Lithium battery health state estimation method and system

Abstract

The invention discloses a lithium battery health state estimation method and system. Belongs to the technical field of lithium batteries, and comprises the following steps: firstly, collecting single lithium battery data under the condition of continuous charging and discharging, then introducing an adaptive factor to improve Kalman filtering to obtain adaptive Kalman filtering, and using the improved adaptive Kalman filtering to carry out missing value filling on the collected data to improve data integrity. Secondly, improving an information acquisition optimization algorithm by using a quadratic interpolation method and an elite random reverse learning strategy to obtain an IIAO, optimizing hyper-parameters of a time sequence analysis model LightTS model by using the IIAO, inputting the processed lithium battery data into the optimized LightTS model for prediction to obtain a preliminary prediction result, and carrying out prediction on the preliminary prediction result; and finally, performing error correction on an error between the preliminary prediction result and the true value by using an online sequential reduced order kernel extreme learning machine (OS-RKELM) to obtain an error prediction value, and finally, superposing the preliminary prediction value and the error prediction value to obtain a final lithium battery health state prediction value.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium batteries, and relates to a method and system for estimating the health state of a lithium battery. Background Art

[0002] Traditional fuels are gases emitted after incomplete combustion of fossil fuels, which are converted into mechanical energy by internal combustion engines. Their gases are toxic and cause increasingly serious air pollution. Moreover, fossil energy is a one-time non-renewable energy source. Due to the shortage of fossil energy resources and the increasingly serious environmental pollution, the demand for renewable energy has been growing rapidly.

[0003] At present, the development of renewable energy has become a consistent strategic choice for energy transformation; clean energies such as solar energy and wind energy are affected by the environment and have instability and volatility; lithium-ion batteries have low noise during operation, a wide operating temperature range, a long charge-discharge cycle life, a high energy density, and do not affect the environment during use, and are widely used in the storage of clean energy. Summary of the Invention

[0004] In view of the above problems, the object of the present invention is to propose a method and system for estimating the health state of a lithium battery, which can improve the estimation accuracy of the SOC of the lithium battery.

[0005] The technical solution of the present invention is as follows: A method for estimating the health state of a lithium battery according to the present invention; the operation steps are as follows:

[0006] Step (1): Compose a training set and a test set from the voltage, current, sampling time, and capacity data of a battery cell under continuous charge and discharge conditions;

[0007] Step (2): Introduce an adaptive factor to improve the Kalman filter, increase the Kalman gain, obtain an adaptive Kalman filter, use the adaptive Kalman filter to fill in missing values for the collected data, discretize the functional relationship between battery model parameters and the expression of the life of a single lithium battery according to the basic principle of the extended Kalman filter algorithm, obtain the state equation and observation equation of the system, and use the extended Kalman filter algorithm to iteratively calculate the SOH of the lithium battery;

[0008] Step (3): Improve the information acquisition optimization algorithm, and use quadratic interpolation and an elite random reverse learning strategy to improve the first stage and the second stage of the algorithm, improve the global search ability of the algorithm, and obtain the improved information acquisition optimization algorithm IIAO;

[0009] Step (4): Use the method of deep learning to predict long - sequence time - series prediction, construct a LightTS model, based on the MLP structure, map the number of rows, and at the same time use the optimized information acquisition to optimize the model hyperparameters to obtain the best hyperparameters, improve the prediction effect of the model, and obtain a preliminary prediction result;

[0010] Step (5): Construct an SOH error time - series as the input matrix, use the OS - RKELM time - series model to correct the error between the initial prediction result obtained in step (4) and the true value, and finally superimpose the error prediction value and the preliminary prediction value to obtain the final prediction result.

[0011] Further, in step (2), the adaptive Kalman filter data - processing method includes the following steps:

[0012] Step (2.1): According to the basic principle of Kalman filter, discretize the functional relationship between battery model parameters and the expression of the single - cell lithium - battery life to obtain the state equation and observation equation of the system, and use the Kalman filter algorithm to perform iterative calculations on the SOH of the lithium - battery;

[0013] The essence of the Kalman (KF) data - processing algorithm is to find the estimated value of x under the minimum mean - square error k of Its characteristic is to calculate the current x by a recursive method k , and the data storage volume of the estimation process is relatively small, which is convenient for the system to perform fast real - time update processing on the data;

[0014] Let the equations of the dynamic system and the observation equation be respectively:

[0015] x k =A k-1,k x k-1 +B k-1,k f k-1 +w k-1

[0016] e k =H k x k +v k

[0017] In the formula, x k is the system state vector at time k, A k-1,k is the state - transition matrix from time k - 1 to time k, B k-1,k is the input system control matrix from time k - 1 to time k, voltage A k-1,k and B k-1,k can change with the change of time k, F is the observation vector at time k, H k is the observation matrix at time, wk is the dynamic noise at time k, v k is the observation noise at time k; usually follows the N(0, 1) distribution and is assumed to be Gaussian white noise;

[0018] Step (2.2): If the estimated state, dynamic noise, system observation, and observation noise satisfy the conditional assumptions of the equations in step

[0019] (2.1), then the observation x k at time k The solution process is as follows:

[0020] The specific process of the Kalman filter algorithm is as follows:

[0021] Pre - estimation of the state of health (SOH) of lithium - ion batteries:

[0022]

[0023] Calculate the covariance matrix of the pre - estimated SOH of lithium - ion batteries:

[0024]

[0025] Calculate the Kalman gain matrix of the SOH of lithium - ion batteries:

[0026]

[0027] Update the SOH estimate of lithium - ion batteries:

[0028]

[0029] Calculate the covariance matrix of the updated SOH estimate of lithium - ion batteries:

[0030]

[0031] Recursive loop calculation:

[0032]

[0033] Given the initial values x 0 and P 0 , according to the observed value e k at time k, the state estimate at time k can be recursively calculated Repeating each step can recursively calculate for state prediction;

[0034] Step (2.3): The KF calculation is divided into a time update process and a measurement update process;

[0035] Step (2.4): The final output lithium - ion battery dataset includes sampling time, current, voltage, charging capacity, and discharging capacity.

[0036] Further, in step (2.3), the KF calculation is divided into a time update process and a measurement update process, which are specifically as follows:

[0037] Step (2.3.1): Time update process:

[0038] and P k-1 are the initial estimates;

[0039] (a), Project the state variable forward:

[0040]

[0041] (b), Project the error covariance forward:

[0042]

[0043] Step (2.3.2): Measurement update process. Here, an adaptive factor is introduced to improve the Kalman filter and increase the Kalman gain;

[0044] (a), The calculation formula for the Kalman gain after introducing the adaptive factor is as follows:

[0045]

[0046] (b), Update e from the observed variable k Estimate:

[0047]

[0048] (c), The error covariance formula for the improved adaptive Kalman filter is:

[0049]

[0050] Further, in step (3), the information acquisition phase and the information filtering and evaluation phase of the IAO algorithm are improved by using the quadratic interpolation method and the elite stochastic reverse learning strategy to obtain the improved information acquisition optimization algorithm IIAO, including the following steps:

[0051] Step (3.1): In the information acquisition phase, individuals collect information from different sources using various methods, which can actually be regarded as a differential evolution, and its formula is:

[0052]

[0053] In the formula, died represents the current iteration number, and θ is a random number between [0, 1], which is used to characterize the influencing factors in the information collection process;

[0054] Step (3.2): The information screening and evaluation stage is crucial for an individual to quickly identify valuable and relevant information. This mechanism not only effectively excludes inaccurate and misleading information but also significantly improves the overall quality of information. Its formula is expressed as:

[0055]

[0056] In the formula, rand is a random number generated in [0,1], and Δ is the error generated when subjective factors filter and evaluate information. The expression of Δ is as follows:

[0057]

[0058] In the formula, Γ represents the subjective influence factor, which is a quantitative index representing individual subjectivity. The calculation formula of Γ is as follows:

[0059]

[0060] In the formula, Max_died is the maximum number of iterations. The mathematical model of Γ consists of three main parts: the sine function part, the logarithmic function part, and the information quality factor φ. The formula for the information quality factor φ is:

[0061]

[0062] In the formula, φ represents a function of the number of iterations died, and it affects the quality of information acquisition by adjusting the phase and amplitude of the cosine function. In this way, the IAO algorithm can simulate the adaptive adjustment behavior of an individual when facing different information, so as to achieve optimal information acquisition.

[0063] Step (3.3): The information analysis and organization stage is to identify the existing useful information from the filtered information and convert the convertible information identified in the previous stage into useful information. Its formula is expressed as:

[0064]

[0065] In the formula, represents the best information body generated in the previous iteration process, represents the average value of the best information body generated in the previous iteration process. Λ represents the control factor for analyzing and organizing information, which is defined as:

[0066]

[0067] During the process of information organization and analysis, individuals can dynamically adjust the depth and breadth of analysis based on the quality and reliability of information, thereby significantly improving the accuracy and practicality of information. This adaptive adjustment mechanism is reflected in that by dynamically adjusting the value of the reliability factor Γ, individuals can correspondingly adjust the rigor and detail of information processing. When the value of Γ is high, it indicates that the information is more reliable, so the depth and breadth Λ of analysis increase accordingly, prompting the system to conduct more in-depth information mining and more extensive information integration. Conversely, when the value of Γ is low, it indicates that the information is less reliable, and the value of Λ will also decrease accordingly, meaning that more cautious information analysis is required to avoid relying on low-quality data, thus ensuring the efficiency and accuracy of information processing.

[0068] Step (3.4): To effectively improve the global search ability of the information acquisition optimization algorithm and overcome the problem of its easy entrapment in local optimal solutions, an elite stochastic opposition-based learning strategy and a random factor are added in the information filtering and evaluation stage of the information acquisition optimization algorithm to further enhance the diversity of the population.

[0069] The new information quality factors are calculated reversely based on the calculated information quality factors, and then these new and old information quality factors are compared uniformly, and the ones with better quality factors are selected for retention. The calculation formula is as follows:

[0070] X new = α(X max + X min ) - rX old (10)

[0071] In the formula, X max and X min are the upper and lower limits of the dynamic boundary, r is a random number; by using the dynamic boundary, the search history of information filtering and evaluation can be saved, thereby reducing the convergence time of the entire algorithm; while the random number r enhances the diversity of the population; α is a scaling factor, which is adjusted to adapt to different optimization problems.

[0072] Step (3.5): To improve the convergence speed and local exploration ability of the algorithm, the quadratic interpolation method is introduced in the information collection stage.

[0073] Quadratic interpolation is a commonly used interpolation method, usually used for quadratic function curve estimation between three given data points. The formula for the quadratic polynomial is as follows:

[0074]

[0075] In the formula, A, B, and C are data points, and f(A), f(B), and f(C) are fitness values.

[0076] The secondary interpolation information collection is updated to perform secondary interpolation update every three individuals. The collected information is updated by two - foot interpolation to generate new information, and then compared with the original information to select more excellent information. This method enhances the global search ability of the algorithm.

[0077] The improved information acquisition optimization algorithm can better optimize the parameters of the LightTs model and improve the prediction of the health state of lithium batteries by the LightTs model.

[0078] Furthermore, in step (4), the construction of the LightTs model includes the following steps:

[0079] Step (4.1): Establish the LightTs model, where the multivariate time - series prediction value is defined as:

[0080] X t′ ={x t′C+1 ,...,x t′-1 ,x t′ |x y ∈R z}

[0081] Multi - step prediction value:

[0082] τ={X t′+1 ,...,X t′-G}

[0083] In the formula, t′ is time, C is the time - window length, Z is the number of time series, and G is the prediction range;

[0084] Step (4.2): Perform continuous and sampling; The continuous sampling method is to convert each time series with length C into several non - overlapping subsequences with length M, and mark M as a subsequence; For an input sequence X t′ ∈R C ,sample it into the subsequence to obtain a 2D matrix where the y - th column is:

[0085]

[0086] Interval sampling method:

[0087]

[0088] Step (4.3): The basic building - block information - exchange block (IEBlock) uses a bottleneck design;

[0089] Step (4.4): LightTS training process: Use IEBlocks to extract the time features in the first part C and map them to a feature vector of dimension U; then connect the features of continuous sampling and interval sampling in the time dimension in the first part with all time series variables in the channel dimension to obtain an input matrix of shape R 2U×Nα ; finally, obtain an IEBlock-C of shape , whose output is the final prediction.

[0090] Furthermore, in step (4.3), the specific steps for the basic building block information exchange block to use the bottleneck design are as follows:

[0091] Step (4.3.1): The input two-dimensional matrix is:

[0092] Z α =(z oy )H α ×W

[0093] The i-th column and the j-th row are respectively represented as:

[0094]

[0095] z y =(z y1 ,z y2 ,...,z yW )

[0096] Apply an MLP on each column, where U' << U, then the time projection is:

[0097]

[0098] Step (4.3.2): Extract features from the time projection along the time dimension, use weight sharing for all columns, and apply an R W →R W MLP on each row, then the channel projection is:

[0099]

[0100] Step (4.3.3): The channel projection keeps the input shape unchanged, exchanges information between different channels, uses weight sharing for all rows, and thus applies another R U' →R U MLP on each column to map the feature dimension from U' to U, then the output projection is:

[0101]

[0102] Step (4.3.4): When the input time series is long, to save computational costs, an improved bottleneck design is used; first, a temporal projection is used to map the number of rows from H α to U for mapping; then, a channel projection of U' time is applied to communicate information; finally, an output projection is used to map the feature U' to the desired hyperparameter output length U.

[0103] Furthermore, in step (4), the hyperparameters of the LightTs model are optimized by adopting an improved information acquisition optimization algorithm, including the following steps:

[0104] Initialize the relevant parameters of the IIAO algorithm, including initializing the particle positions, the upper and lower limits of the search space, the maximum number of iterations, and the current number of iterations;

[0105] Input the training set of lithium battery capacitance data into the LightTs model to optimize the hyperparameters of the model;

[0106] After the gradual optimization by the IIAO algorithm, calculate the fitness value of each particle. When the preset termination condition of the model is met, determine the best particle position, which is the optimal value of the hyperparameters of the LightTs model;

[0107] The optimized LightTs model performs SOH prediction on the lithium battery data processed by Kalman filtering to obtain the preliminary lithium battery SOH prediction result.

[0108] Furthermore, in step (5), the OS-RKELM model is used to correct the prediction error. The error prediction process of the OS-RKELM model includes the following steps:

[0109] Step (5.1): Use a part of the initial training data from the given training set to initialize the model, and randomly select a subset from it where N 0 represents the number of samples;

[0110] Step (5.2): Calculate the initial kernel matrix K 0 :

[0111] K 0 = к(τ 0 ,τ L )

[0112] The formula for the initial output weight SH (0) is as follows:

[0113]

[0114] where

[0115] Step (5.3): Calculate Partial data block

[0116]

[0117] Step (5.4): Kernel matrix of partial data block The calculation formula is as follows:

[0118]

[0119] Step (5.5): Calculate the output weight

[0120]

[0121] Step (5.6): Use the OS-RKELM model to correct the prediction error.

[0122] Furthermore, in step (5.6), the operation steps of using the OS-RKELM model to correct the prediction error are as follows:

[0123] Step (5.6.1): Subtract the initial prediction result from the true value at time z to obtain the error time series, and its calculation method is as follows:

[0124] Q z = R dz - S pz

[0125] In the formula, Q z represents the error value at time z, R dz represents the true value at time z, S pz represents the initial predicted value at time z;

[0126] Step (5.6.2): Construct the error time series of the lithium battery as the input matrix;

[0127] Step (5.6.3): Use the OSRKELM model to correct the error of the lithium battery error time series to obtain the lithium battery error prediction value, and add the initial prediction result and the error prediction result to obtain the final prediction result, and its calculation method is as follows:

[0128]

[0129] In the formula, represents the final predicted value of the lithium battery at time z, Y dz represents the initial prediction result of the lithium battery at time z, δ zIndicates the error prediction result.

[0130] Furthermore, a lithium battery state of health estimation system includes a data acquisition and processing unit that collects various data during the charging and discharging processes of the lithium battery through multiple sensors and stores them according to the number of charge-discharge cycles; the collected data is filled using an adaptive Kalman filter to further improve the data set and construct the input data for the lithium battery state of health.

[0131] The algorithm improvement unit improves the information acquisition optimization algorithm using the quadratic interpolation method and the elite stochastic reverse learning strategy to enhance the global search ability of the algorithm and improve the efficiency of optimizing the model parameters.

[0132] The model optimization unit optimizes the parameters of the LighTS model using the improved information acquisition optimization algorithm; through the improved information acquisition optimization algorithm, a balance is maintained between local search and global search to optimize the parameters of the LighTS model and improve the accuracy and stability of the prediction model.

[0133] The error correction unit subtracts the observed value from the initial prediction value, inputs the obtained error value into the OSRKELM model for error correction, and superimposes the obtained error-corrected prediction value and the initial prediction value to obtain the final lithium battery state of health prediction result.

[0134] The beneficial effects of the present invention are as follows: 1. In view of the characteristics of the lithium battery capacitance data, such as correlation and large data volume, which will interfere with the prediction accuracy of the prediction model, the Kalman filter with an introduced adaptive factor is used to fill in the missing data values of the historical lithium battery continuous charging and discharging data, improving the data integrity and enhancing the prediction accuracy of the model for the data; 2. Aiming at the limitations of the traditional information acquisition optimization algorithm, the quadratic interpolation method and the elite stochastic reverse learning strategy are used to improve the IAO algorithm, enhancing the global search ability and obtaining a better convergence effect; 3. The present invention uses the improved IIAO algorithm to optimize the parameters of the LightTS model to find the parameters with the highest prediction accuracy for the lithium battery state of health. This self-parameter optimization method improves the performance of the model, enables the model to better fit the lithium battery data, enhances the generalization ability of the model, and improves the prediction accuracy and reliability; 4. Aiming at the problem that a single model is prone to large errors, a variant model based on the extreme learning machine is used for error correction prediction. The online sequential reduced-order kernel extreme learning machine can perform online sequential learning compared with the extreme learning machine, process data one by one, and has a faster learning rate and stronger generalization performance, making the prediction of the model more accurate. Therefore, the OSRKELM corrects the prediction error, which can further improve the prediction accuracy of the lithium battery state of health. Description of the Drawings

[0135] Figure 1 is the operation flow chart of the present invention;

[0136] Figure 2 is the schematic diagram of the Kalman filter calculation process in the present invention;

[0137] Figure 3 is the schematic diagram of the improved IAO algorithm (IIAO) process in the present invention;

[0138] Figure 4 is the error correction flow chart in the present invention. Specific embodiments

[0139] The following further elaborates on the specific technical solutions of the present invention in combination with specific examples.

[0140] As shown in the figure, the present invention provides a method for estimating the health state of a lithium battery. First, the data of a single lithium battery under continuous charge and discharge is collected. Then, an adaptive factor is introduced to improve the Kalman filter, increasing the Kalman gain of the filter to obtain an adaptive Kalman filter. The adaptive Kalman filter is used to fill in the missing values in the collected data to improve data integrity. Next, the information acquisition optimization algorithm is improved using the quadratic interpolation method and the elite stochastic reverse learning strategy to obtain IIAO, and the hyperparameters of the time series analysis model LightTS model are optimized using IIAO. The processed lithium battery data is input into the optimized LightTS model for prediction to obtain a preliminary prediction result. Finally, the online sequential reduced-order kernel extreme learning machine (OS-RKELM) is used to correct the error between the preliminary prediction result and the true value to obtain an error prediction value. Finally, the preliminary prediction value and the error prediction value are superimposed to obtain the final predicted value of the health state of the lithium battery;

[0141] The specific operation steps are as follows:

[0142] Step (1): Collect the lithium battery single-cell voltage, current, sampling time, and capacity data under continuous charge and discharge to form a training set and a test set;

[0143] Step (2): Introduce an adaptive factor to improve the Kalman filter, increasing the Kalman gain to obtain an adaptive Kalman filter. Use the adaptive Kalman filter to fill in the missing values in the collected data. According to the basic principle of the extended Kalman filter algorithm, discretize the functional relationship between the battery model parameters and the expression of the single-cell lithium battery life to obtain the state equation and the observation equation of the system, and use the extended Kalman filter algorithm to perform iterative calculations on the SOH of the lithium battery;

[0144] Among them, the adaptive Kalman filter data processing method is as follows:

[0145] Step (2.1): Based on the basic principle of Kalman filtering, the functional relationship between the battery model parameters and the expression of the life of a single lithium battery are discretized to obtain the state equation and observation equation of the system, and the Kalman filtering algorithm is used to iteratively calculate the SOH of the lithium battery;

[0146] The essence of the Kalman (KF) data processing algorithm is to find the minimum mean square error x k Estimated value of Its characteristic is to calculate the current x by recursive method. k ,The storage volume of the estimation process data is relatively small, which facilitates the system to quickly update and process the data in real time;

[0147] Assume that the equations of the dynamic system and the observation equations are:

[0148] x k =A k-1,k x k-1 +B k-1,k f k-1 +w k-1

[0149] e k =H k x k +v k

[0150] In the formula, x k is the system state vector at time k, A k-1,k is the state transition matrix from time k-1 to time k, B k-1,k is the input system control matrix from time k-1 to time k, the voltage A k-1,k and B k-1,k It can change with the change of time k. F is the observation vector at time k, and H k is the observation matrix at the moment, w k is the dynamic noise at time k, v k is the observation noise at time k; usually it follows N(0,1) distribution and is assumed to be Gaussian white noise;

[0151] Step (2.2): If the estimated state and dynamic noise as well as the system observation and observation noise satisfy the conditional assumptions of the equations in step (2.1), then the observation x at time k is k Estimates The solution process is as follows;

[0152] The specific process of the Kalman filter algorithm is as follows:

[0153] Lithium battery SOH estimation:

[0154]

[0155] Calculate the covariance matrix of the preliminary estimation of the state of health (SOH) of a lithium battery:

[0156]

[0157] Calculate the Kalman gain matrix of the SOH of a lithium battery:

[0158]

[0159] Update the estimation of the SOH of a lithium battery:

[0160]

[0161] Calculate the covariance matrix of the updated SOH estimation of a lithium battery:

[0162]

[0163] Recursive loop calculation:

[0164]

[0165] At the given initial values \(x\) 0 and \(P\) 0 and based on the observation value \(e\) at time \(k\) k , the state estimation at time \(k\) can be recursively calculated. Repeating each step allows for recursive calculation for state prediction;

[0166] Step (2.3): The KF calculation is divided into a time update process and a measurement update process, as follows:

[0167] Step (2.3.1): Time update process:

[0168] and \(P\) k-1 are the initial estimations;

[0169] (a)、Predict the state variable forward:

[0170]

[0171] (b)、Predict the error covariance forward:

[0172]

[0173] Step (2.3.2): Measurement update process, where an adaptive factor is introduced to improve the Kalman filter and increase the Kalman gain;

[0174] (a)、The formula for the Kalman gain after introducing the adaptive factor is as follows:

[0175]

[0176] (b), Update e from the observed variables k Estimation:

[0177]

[0178] (c), The error covariance formula of the improved adaptive Kalman filter is:

[0179]

[0180] Step (2.3): By comparing the experimental data of step (2.22) and step (2.3), it can be judged that the Kalman filter with an adaptive factor has a better processing effect on the collected lithium battery data. Therefore, the finally output lithium battery data set includes sampling time, current, voltage, charging capacity, and discharging capacity;

[0181] Step (3): Improve the information acquisition stage and information filtering and evaluation stage of the IAO algorithm by using the quadratic interpolation method and the elite random reverse learning strategy to obtain the improved information acquisition optimization algorithm IIAO, including the following steps:

[0182] Step (3.1): In the information acquisition stage, individuals use various methods to collect information from different sources, which can actually be regarded as a differential evolution, and its formula is:

[0183]

[0184] In the formula, died represents the current iteration number, and θ is a random number between [0,1], which is used to characterize the influencing factors in the information collection process;

[0185] Step (3.2): The information screening and evaluation stage is crucial for individuals to quickly identify valuable and relevant information. This mechanism not only effectively excludes inaccurate and misleading information, but also significantly improves the overall quality of information, and its formula is expressed as:

[0186]

[0187] In the formula, rand is a random number generated in [0,1], and Δ is the error generated when the subjective factor filters and evaluates the information. The expression of Δ is as follows:

[0188]

[0189] In the formula, Γ represents the subjective influence factor, which is a quantitative index representing the individuality of an individual. The calculation formula of Γ is as follows:

[0190]

[0191] In the formula, Max_died is the maximum number of iterations. The mathematical model of Γ consists of three main parts: the sine function part, the logarithmic function part, and the information quality factor φ. The formula for the information quality factor φ is as follows:

[0192]

[0193] In the formula, φ represents a function of the iteration number died, which affects the quality of information acquisition by adjusting the phase and amplitude of the cosine function. In this way, the IAO algorithm can simulate the adaptive adjustment behavior of individuals when facing different information, thus achieving optimal information acquisition.

[0194] Step (3.3): The information analysis and organization stage is to identify the existing useful information from the filtered information and convert the convertible information identified in the previous stage into useful information. Its formula is expressed as:

[0195]

[0196] In the formula, represents the best information body generated in the previous iteration process, represents the average value of the best information bodies generated in the previous iteration process. Λ represents the control factor for analyzing and organizing information, which is defined as:

[0197]

[0198] In the process of information organization and analysis, individuals can dynamically adjust the depth and breadth of analysis according to the quality and reliability of information, thereby significantly improving the accuracy and practicality of information. This adaptive adjustment mechanism is reflected in that by dynamically adjusting the value of the reliability factor Γ, individuals can correspondingly adjust the rigor and detail of information processing. When the value of Γ is high, it indicates that the reliability of the information is strong, so the depth and breadth Λ of the analysis increase accordingly, prompting the system to conduct more in-depth information mining and more extensive information integration. On the contrary, when the value of Γ is low, it indicates that the reliability of the information is low, and the value of Λ will also decrease accordingly, meaning that more cautious information analysis is required to avoid relying on low-quality data, thus ensuring the efficiency and accuracy of information processing.

[0199] Step (3.4): In order to effectively improve the global search ability of the information acquisition optimization algorithm and overcome the problem of its easy entrapment in local optimal solutions, an elite random reverse learning strategy and a random factor are added to the information filtering and evaluation stage of the information acquisition optimization algorithm to further enhance the diversity of the population.

[0200] Calculate the new information quality factor by reverse calculation according to the calculated information quality factor, and then compare these new and old information quality factors uniformly, and select the one with a better quality factor for retention. The calculation formula is as follows:

[0201] X new = α(X max + X min ) - rX old

[0202] where X max and X min are the upper and lower limits of the dynamic boundary, r is a random number;

[0203] Using the dynamic boundary, the search history of information filtering and evaluation can be saved, thereby reducing the convergence time of the entire algorithm, while the random number r enhances the diversity of the population; α is a scaling factor, adjusted to adapt to different optimization problems;

[0204] Step (3.5): To improve the convergence speed and local exploration ability of the algorithm, the quadratic interpolation method is introduced in the information acquisition stage; quadratic interpolation is a commonly used interpolation method, usually used for quadratic function curve estimation between three given data points; the formula for the quadratic polynomial is as follows:

[0205]

[0206] where A, B, C are data points, and f(A), f(B), f(C) are fitness values;

[0207] The quadratic interpolation information acquisition is updated by performing quadratic interpolation update every three individuals, performing two-foot interpolation update on the acquired information to generate new information, and then comparing it with the old information to select more excellent information; this method enhances the global search ability of the algorithm;

[0208] The improved information acquisition optimization algorithm can better optimize the parameters of the LightTs model and improve the prediction of the health state of lithium batteries by the LightTs model;

[0209] Step (4): Construct the LightTs model, and the specific steps are as follows:

[0210] Step (4.1): Establish the LightTs model, where the multivariate time series prediction value is defined:

[0211] X t′ = {x t′-C+1 ,..., x t′-1 , x t′ |x y ∈ R z}

[0212] Multi-step prediction value:

[0213] τ = {X t′+1 ,..., Xt′-G}

[0214] where t′ is the time, C is the length of the time window, Z is the number of time series, and G is the prediction range;

[0215] Step (4.2): Conduct continuous and sampling; The continuous sampling method is to convert each time series of length C into several non-overlapping subsequences of length M; Mark M as a subsequence;

[0216] For an input sequence X t′ ∈R C , sample it into subsequences to obtain a 2D matrix where the y-th column is:

[0217]

[0218] Interval sampling method

[0219]

[0220] Step (4.3): The basic building block information exchange block (IEBlock) uses a bottleneck design:

[0221] Step (4.3.1): The input two-dimensional matrix is:

[0222] Z α =(z oy )H α ×W

[0223] The i-th column and the j-th row are respectively expressed as:

[0224]

[0225] z y′ =(z y1 ,z y2 ,...,z yW )

[0226] Apply an MLP on each column, where U′ << U, then the time projection is:

[0227]

[0228] Step (4.3.2): Extract features along the time dimension for the time projection, share weights for all columns, and apply an MLP from R W →R W on each row, then the channel projection is:

[0229]

[0230] Step (4.3.3): The channel projection keeps the input shape unchanged, exchanges information between different channels, and uses weight sharing for all rows, so as to apply another R U' →R U The MLP of maps the feature dimension from U' to U, and the output projection is:

[0231]

[0232] Step (4.3.4): When the input time series is long, in order to save computational cost, an improved bottleneck design is used; first, a time projection is used to project the number of rows from H α to U for mapping; then, the time of U' of the channel projection is used to communicate information; finally, the output projection is used to map the feature U' to the required hyperparameter output length U;

[0233] Step (4.4): The LightTS training process; use IEBlocks to extract the time features in the first part C and map them to feature vectors with dimension U; then, connect the features of continuous sampling and interval sampling in the time dimension in the first part with all time series variables in the channel dimension to obtain an input matrix with the shape of R 2U×Nα Finally, the IEBlock-C with the shape of is obtained, and its output is the final prediction;

[0234] In step (4), the hyperparameters of the LightTs model are optimized by using the improved information acquisition optimization algorithm, including the following steps:

[0235] Initialize the relevant parameters of the IIAO algorithm, including initializing the particle position, the upper and lower limits of the search space, the maximum number of iterations, and the current number of iterations;

[0236] Input the training set of lithium battery capacitance data into the LightTs model to optimize the hyperparameters of the model;

[0237] After the step-by-step optimization of the IIAO algorithm, calculate the fitness value of each particle. When the preset termination condition of the model is met, determine the best particle position, which is the optimal value of the hyperparameters of the LightTs model;

[0238] The optimized LightTs model performs SOH prediction on the lithium battery data processed by Kalman filtering to obtain the preliminary lithium battery SOH prediction result;

[0239] In step (5), the OS-RKELM model is used to correct the prediction error. The error prediction process of the OS-RKELM model includes the following steps:

[0240] Step (5.1): Initialize the model using a portion of the initial training data from the given training set, and randomly select a subset from it where N denotes the number of samples; 0

[0241] Step (5.2): Calculate the initial kernel matrix K 0 :

[0242] K 0 = κ(τ 0 , τ L )

[0243] The formula for the initial output weight SH (0) is as follows:

[0244]

[0245] where

[0246] Step (5.3): Calculate partial data blocks

[0247]

[0248] Step (5.4): The kernel matrix of the partial data block is calculated as follows:

[0249]

[0250] Step (5.5): Calculate the output weights

[0251]

[0252] Step (5.6): Use the OS-RKELM model to correct the prediction error, error correction;

[0253] Specifically, it includes the following steps:

[0254] Step (5.6.1): Use the OS-RKELM model to correct the error of the lithium battery health state:

[0255] Step (5.6.2): Subtract the initial prediction result from the true value at time z to obtain the error time series, and its calculation method is as follows:

[0256] Q z = R dz - S​pz

[0257] In the formula, Q z represents the error value at time z, R dz represents the true value at time z, S pz represents the initial predicted value at time z;

[0258] Step (5.6.3): Construct the error time series of the lithium battery as the input matrix;

[0259] Step (5.6.4): Use the OSRKELM model to correct the error of the lithium battery error time series to obtain the error prediction value of the lithium battery. Add the initial prediction result and the error prediction result to obtain the final prediction result. The calculation method is as follows:

[0260]

[0261] In the formula, represents the final predicted value of the lithium battery at time z, Y dz represents the initial prediction result of the lithium battery at time z, δ z represents the error prediction result.

[0262] Furthermore, a lithium battery state of health estimation system includes:

[0263] A data acquisition and processing unit collects various data during the charging and discharging process of the lithium battery through multiple sensors and stores them according to the number of cycles; performs data filling on the collected data using adaptive Kalman filtering to further improve the data set and construct the input data for the lithium battery state of health;

[0264] An algorithm improvement unit improves the information acquisition optimization algorithm using quadratic interpolation and elite stochastic opposition-based learning strategy to improve the global search ability of the algorithm and enhance the efficiency of optimizing the model parameters;

[0265] A model optimization unit uses the improved information acquisition optimization algorithm to optimize the parameters of the LighTS model. Through the improved information acquisition optimization algorithm, maintain a balance between local search and global search, optimize the parameters of the LighTS model, and improve the accuracy and stability of the prediction model;

[0266] An error correction unit subtracts the observed value from the initial predicted value, inputs the obtained error value into the OSRKELM model for error correction, and superimposes the obtained error correction prediction value and the initial predicted value to obtain the final lithium battery state of health prediction result.

[0267] The system feeds back the health information to the technical personnel according to the final lithium battery state of health prediction result through the interface, providing support for subsequent lithium battery maintenance.

[0268] This invention conducts research on the prediction of the health state of lithium batteries by using the adaptive Kalman filtering method, improving the IAO algorithm to optimize the parameters of the LightTS model, and introducing the error correction technology. This method only needs to obtain a large amount of historical lithium battery data and decompose these data to evaluate the performance of the battery. This method has a faster response speed and higher prediction accuracy, and can effectively predict the health state of lithium batteries.

Claims

1. A method for estimating the health status of a lithium battery, characterized in that: The operation steps are as follows: Step (1): The voltage, current, sampling time and capacity data of the battery cells under continuous charge and discharge conditions are used to form a training set and a test set; Step (2): introducing an adaptive factor to improve the Kalman filter to obtain an adaptive Kalman filter, using the adaptive Kalman filter to fill missing values ​​in the collected data, discretizing the functional relationship between the battery model parameters and the expression of the life of a single lithium battery based on the basic principle of the extended Kalman filter algorithm, obtaining the state equation and observation equation of the system, and using the extended Kalman filter algorithm to iteratively calculate the SOH of the lithium battery; Step (3): Improve the information acquisition optimization algorithm by using the quadratic interpolation method and the elite random reverse learning strategy to improve the first and second stages of the algorithm, and obtain the improved information acquisition optimization algorithm IIAO; Step (4): Use deep learning methods to predict long-sequence time series predictions, build a LightTS model, map the number of rows based on the MLP structure, and use the optimized information acquisition to optimize the model hyperparameters to obtain the best hyperparameters and finally obtain preliminary prediction results; Step (5): Construct the SOH error time series as the input matrix, use the OS-RKELM time series model to correct the error between the initial prediction result obtained in step (4) and the true value, and finally superimpose the error prediction value and the preliminary prediction value to obtain the final prediction result.

2. A lithium battery health status estimation method according to claim 1, characterized in that: In step (2), the operation steps of the adaptive Kalman filter data processing method are as follows: Step (2.1): Based on the basic principle of Kalman filtering, the functional relationship between the battery model parameters and the expression of the life of a single lithium battery are discretized to obtain the state equation and observation equation of the system, and the Kalman filtering algorithm is used to iteratively calculate the SOH of the lithium battery; The essence of the KF data processing algorithm is to find the minimum mean square error x k Estimated value of It uses the recursive method to calculate the current x k ; Assume that the equations of the dynamic system and the observation equations are: x k =A k-1,k x k-1 +B k-1,k f k-1 +w k-1 e k =H k x k +v k In the formula, x k is the system state vector at time k, A k-1,k is the state transition matrix from time k-1 to time k, B k-1,k is the input system control matrix from time k-1 to time k, the voltage A k-1,k and B k-1,k changes with the time k, F is the observation vector at time k, H k is the observation matrix at the moment, w k is the dynamic noise at time k, v k is the observation noise at time k; Step (2.2): If the estimated state and dynamic noise and the system observation and observation noise satisfy the conditional assumptions of the equations in step (2.1), then the observation x at time k is k Estimates The solution process is as follows: The specific process of the Kalman filter algorithm is as follows: Lithium battery SOH estimation: Calculate the lithium battery SOH estimated covariance matrix: Calculate the lithium battery SOH Kalman gain matrix: Updated lithium battery SOH estimation: Calculate the updated lithium battery SOH estimation covariance matrix: Recursive loop calculation: Under the given initial values ​​x0 and P0, according to the observed value e at time k k , recursively calculate the state estimate at time k Repeating each step will recursively calculate and predict the state; Step (2.3): KF calculation is divided into two processes: time update process and measurement update process; Step (2.4): The final output lithium battery data set includes sampling time, current, voltage, charging capacity and discharging capacity.

3. A lithium battery health status estimation method according to claim 2, characterized in that: In step (2.3), the KF calculation is divided into two processes: time update process and measurement update process as shown below: Step (2.3.1): Time update process: and P k-1 For initial estimate (a) Forward calculation of state variables: (b) Forward error covariance: Step (2.3.2): Calculate the update process and introduce adaptive factors to improve the Kalman filter; (a) The Kalman gain calculation formula after introducing the adaptive factor is as follows: (b) Update e by observed variables k estimate: (c) The error covariance formula of the improved adaptive Kalman filter is:

4. A lithium battery health status estimation method according to claim 1, characterized in that: The operation steps of step (3) are as follows: Step (3.1): In the information collection stage, individuals use various methods to collect information from different sources, which is actually regarded as a differential evolution. The formula is: In the formula, died represents the current iteration number, and θ is a random number between [0,1], which is used to characterize the influencing factors in the information collection process; Step (3.2): Information screening and evaluation stage The formula for individuals to quickly identify valuable information is expressed as: Where rand is a random number generated in [0,1], and Δ is the error caused by subjective factors when filtering and evaluating information. The expression of Δ is as follows: In the formula, Γ represents the subjective impact factor, which is a quantitative indicator of individual subjectivity. The calculation formula of Γ is as follows: Where Max_died is the maximum number of selections. The mathematical model of Γ consists of three parts: the sine function part, the logarithmic function part, and the information quality factor φ. The formula of the information quality factor φ is: In the formula, φ represents the function of the number of iterations died, which affects the quality of information acquisition by adjusting the phase and amplitude of the cosine function; Step (3.3): The information analysis and organization stage is to identify the existing useful information from the filtered information and convert the convertible information identified in the previous stage into useful information. The formula is expressed as: In the formula, represents the best information body generated in the previous iteration, represents the average value of the best information body generated in the previous iteration, and Λ represents the control factor for analyzing and organizing information, which is defined as: When the Γ value is high, it indicates that the information is highly reliable, so the depth and breadth of the analysis Λ increase; conversely, when the Γ value is low, it indicates that the information is less reliable, and the Λ value will also decrease, indicating that the information analysis is more cautious to ensure the efficiency and accuracy of information processing; Step (3.4): In order to effectively improve the global search capability of the information acquisition optimization algorithm, an elite random reverse learning strategy and a random factor are added to the information filtering and evaluation stage of the information acquisition optimization algorithm to enhance the diversity of the population; The new information quality factor is calculated in reverse based on the calculated information quality factor, and then the new and old information quality factors are compared uniformly, and the one with better quality factor is selected for retention. The calculation formula is as follows: X new =α(X max +X min )-rX old Where, X max With X min are the upper and lower limits of the dynamic boundary, and r is a random number. The dynamic boundary is used to save the search history of information filtering and evaluation. The random number r enhances the diversity of the population. α is a scaling factor that is adjusted to adapt to different optimization problems. Step (3.5): In order to improve the convergence speed and local exploration ability of the algorithm, the quadratic interpolation method is introduced in the information collection stage; Quadratic interpolation is an interpolation method used to estimate a quadratic function curve between three given data points; the formula for the quadratic polynomial is as follows: In the formula, A, B, and C are data points, and f(A), f(B), and f(C) are fitness values; The quadratic interpolation information collection and update is performed once for every three individuals. The collected information is updated by quadratic interpolation to generate new information, which is then compared with the old information to select the better information.

5. A lithium battery health status estimation method according to claim 1, characterized in that: In step (4), the steps of constructing the LightTs model are as follows: Step (4.1): Establish the LightTs model, in which the multivariate time series forecast value is defined: X t′ ={x t′-C+1 ,...,x t′-1 ,x t′ |x y ∈R z } Multi-step forecast values: τ={X t′+1 ,...,1...,X t′-C } In the formula, t′ is time, C is the time window length, Z is the number of time series, and G is the prediction range; Step (4.2): Perform continuous sampling; the continuous sampling method is to convert each time series of length C into several non-overlapping subsequences of length M, marking M as a subsequence; for an input sequence X t′ ∈R C , sample it into subsequences Get a 2D matrix The yth column is: Interval sampling method: Step (4.3): Basic building block information exchange block uses bottleneck design; Step (4.4): LightTS training process: Use IEBlocks to extract the time features in the first part C and map them to a feature vector of dimension U; then connect the features of continuous sampling and interval sampling on the time dimension in the first part with all time series variables on the channel dimension to obtain a shape of R 2U×Nα The input matrix is IEBlock-C, whose output is the final prediction.

6. A lithium battery health status estimation method according to claim 1, characterized in that: In step (4.3), the basic building block information exchange block uses the following steps of bottleneck design: Step (4.3.1): The input two-dimensional matrix is: Z α =(z oy )H α ×W The i-th column and j-th row are represented as: With y· =(from y1 ,With y2 ,...,With yW ) Apply a MLP, where U′<<U, then the time projection is: Step (4.3.2): Project the time dimension to extract features, use weight sharing for all columns, and apply an R on each row. W →R W The MLP of , then the channel projection is: Step (4.3.3): Channel projection keeps the input shape unchanged, exchanges information between different channels, uses weight sharing for all rows, and thus applies another R on each column U' →R U The MLP maps the feature dimension from U' to U, and the output projection is: Step (4.3.4): When the input time series is long, use the improved bottleneck design; first use a time projection to convert the number of rows from H α Map it to U; then apply channel projection U' time to communicate information; finally use output projection to map the feature U' to the required hyperparameter output length U.

7. A lithium battery health status estimation method according to claim 1, characterized in that: In step (4), the process of optimizing the hyperparameters of the LightTs model using the improved information acquisition optimization algorithm is as follows: Initialize the relevant parameters of the IIAO algorithm, including the initial particle position, the upper and lower limits of the search space, the maximum number of iterations, and the current number of iterations; Input the training set of lithium battery capacitance data into the LightTs model and optimize the model's hyperparameters; After the IIAO algorithm gradually optimizes, the fitness value of each particle is calculated. When the preset termination condition of the model is met, the optimal particle position is determined, which is the optimal value of the LightTs model hyperparameters. The optimized LightTs model predicts the SOH of lithium battery data after Kalman filtering and obtains preliminary lithium battery SOH prediction results.

8. A lithium battery health status estimation method according to claim 1, characterized in that: In step (5), the prediction error is corrected using the OS-RKELM model. The operating steps of the error prediction process of the OS-RKELM model are as follows: Step (5.1): Use a portion of the initial training data from the given training set To initialize the model, randomly select a subset In the formula, N0 represents the number of samples; Step (5.2): Calculate the initial kernel matrix K0: K0=к(τ0,τ L ) Initial output weight SH (0) The formula is as follows: In the formula, Step (5.3): Calculate Partial data block Step (5.4): Kernel matrix of some data blocks The calculation formula is as follows: Step (5.5): Calculate output weights Step (5.6): Use the OS-RKELM model to correct the prediction error.

9. A lithium battery health status estimation method and system according to claim 8, characterized in that: In step (5.6), the operation steps of correcting the prediction error using the OS-RKELM model are as follows: Step (5.6.1): Subtract the initial forecast result from the true value at time z to obtain the error time series, which is calculated as follows: Q z =R dz -S pz In the formula, Q z Represents the error value at time z, R dz represents the true value at time z, S pz represents the initial prediction value at time z; Step (5.6.2): ​​Construct the lithium battery error time series as the input matrix; Step (5.6.3): Use the OSRKELM model to perform error correction on the lithium battery error time series to obtain the lithium battery error prediction value. Add the initial prediction result and the error prediction result to obtain the final prediction result. The calculation method is as follows: In the formula, represents the final predicted value of the lithium battery at time z, Y dz represents the initial prediction result of lithium battery at time z, δ z Represents the error prediction result.

10. A lithium battery health status estimation system, characterized in that: It includes a data acquisition and processing unit, an algorithm improvement unit, a model optimization unit and an error correction unit.

Images