, the life cycle prediction model is a machine learning model used to predict the charging and discharging process, remaining life and other information of the battery. The model needs to input the state matrix of the battery in order to track and predict the behavior of the battery. Therefore, the battery state matrix is ​​input into the life cycle prediction model for processing. The life cycle prediction model includes: a bidirectional long short-term memory network, a first codec network and a second codec network. The life cycle prediction model is a hybrid model composed of three sub-models, including: a bidirectional long short-term memory network (Bi-LSTM): used to extract features from the matrix. Bi-LSTM has a certain memory capacity and can perform well on long-term dependency problems. By using Bi-LSTM to extract features from the matrix, the dynamic change information of the battery state can be better captured. The first codec network (AE1): used to learn a low-dimensional representation of the state matrix. By compressing the battery state matrix into a low-dimensional representation, the prediction and tracking of the charge and discharge life cycle can be better achieved. The second codec network (AE2): used to learn to predict the remaining life of the battery. By learning the evolution of the battery state, the remaining life of the battery can be more accurately predicted, and charging and discharging operations can be performed when necessary to extend the battery life. In the life cycle prediction model, a bidirectional long short-term memory network (Bi-LSTM) is used to extract features from the battery state matrix. Bi-LSTM can capture long-term dependencies in continuous time series data, thereby improving the accuracy of the model. By inputting the battery state matrix into the Bi-LSTM model, the features of each time step can be extracted.

[0040] The first codec network is an autoencoder model for learning a low-dimensional representation of the state matrix. In the charging cycle prediction, the target feature matrix is ​​input into the first codec network to learn the low-dimensional representation of the state matrix. When predicting the charging cycle, the learned low-dimensional features are input into some classification models, such as decision trees or support vector machines, to obtain the charging cycle prediction results and form a charging cycle prediction data set. Similarly, for the discharge cycle prediction, the target feature matrix is ​​input into the second codec network to learn its low-dimensional representation. When predicting the discharge cycle, the learned low-dimensional features are also used as input to perform a classification model to obtain the discharge cycle prediction results and form a discharge cycle prediction data set. After the charging and discharge cycle prediction data sets are obtained, they can be curve mapped, and the results of the curve mapping are used as the predicted values ​​of the charging and discharge cycle curves to obtain the life prediction data. When establishing a prediction model, a variety of data models can be selected for training, and the specific selection needs to be made according to the actual situation. This application does not make specific restrictions here.

[0041] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application calculates the correlation coefficient between internal electrochemical characteristic variables and external environmental variables and battery life, screens out key variables that affect battery life prediction, and forms a key variable set; and constructs a correlation matrix, sample data matrix, and time series matrix to further improve the accuracy and comprehensiveness of the prediction.

[0042] Embodiment 2, in some embodiments, step S200 further includes: S210: Obtain variable parameters in the state matrix group, and divide the variable parameters into levels; establish a hierarchical nesting model of the variable parameters, analyze the nesting relationship between variable parameters at different levels, and obtain a nesting relationship analysis result.

[0043] The hierarchical division includes micro-level parameters, meso-level parameters and macro-level parameters; for example, the electrical characteristic parameters (terminal voltage, charge and discharge current, internal resistance and open circuit voltage) in the internal electrochemical characteristic variables are taken as micro-level parameters; the chemical state parameters (charge state, health state, lithium ion concentration gradient and electrolyte decomposition product concentration) are taken as meso-level parameters; the thermal characteristic parameters (internal temperature distribution, pole ear temperature rise rate and thermal runaway critical threshold) and kinetic parameters (lithium ion diffusion coefficient and solid electrolyte interface film impedance) are taken as macro-level parameters. The mode variable data and extreme variable data in the external environmental variables are also divided into different levels according to their influence range and nature.

[0044] In some embodiments, constructing a hierarchical nested model specifically includes: collecting observation data of multiple target batteries, including data on internal electrochemical characteristic variables and external environmental variables; preprocessing the data, including data cleaning (processing missing values ​​and outliers), normalization, etc.; constructing a correlation matrix, a sample data matrix, and a time series matrix based on the preprocessed data; determining the hierarchical structure and constructing a hierarchical nested model of parameters.

[0045] In some embodiments, based on the hierarchical nested model, a correlation matrix between parameters at different levels is obtained, such as a correlation matrix between micro-level parameters and meso-level parameters, wherein rows represent different electrical characteristic parameters at the micro-level, columns represent different chemical state parameters at the meso-level, and matrix elements are correlation coefficients; the nested relationship between the variable parameters at different levels is analyzed to obtain the nested relationship analysis results; the nested relationship analysis results are preliminary analysis results between the parameters at different levels, which are mainly reflected in the clear mutual influence paths and logical associations between the parameters at different levels, specifically including: the influence of the micro-level on the meso-level: the electrical characteristic parameters at the micro-level, such as terminal voltage, charge and discharge current, internal resistance and open circuit voltage, will directly affect the chemical state parameters at the meso-level. For example, changes in terminal voltage will reflect the activity of the electrochemical reaction inside the battery, thereby affecting the state of charge (SOC). When the terminal voltage gradually changes during normal charge and discharge, the SOC will increase or decrease accordingly. The size and rate of change of the charge and discharge current will affect the lithium ion migration and chemical reaction rate inside the battery, thereby affecting the state of health (SOH). If the charge and discharge current is too large, it may accelerate the aging of the battery and cause the SOH to decrease.

[0046] The impact of the meso-level on the macro-level: The chemical state parameters at the meso-level will further affect the thermal characteristic parameters and kinetic parameters at the macro-level. For example, changes in the state of charge will affect the chemical reaction heat inside the battery, thereby changing the internal temperature distribution. When the SOC is higher, the chemical reaction inside the battery may be more intense, generating more heat, causing the internal temperature to rise. The deterioration of the health state will affect the diffusion coefficient of lithium ions in the electrolyte, and then affect the kinetic parameter of the lithium ion diffusion coefficient. At the same time, changes in chemical state parameters will also affect thermal characteristic parameters such as the temperature rise rate of the tab and the critical threshold of thermal runaway.

[0047] S220: Based on the nested relationship analysis results, a coupling model between parameters at different levels is established; based on the coupling model, multi-scale features are extracted respectively, and cross-scale collaborative analysis between multi-level parameters is performed based on the multi-scale features.

[0048] In some embodiments, a coupling model between parameters at different levels is established. The coupling model construction can select the above-mentioned multivariate linear regression model or neural network model, etc., which can be adjusted and selected according to the actual situation. This application will not go into details here. Based on the coupling model, multi-scale features are extracted respectively, for example, the short-term fluctuation characteristics of electrical characteristic parameters are extracted from the microscopic level, the long-term change trend characteristics of chemical state parameters are extracted from the mesoscopic level, and the periodic change characteristics of thermal characteristic parameters and kinetic parameters are extracted from the macroscopic level. Then analyze how the short-term fluctuation characteristics at the microscopic level affect the long-term change trend characteristics at the mesoscopic level, and how the periodic change characteristics at the macroscopic level interact with the characteristics at the microscopic and mesoscopic levels. Specifically including: short-term fluctuation characteristic data at the microscopic level, such as the fluctuation range of the terminal voltage per minute; long-term change trend characteristic data at the mesoscopic level, such as the weekly change rate of the state of charge; periodic change characteristic data at the macroscopic level, such as the daily cycle peak value of the internal temperature distribution.

[0049] In some embodiments, a cross-scale synergistic analysis is performed between multi-level parameters based on multi-scale features, for example, analyzing the synergistic effects of the charge and discharge current at the micro level, the lithium ion concentration gradient at the meso level, and the internal temperature distribution at the macro level on the battery life. The result of the cross-scale synergistic analysis is: the specific interaction effects between different levels, which are used to indicate the specific data relationship between parameters at different levels, for example: when the charge and discharge current at the micro level exceeds 2A, and the rate of change of the lithium ion concentration gradient at the meso level is greater than 0.1 / h, and the peak value of the internal temperature distribution at the macro level exceeds 40°C, the battery life is expected to be shortened by 20%.

[0050] S230: Filter out dominant features based on the cross-scale collaborative analysis results, and combine the filtered dominant features to form a dominant feature set.

[0051] In this embodiment, by establishing a hierarchical nested model, the battery status is analyzed simultaneously from the macro, meso and micro levels, which can more comprehensively understand the interaction between the internal and external factors of the battery and avoid the limitations of single-level analysis; cross-scale collaborative analysis can discover patterns that cannot be discovered by single parameter analysis, and considers the synergistic effects between parameters at different levels, thereby improving the comprehensiveness and accuracy of battery life prediction; multi-scale characteristic analysis reveals the characteristics of the battery status at different scales, so that the prediction model can better adapt to the changes of the battery under different working conditions and environments, and improves the generalization ability of the model; the output multi-scale characteristics and cross-scale laws provide a deeper decision-making basis for the design, use and maintenance of the battery. For example, the battery usage strategy can be optimized according to these laws to extend the battery life.

[0052] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application establishes a hierarchical nested model to simultaneously analyze the battery status from the macro, meso and micro levels, which can more comprehensively understand the interaction between internal and external factors of the battery and avoid the limitations of single-level analysis; Cross-scale collaborative analysis can discover patterns that cannot be discovered by single parameter analysis, and consider the synergistic effects between parameters at different levels, further improving the comprehensiveness and accuracy of battery life prediction; Multi-scale characteristic analysis reveals the characteristics of the battery state at different scales, enabling the prediction model to better adapt to the changes of the battery under different working conditions and environments, and improving the generalization ability of the model.

[0053] Embodiment 3, as Figure 2 As shown, the method also includes: S500: constructing an uncertainty quantification model, determining the probability distribution of each dominant feature based on the dominant feature set, and generating a number of random samples according to the probability distribution of each dominant feature; S600: inputting the generated random samples into a preset life cycle prediction model to obtain a battery life prediction value corresponding to each group of samples; according to the statistical characteristics of all battery life prediction values, obtaining the contribution degree of the dominant characteristics to the uncertainty of the prediction result, and then calculating the confidence interval of the prediction result; S700: inputting the confidence intervals of different dominant characteristics and the corresponding personalized state matrix into a preset life cycle prediction model to obtain life prediction data of the battery to be predicted.

[0054] In some embodiments, a Monte Carlo simulation method is selected as an uncertainty quantification model. Monte Carlo simulation is a method for approximating the behavior of a complex system through a large number of random samplings, which is suitable for processing systems with uncertainty. Based on the set of dominant features, the probability distribution of each dominant feature is determined. For example, for the dominant feature of temperature, it is assumed that it obeys a normal distribution, and its mean and standard deviation are estimated through historical data; for discrete features such as charge and discharge rate, its possible value range and corresponding probability are determined.

[0055] A random number generator is used to generate a large number of random samples according to the probability distribution of each dominant feature. For example, 10,000 random combinations of different dominant features are generated, each of which represents a characteristic value of the battery under a possible working state.

[0056] Input the generated random samples into the preset life cycle prediction model to obtain the battery life prediction value corresponding to each group of samples. Calculate the statistical characteristics of all predicted values, such as mean, standard deviation, minimum value, maximum value, etc. The standard deviation reflects the degree of dispersion of the prediction results. The larger the standard deviation, the higher the uncertainty of the prediction results. Draw the frequency distribution histogram and probability density function curve of the prediction results to intuitively display the distribution of the prediction results. For example, the histogram can be used to observe the frequency of predicted life in different intervals, and the probability density function curve can more clearly reflect the probability distribution of the predicted life.

[0057] In some embodiments, the contribution of the dominant feature to the uncertainty of the prediction result is obtained, including: performing local sensitivity analysis and global sensitivity analysis on each dominant feature to quantify the contribution; the local sensitivity analysis is based on the influence of a single dominant feature on the prediction result; the global sensitivity analysis is to decompose the variance of the prediction result into the contribution analysis of each dominant feature and its interaction.

[0058] Specifically, a local sensitivity analysis method, such as calculating partial derivatives, is used to analyze the influence of each dominant feature on the prediction results. Specifically, for each dominant feature, the value of the feature is slightly changed while other features remain unchanged, and the changes in the prediction results are observed. The greater the change, the more sensitive the feature is to the prediction results.

[0059] The Sobol method is used to perform global sensitivity analysis, taking into account the interactions between dominant features. The Sobol method can decompose the variance of the prediction results into the contributions of each dominant feature and its interaction, thereby quantifying the contribution of each feature to the uncertainty of the prediction results. The Sobol method is a method for evaluating the influence of model input parameters on output results. It belongs to the variance decomposition method. It decomposes the total variance of the model output into the variance of each parameter's individual contribution and the interaction between parameters, and quantifies the importance of the parameters.

[0060] According to the results of sensitivity analysis, the dominant features that contribute more to the uncertainty of the prediction results, that is, the key factors, are determined. According to the analysis results of the Sobol method, the contribution of each feature to the uncertainty of the prediction results is quantified.

[0061] In some embodiments, calculating the confidence interval of the prediction result includes: selecting an appropriate confidence level, such as 95%, according to actual needs; judging according to the distribution type of the prediction result (such as normal distribution); calculating the confidence interval using corresponding statistical methods; and presenting the calculated confidence interval together with the prediction result. The confidence interval refers to the estimated interval of the population parameter constructed by the sample statistic, which represents a range in which the true value of the population parameter may fall under a given confidence level.

[0062] For example, a 95% confidence level means that in the confidence interval constructed by a large number of repeated samplings, approximately 95% of the intervals will contain the true value of the population parameter. In other words, there is 95% confidence that the true value of the population parameter falls within the calculated confidence interval. Taking battery life prediction as an example, the battery life is predicted to be 1000 cycles, and the 95% confidence interval is [804, 1196] cycles. This means that if multiple battery life predictions are made and a 95% confidence interval is calculated each time, then in these intervals, approximately 95% of the intervals will contain the true value of the battery life. At the same time, this interval also understands the credible range of the prediction results, that is, there is a 95% probability that the true value of the battery life falls between 804 cycles and 1196 cycles. This further increases the accuracy and controllability of life prediction.

[0063] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application generates a large number of random samples and comprehensively considers the uncertainty of the dominant characteristics, which can more accurately evaluate the uncertainty of the battery life prediction results; The contribution of each dominant characteristic to the uncertainty of the prediction results is quantified, the key factors are clarified, the confidence interval of the prediction results is calculated, and the credible range of the prediction results is given, which increases the accuracy and controllability of life prediction and can more accurately reflect the actual conditions of different batteries.

[0064] Embodiment 4, as Figure 2 As shown, the method also includes: S800: establishing a battery group to be predicted, the battery group to be predicted includes a plurality of batteries to be predicted; obtaining a personalized state matrix of each battery to be predicted to form a cross-group matrix; based on the cross-group matrix, establishing a cross-group life collaborative prediction mechanism, determining the collaborative rules between different batteries, and obtaining a cross-group collaborative rule base.

[0065] In some embodiments, a battery group to be predicted is established, and the battery group to be predicted includes several batteries to be predicted. Battery group-level feature fusion is introduced to calculate the variable correlation matrix between batteries (such as the correlation between the temperature of battery A and the charge and discharge rate of battery B); a cross-group matrix is ​​formed according to the personalized state matrix of each battery to be predicted, and the time-aligned observation values ​​of key variables of different battery groups are recorded (such as the synchronous change sequence of the terminal voltage of battery A and battery B). Extract cross-group multi-scale features and establish a cross-group life collaborative prediction mechanism. The cross-group life collaborative prediction mechanism refers to the life collaborative prediction achieved through the parameter influence between different batteries in the same group. For example: the inhibitory effect of the temperature change of battery A on the lithium ion diffusion coefficient of battery B, the accelerated decay effect of the charge and discharge rate fluctuation of battery B on the health state (SOH) of battery A, and obtain: when the current of battery A>2A and the temperature of battery B>40℃, the life of battery A is shortened by 15%. The cross-group multi-scale features include short-term features: the transient effect of the current fluctuation of battery A on the voltage of battery B; long-term features: the cumulative effect of the temperature drift of battery B on the capacity decay of battery A.

[0066] In some embodiments, the coordination rules between different batteries are determined to obtain a cross-group coordination rule base. According to the coordination rule base, the weights of variables affected by other battery groups in the cross-group matrix are adjusted. For example, if the temperature of group B has a significant impact on the life of group A, the weight of the temperature variable in the state matrix of group A is increased.

[0067] In this embodiment, the overall life prediction of the battery group to be predicted is realized. By identifying the mutual influence relationship and mutual coordination mechanism between batteries in the battery group to be predicted, the action rules between batteries are determined. According to the action rules, the optimal life use and accurate prediction of the life are achieved.

[0068] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application realizes the overall life prediction of the battery group to be predicted, rather than the isolated prediction of a single battery, by establishing a battery group to be predicted and a cross-group matrix; it achieves the effect of accurately identifying the mutual influence and coordination mechanism between batteries, and effectively improves the overall life prediction and life optimization of multiple batteries.

[0069] Embodiment 5, as Figure 3 As shown, this embodiment also provides a battery life prediction system based on a BMS system, which is used to implement the above method. The system includes: The data acquisition and processing module is used to obtain the observation data of multiple target batteries, extract the internal electrochemical characteristic variables and external environmental variables based on the observation data, preliminarily screen out the key variable set according to the influence of each variable on the battery life prediction; and construct a state matrix group based on the key variable set; A commonality matrix generation module is used to obtain a dominant feature set according to a matrix association model and a state matrix group, and obtain a commonality matrix based on the dominant feature set; A personalized state matrix generation module is used to obtain actual observation data of the battery to be predicted, and obtain a personalized state matrix based on the actual observation data and the commonality matrix; The uncertainty analysis module is used to determine the probability distribution of each dominant feature according to the dominant feature set and the uncertainty quantification model, and generate a number of random samples according to the probability distribution of each dominant feature; input the generated random samples into the preset life cycle prediction model to obtain the battery life prediction value corresponding to each group of samples; according to the statistical characteristics of all battery life prediction values, obtain the contribution of the dominant feature to the uncertainty of the prediction result, and then calculate the confidence interval of the prediction result; The cross-group collaboration module is used to establish the battery group to be predicted, obtain the personalized state matrix of each battery to be predicted, and form a cross-group matrix; based on the cross-group matrix, a cross-group life collaborative prediction mechanism is established, the collaborative rules between different batteries are determined, and a cross-group collaborative rule base is obtained; The life prediction module is used to input the personalized state matrix and confidence interval into the preset life cycle prediction model to obtain the life prediction data of the battery to be predicted.

[0070] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0071] It should be understood by those skilled in the art that embodiments of the present invention may provide methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0075] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0076] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A battery life prediction method based on a BMS system, characterized in that: The method comprises: S100: Obtain observation data of multiple target batteries, extract internal electrochemical characteristic variables and external environmental variables based on the observation data, preliminarily screen out a key variable set according to the influence of each variable on battery life prediction; and construct a state matrix group based on the key variable set; S200: obtaining a dominant feature set based on the matrix association model and the state matrix group, and obtaining a commonality matrix based on the dominant feature set; S300: Acquire actual observation data of the battery to be predicted, and obtain a personalized state matrix based on the actual observation data and the commonality matrix; S400: Inputting the personalized state matrix into a preset life cycle prediction model to obtain life prediction data of the battery to be predicted.

2. The method for predicting battery life based on a BMS system according to claim 1, characterized in that: Step S200 also includes: S210: Obtain variable parameters in the state matrix group, and divide the variable parameters into levels; establish a hierarchical nesting model of the variable parameters, analyze the nesting relationship between variable parameters at different levels, and obtain a nesting relationship analysis result; S220: Based on the nested relationship analysis results, a coupling model between parameters at different levels is established; multi-scale features are extracted based on the coupling model, and cross-scale collaborative analysis between multi-level parameters is performed based on the multi-scale features; S230: Filter out dominant features based on the cross-scale collaborative analysis results, and combine the filtered dominant features to form a dominant feature set; The hierarchical division includes micro-level parameters, meso-level parameters and macro-level parameters.

3. The method for predicting battery life based on a BMS system according to claim 1, characterized in that: The internal electrochemical characteristic variables refer to quantifiable physical or chemical parameters that characterize the electrochemical state and reaction process of the battery, and the internal electrochemical characteristic variables include electrical characteristic parameters, chemical state parameters, thermal characteristic parameters and kinetic parameters; The external environmental variables include mode variable data and extreme variable data. The mode variable data are high-frequency environmental variables that characterize the statistics of the battery in a normal operation cycle; the extreme variable data refer to extreme environmental variables that are statistically analyzed in the full operation cycle of the battery, and are defined as the physical limits or statistical boundary values ​​of each variable. Among them, the environmental data of all operating cycles of the target battery are obtained, the environmental variables in the environmental data and the frequency of occurrence of each environmental variable are counted, the frequency of occurrence of specific variable values ​​is arranged in descending order, the selection ratio is set, and the environmental variables are selected in turn according to the selection ratio and defined as high-frequency environmental variables.

4. The method for predicting battery life based on a BMS system according to claim 1, characterized in that: Calculate the correlation coefficient between each variable and battery life, including: A multivariate linear regression model was established, with battery life as the dependent variable and each variable as the independent variable; the influence of each variable on battery life was determined by the size and significance test of the regression coefficient.

5. The method for predicting battery life based on a BMS system according to claim 1, characterized in that: The state matrix group includes a correlation matrix, a sample data matrix and a time series matrix; The correlation matrix is ​​a matrix formed according to the correlation coefficients between different key variables in the key variable set; The sample data matrix is ​​a matrix formed based on the observation data of the key variables, with the observation data as rows and the key variables as columns; The time series matrix has time points as rows and key variables as columns, and is used to analyze the changing trends of key variables over time.

6. The method for predicting battery life based on a BMS system according to claim 2, characterized in that: The nested relationship analysis result is a preliminary analysis result between parameters at different levels, which is used to reflect the clear mutual influence paths and logical associations between parameters at different levels; the cross-scale collaborative analysis result is used to indicate the specific data relationship between parameters at different levels.

7. The method for predicting battery life based on a BMS system according to claim 1, characterized in that: The method further comprises: S500: constructing an uncertainty quantification model, determining the probability distribution of each dominant feature based on the dominant feature set and the uncertainty quantification model, and generating a number of random samples according to the probability distribution of each dominant feature; S600: inputting the generated random samples into a preset life cycle prediction model to obtain a battery life prediction value corresponding to each group of samples; according to the statistical characteristics of all battery life prediction values, obtaining the contribution degree of the dominant characteristics to the uncertainty of the prediction result, and then calculating the confidence interval of the prediction result; S700: Input the confidence intervals of different dominant features and the corresponding personalized state matrix into a preset life cycle prediction model, and execute step S400.

8. The method for predicting battery life based on a BMS system according to claim 7, characterized in that: The contribution of the dominant feature to the uncertainty of the prediction result is obtained, including: performing local sensitivity analysis and global sensitivity analysis on each dominant feature to quantify the contribution; the local sensitivity analysis is based on the influence of a single dominant feature on the prediction result; the global sensitivity analysis is to decompose the variance of the prediction result into the contribution analysis of each dominant feature and its interaction.

9. The method for predicting battery life based on a BMS system according to claim 1, characterized in that: The method further comprises: S800: Establish a battery group to be predicted, wherein the battery group to be predicted includes a plurality of batteries to be predicted; obtain a personalized state matrix of each battery to be predicted to form a cross-group matrix; based on the cross-group matrix, establish a cross-group life collaborative prediction mechanism, determine the collaborative rules between different batteries, and obtain a cross-group collaborative rule base.

10. A battery life prediction system based on a BMS system, characterized in that: The system comprises: The data acquisition and processing module is used to obtain the observation data of multiple target batteries, extract the internal electrochemical characteristic variables and external environmental variables based on the observation data, preliminarily screen out the key variable set according to the influence of each variable on the battery life prediction; and construct a state matrix group based on the key variable set; A commonality matrix generation module is used to obtain a dominant feature set according to a matrix association model and a state matrix group, and obtain a commonality matrix based on the dominant feature set; A personalized state matrix generation module is used to obtain actual observation data of the battery to be predicted, and obtain a personalized state matrix based on the actual observation data and the commonality matrix; The uncertainty analysis module is used to determine the probability distribution of each dominant feature according to the dominant feature set and the uncertainty quantification model, and generate a number of random samples according to the probability distribution of each dominant feature; input the generated random samples into the preset life cycle prediction model to obtain the battery life prediction value corresponding to each group of samples; according to the statistical characteristics of all battery life prediction values, obtain the contribution of the dominant feature to the uncertainty of the prediction result, and then calculate the confidence interval of the prediction result; The cross-group collaboration module is used to establish the battery group to be predicted, obtain the personalized state matrix of each battery to be predicted, and form a cross-group matrix; based on the cross-group matrix, a cross-group life collaborative prediction mechanism is established, the collaborative rules between different batteries are determined, and a cross-group collaborative rule base is obtained; The life prediction module is used to input the personalized state matrix and confidence interval into the preset life cycle prediction model to obtain the life prediction data of the battery to be predicted.

Images