Lithium ion battery residual life prediction method based on intelligent algorithm fusion

Abstract

The invention relates to the technical field of battery management, in particular to a lithium ion battery residual life prediction method based on intelligent algorithm fusion. The method comprises the following steps: S1, collecting operation data of the lithium ion battery under different working conditions through a data monitoring system; s2, cleaning and standardizing the operation data to eliminate the influence of noise and abnormal values; s3, extracting key features capable of reflecting the health state of the battery from the operation data by using electrochemical impedance spectroscopy analysis; s4, predicting the residual life of the lithium ion battery through the fusion prediction model; and S5, carrying out error analysis according to a prediction result, and carrying out health state evaluation. Detailed information about the internal structure of the battery can be provided through the EIS, and the parameters directly reflect the aging degree of the battery. And by introducing temperature correction, the accuracy of impedance parameters is further improved, and the actual working conditions are better simulated. Multiple prediction algorithms are combined, and the prediction precision is improved by using respective advantages.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management, and more specifically, to a method for predicting the remaining useful life of a lithium-ion battery based on the fusion of intelligent algorithms. Background Art

[0002] The method for predicting the remaining useful life of a lithium-ion battery based on the fusion of intelligent algorithms is a method that comprehensively utilizes a variety of machine learning and data processing technologies to improve the prediction accuracy. This method usually combines model-based prediction and data-driven prediction. By combining the advantages of different algorithms, the limitations of a single algorithm can be overcome, thereby more accurately predicting the remaining useful life (RUL) of a lithium-ion battery. The method based on the fusion of intelligent algorithms aims to provide a more robust and accurate RUL prediction scheme for lithium-ion batteries by integrating the advantages of multiple algorithms, which is of great significance for optimizing the battery management system, extending the battery life, and enhancing the system safety.

[0003] In the existing methods for predicting the remaining useful life of a lithium-ion battery based on the fusion of intelligent algorithms, the accuracy of a single prediction algorithm is insufficient and cannot represent the remaining useful life of a lithium-ion battery; different application scenarios have different requirements for the working conditions of the battery, and most current prediction models are difficult to adapt to a wide range of changing working environments, resulting in limited generalization ability. Effective features are crucial for improving the performance of the prediction model. However, it is not easy to accurately extract the key features that contribute to predicting the remaining useful life of the battery from the original monitoring data, especially when the data contains a large amount of noise. Therefore, a method for predicting the remaining useful life of a lithium-ion battery based on the fusion of intelligent algorithms is provided. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for predicting the remaining useful life of a lithium-ion battery based on the fusion of intelligent algorithms to solve the problems proposed in the above background art, that is, in the existing methods for predicting the remaining useful life of a lithium-ion battery based on the fusion of intelligent algorithms, the accuracy of a single prediction algorithm is insufficient and cannot represent the remaining useful life of a lithium-ion battery; different application scenarios have different requirements for the working conditions of the battery, and most current prediction models are difficult to adapt to a wide range of changing working environments, resulting in limited generalization ability. Effective features are crucial for improving the performance of the prediction model. However, it is not easy to accurately extract the key features that contribute to predicting the remaining useful life of the battery from the original monitoring data, especially when the data contains a large amount of noise.

[0005] To achieve the above object, the present invention aims to provide a method for predicting the remaining useful life of a lithium-ion battery based on the fusion of intelligent algorithms, including the following steps:

[0006] S1. Collect the operation data of the lithium-ion battery under different working conditions through a data monitoring system;

[0007] S2. Clean and standardize the operation data to eliminate the influence of noise and outliers;

[0008] S3. Use electrochemical impedance spectroscopy analysis to extract key features from the operation data that can reflect the battery health status;

[0009] S4. Predict the remaining life of the lithium-ion battery through a fusion prediction model, where the fusion prediction model is a fusion model obtained by combining an LSTM model, a random forest model, and an XGBoost model;

[0010] S5. Conduct error analysis based on the prediction results and perform health status assessment.

[0011] As a further improvement of this technical solution, in S1, the operation data under different working conditions includes voltage, current, battery temperature, capacitance, and impedance at different temperatures and different charge and discharge rates.

[0012] As a further improvement of this technical solution, in S2, the specific process of cleaning and standardizing the operation data is as follows:

[0013] Outlier detection and removal:

[0014] Calculate the mean μ and standard deviation σ of the data set:

[0015]

[0016] where n represents the total number of data points; d i represents the i-th data point; i represents the index variable;

[0017] Determine the outlier range:

[0018] The outlier range is [μ - 3σ, μ + 3σ];

[0019] When |d i - μ| > 3σ, it is an outlier, and the abnormal data point d i is removed;

[0020] Use linear interpolation to fill in the missing values:

[0021] For two known data points d i-1 and d i+1 , the intermediate missing value data point d i is estimated using the following formula:

[0022]

[0023] where d i-1 represents d iPrevious data point; d i+1 Denote d i Next data point; t i Denote timestamp; t i-1 Denote previous timestamp; t i+1 Denote next timestamp;

[0024] Use Min - Max scaling;

[0025]

[0026] Wherein, X represents the original data point; X min Denote the minimum value in the dataset; X max Denote the maximum value in the dataset; X′ represents the value obtained by converting the original data X through the Min - Max scaling method to a value between 0 and 1;

[0027] Use Z - score normalization:

[0028]

[0029] Wherein, Z represents the value after Z - score normalization.

[0030] As a further improvement of this technical solution, in S3, the specific process involved in using electrochemical impedance spectroscopy analysis to extract key features that can reflect the battery health state from the operation data is:

[0031] Connect the lithium - ion battery to the electrochemical impedance spectroscopy analysis test equipment;

[0032] Perform frequency range selection;

[0033] Use the electrochemical impedance spectroscopy analysis test equipment to apply small - amplitude sine - wave voltage and current signals and measure the impedance;

[0034]

[0035] Wherein, Z(ω) represents the total impedance; R 0 Denote the ohmic resistance; R ct Denote the charge - transfer resistance; C dl Denote the double - layer capacitance; ψ represents the coefficient related to the impedance; ω represents the angular frequency; j represents the imaginary unit.

[0036] As a further improvement of this technical solution, considering that the battery temperature also affects the battery health state, introduce battery temperature to optimize the electrochemical impedance spectroscopy analysis;

[0037] The impedance parameter of the battery, ohmic resistance R 0 、Charge - transfer resistance R ct 、Double - layer capacitance Cdl It will change with temperature;

[0038]

[0039] Among them, R′ 0 (T) represents the ohmic resistance after considering the influence of battery temperature; R′ ct (T) represents the charge transfer resistance after considering the influence of battery temperature; C′ dl (T) represents the double-layer capacitance after considering the influence of battery temperature; represents the activation energy of the ohmic resistance; represents the activation energy of the charge transfer resistance; represents the activation energy of the double-layer capacitance; T represents the current working temperature of the battery; T 0 represents the reference temperature for calibration; R represents the gas constant; R 0 (T 0 ) represents the ohmic resistance of the battery measured at the reference temperature T 0 ; R ct (T 0 ) represents the charge transfer resistance measured at the reference temperature T 0 ; C dl (T 0 ) represents the double-layer capacitance measured at the reference temperature T 0 ; exp represents the abbreviation of the exponential function e x ;

[0040] The optimized total impedance formula is:

[0041]

[0042] Among them, Z(ω,T) represents the total complex impedance of the battery at the angular frequency ω and the battery temperature T.

[0043] As a further improvement of this technical solution, in the S4, the specific process involved in predicting the remaining life of the lithium-ion battery through the fusion prediction model is:

[0044] S4.1. Combine the LSTM model, the random forest model, and the XGBoost model into a fusion prediction model;

[0045] S4.2. Extract the key features and train the LSTM model, the random forest model, and the XGBoost model; among them, the key features include battery temperature, capacitance, and impedance;

[0046] S4.3. Perform model fusion through weighted averaging to obtain the final prediction result.

[0047] As a further improvement of this technical solution, in S4.2, the processes involved in training the LSTM model, random forest model, and XGBoost model are as follows:

[0048] The input feature X″ includes static and dynamic features, expressed as:

[0049] X″ = [X static , X time_series ;

[0050] Among them, X static represents static features; X time_series represents dynamic features;

[0051] The static features include battery temperature, capacitance, and impedance;

[0052] The dynamic features are the historical values in the most recent k time steps:

[0053] X time_series = {x t , x t-1 , …, x t-k+1}), x k ∈ {R 0 , R ct , C dl , T};

[0054] Among them, k represents the index variable; x t represents the feature vector at time t; t represents the time index variable; x k represents the feature vector except the time feature vector; x t-k+1 represents the feature vector of the first k + 1 time steps;

[0055] Input the dynamic features into the LSTM model:

[0056] Input gate:

[0057]

[0058] Forget gate:

[0059]

[0060] Cell gate state update:

[0061] c t = f t ⊙ c t-1 + i t ⊙ tanh(W xc x t + W hc h t-1 + b c );

[0062] Output gate:

[0063]

[0064] Hidden state update:

[0065] h t = o t ⊙ tanh(c t );

[0066] where, i t represents the state of the input gate; represents the activation function; f t represents the state of the forget gate; c t represents the state of the cell gate; o t represents the state of the output gate; h t represents the hidden state; W xi represents the weight matrix input to the input gate; W xf represents the weight matrix input to the forget gate; W xc represents the weight matrix input to the cell gate; W xo represents the weight matrix input to the output gate; W hi represents the weight matrix from the hidden state to the input gate; W hf represents the weight matrix from the hidden state to the forget gate; W hc represents the weight matrix from the hidden state to the cell gate; W ho represents the weight matrix from the hidden state to the output gate; b i represents the bias term of the input gate; b f represents the bias term of the forget gate; b c represents the bias term of the cell gate; b o represents the bias term of the output gate; tanh represents the hyperbolic tangent activation function; h t represents the hidden state; h t-1 represents the hidden state at the previous time step t - 1; c t-1 represents the cell gate state at the previous time step t - 1; ⊙ represents element-wise multiplication;

[0067] Considering that it is not easy to extract the key information of the battery health state change from static and dynamic features, a health index variable is introduced to optimize the LSTM model;

[0068] Then the updated input gate:

[0069]

[0070] The updated forget gate:

[0071]

[0072] Updated unit gate status update:

[0073] c t ′ = f t ⊙ c t-1 ′ + i t ′ ⊙ tanh(W xc [x t ; hi t + W hc h t-1 + b c );

[0074] Updated output gate:

[0075]

[0076] Updated hidden state update:

[0077] h t ′ = o t ′ ⊙ tanh(c t ′);

[0078] Among them, i t ′ represents the status of the updated input gate; f t ′ represents the status of the updated forget gate; c t ′ represents the status of the updated unit gate; o t ′ represents the status of the updated output gate; h t ′ represents the updated hidden state; c t-1 ′ represents the status of the unit gate at the previous time step t - 1;

[0079] The LSTM output is calculated through a fully connected layer for the remaining life:

[0080]

[0081] Among them, represents the predicted remaining service life; W out represents the weight matrix of the final output; h T represents the hidden state vector of the last time step of the LSTM model; b out represents the bias term of the final output;

[0082] Input the static features into the random forest model:

[0083] The random forest consists of multiple decision trees;

[0084] Each tree generates a predicted value based on the input features;

[0085] The prediction of a single decision tree is:

[0086]

[0087] Among them, represents the prediction result of a single decision tree; f tree represents the prediction function of the decision tree;

[0088] Considering that the temperature and humidity in the environment also affect the battery life, temperature and humidity variables are introduced to optimize the random forest model;

[0089]

[0090] Among them, represents the prediction result of a single decision tree after introducing temperature and humidity variables; E represents the temperature and humidity variables;

[0091] The predicted value of the random forest is the average of the predicted values of all decision trees:

[0092]

[0093] Among them, represents the overall prediction result of the random forest; N represents the number of decision trees in the random forest model; represents the predicted value of the r-th decision tree; r represents the r-th decision tree;

[0094] Input the static features into the XGBoost model:

[0095] XGBoost uses a weighted tree model and iteratively trains to generate M weighted decision trees;

[0096] The prediction of the m-th tree is:

[0097]

[0098] Among them, represents the predicted value of the m-th decision tree; f m represents the prediction function of the m-th decision tree;

[0099] The model prediction is the weighted sum of all trees:

[0100]

[0101] Among them, represents the overall prediction result of the XGBoost model; α m represents the weight of the m-th decision tree; M represents the number of decision trees in the XGBoost model; m represents the m-th decision tree.

[0102] As a further improvement of this technical solution, in step S4.3, the specific process of obtaining the final prediction result through weighted average for model fusion is as follows:

[0103] Use the weighted average method to synthesize the prediction results of different models:

[0104]

[0105] w LSTM +w RF +w XGB = 1;

[0106] where w LSTM represents the weight of the LSTM model; w RF represents the weight of the random forest model; w XGB represents the weight of the XGBoost model; represents the prediction result after fusion.

[0107] As a further improvement of this technical solution, in step S5, the specific steps involved in error analysis based on the prediction result and health status assessment are as follows:

[0108] S5.1. Calculate the error index according to the prediction result;

[0109] S5.2. Define the health status index;

[0110] where the defined health status index is:

[0111]

[0112] where SOH represents the health status score; I Z represents the Zth health index; w Z represents the weight of the Zth health index; Z represents the index variable; z represents the total number of health indices.

[0113] As a further improvement of this technical solution, in step S5.1, the specific process of calculating the error index according to the prediction result is as follows:

[0114] Mean Square Error:

[0115]

[0116] where MSE represents the Mean Square Error; represents the actual remaining life;

[0117] Mean Absolute Error:

[0118]

[0119] Among them, MAE represents the mean absolute error;

[0120] Relative error:

[0121]

[0122] Among them, RE represents the relative error.

[0123] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0124] 1. In the method for predicting the remaining life of a lithium-ion battery based on the fusion of intelligent algorithms, by collecting data under different temperatures, charge-discharge rates and other conditions, the performance changes of the battery under various actual operating environments can be captured. By removing noise and outliers, the consistency and accuracy of the data are ensured, and misleading results are avoided. The stability and accuracy of model training are improved, and the prediction error caused by data quality problems is reduced.

[0125] 2. In the method for predicting the remaining life of a lithium-ion battery based on the fusion of intelligent algorithms, EIS can provide detailed information about the internal structure of the battery, and these parameters directly reflect the aging degree of the battery. By introducing temperature correction, the accuracy of the impedance parameters is further improved, and the actual working conditions are better simulated. Combining multiple prediction algorithms and utilizing their respective advantages can improve the prediction accuracy. By comparing with actual data, the accuracy of the prediction model is evaluated, potential problems are identified and improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0126] Figure 1 It is the overall method flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0127] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0128] Embodiment:

[0129] Please refer to Figure 1 As shown, this embodiment provides a method for predicting the remaining life of a lithium-ion battery based on the fusion of intelligent algorithms, including the following steps:

[0130] S1. Collect the operation data of the lithium-ion battery under different working conditions through a data monitoring system;

[0131] In this embodiment, the operating data under different working conditions include voltage, current, battery temperature, capacitance, and impedance at different temperatures and different charge-discharge rates.

[0132] Specifically, in the method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion, collecting the operating data of the lithium-ion battery under different working conditions (such as voltage, current, battery temperature, capacitance, and impedance at different temperatures and charge-discharge rates) through a data monitoring system is a crucial first step. These multi-dimensional data provide comprehensive health status information of the battery under various actual operating conditions, laying a foundation for subsequent data preprocessing, feature extraction, and model training. Specifically, these rich operating data help accurately capture the changing trend of battery performance over time, identify key factors affecting battery life, thereby improving the accuracy and reliability of remaining life prediction. In this way, the degree of battery aging can be more effectively evaluated, and maintenance measures can be taken in a timely manner to extend the battery service life.

[0133] S2. Clean and standardize the operating data to eliminate the influence of noise and outliers;

[0134] In this embodiment, the specific process of cleaning and standardizing the operating data is as follows:

[0135] Outlier detection and removal:

[0136] Calculate the mean μ and standard deviation σ of the data set:

[0137]

[0138] where n represents the total number of data points; d i represents the i-th data point; i represents the index variable;

[0139] Determine the outlier range:

[0140] The outlier range is [μ - 3σ, μ + 3σ];

[0141] When |d i - μ| > 3σ, it is an outlier, and the abnormal data point d i is removed;

[0142] Use linear interpolation to fill in the missing values:

[0143] For two known data points d i-1 and d i+1 , the missing value data point d i in the middle is estimated by the following formula:

[0144]

[0145] Among them, d i-1 represents the previous data point of d i ; d i+1 represents the next data point of d i ; t i represents the timestamp; t i-1 represents the previous timestamp; t i+1 represents the next timestamp;

[0146] Use Min - Max scaling;

[0147]

[0148] Among them, X represents the original data point; X min represents the minimum value in the dataset; X max represents the maximum value in the dataset; X′ represents the value obtained by converting the original data X through the Min - Max scaling method to a value between 0 and 1;

[0149] Use Z - score normalization:

[0150]

[0151] Among them, Z represents the value after Z - score normalization.

[0152] Specifically, through steps such as outlier detection and removal, linear interpolation to fill in missing values, and using Min - Max scaling and Z - score normalization, the influence of noise and outliers can be effectively eliminated, the bias and variance in the data can be reduced, making subsequent feature extraction and model training more robust. These processing steps help to accurately capture the changing trend of battery performance, improve the performance of the remaining life prediction model, and ultimately achieve more accurate battery health state assessment and remaining life prediction. This can not only improve the credibility of the prediction results, but also provide reliable data support for the battery management system, extend the battery life and ensure its safe operation.

[0153] S3. Use electrochemical impedance spectroscopy analysis to extract key features from the operation data that can reflect the battery health state;

[0154] In this embodiment, the specific process involved in using electrochemical impedance spectroscopy analysis to extract key features from the operation data that can reflect the battery health state is as follows:

[0155] Connect the lithium - ion battery to the electrochemical impedance spectroscopy analysis test equipment;

[0156] Perform frequency range selection;

[0157] Use the electrochemical impedance spectroscopy analysis test equipment to apply small - amplitude sine - wave voltage and current signals and measure the impedance;

[0158]

[0159] Among them, Z(ω) represents the total impedance; R 0 represents the ohmic resistance; R ct represents the charge transfer resistance; C dl represents the double-layer capacitance; ψ represents the coefficient related to impedance; ω represents the angular frequency; j represents the imaginary unit.

[0160] Specifically, in the method for predicting the remaining life of a lithium-ion battery based on the fusion of intelligent algorithms, the specific process of using electrochemical impedance spectroscopy (EIS) analysis to extract key features from operation data that can reflect the battery health state includes connecting the battery to the EIS device, selecting the frequency range, applying a small-amplitude sine wave signal, and measuring the impedance. The total impedance obtained through this step can accurately capture the changes in key parameters such as the internal ohmic resistance, charge transfer resistance, and double-layer capacitance of the battery. These parameters directly reflect the aging degree and health state of the battery, providing an important basis for subsequent data processing and model training, thus significantly improving the accuracy and reliability of the remaining life prediction. Specifically, EIS analysis helps to identify early aging signs, optimize maintenance strategies, and extend the service life of the battery.

[0161] In this embodiment, considering that the battery temperature also affects the battery health state, the battery temperature is introduced to optimize the electrochemical impedance spectroscopy analysis;

[0162] The impedance parameters of the battery, the ohmic resistance R 0 , the charge transfer resistance R ct , and the double-layer capacitance C dl will change with the temperature;

[0163]

[0164]

[0165] Among them, R′ 0 (T) represents the ohmic resistance after considering the influence of the battery temperature; R′ ct (T) represents the charge transfer resistance after considering the influence of the battery temperature; C′ dl (T) represents the double-layer capacitance after considering the influence of the battery temperature; represents the activation energy of the ohmic resistance; represents the activation energy of the charge transfer resistance; represents the activation energy of the double-layer capacitance; T represents the current working temperature of the battery; T 0 represents the reference temperature for calibration; R represents the gas constant; R 0 (T 0) represents the ohmic resistance of the battery measured at the reference temperature T 0 ; R ct (T 0 ) represents the charge transfer resistance measured at the reference temperature T 0 ; C dl (T 0 ) represents the double-layer capacitance measured at the reference temperature T 0 ; exp represents the abbreviation of the exponential function e x ;

[0166] The optimized total impedance formula is:

[0167]

[0168] where Z(ω,T) represents the total complex impedance of the battery at the angular frequency ω and the battery temperature T;

[0169] where the key features include the battery temperature, capacitance, and impedance.

[0170] Specifically, in the method for predicting the remaining life of a lithium-ion battery based on the fusion of intelligent algorithms, the battery temperature is introduced to optimize the electrochemical impedance spectroscopy (EIS) analysis. By correcting the temperature dependence of the ohmic resistance, charge transfer resistance, and double-layer capacitance, the total impedance formula can more accurately reflect the health state of the battery under actual working conditions. This optimization step can capture the influence of internal physical and chemical changes in the battery on the impedance parameters, significantly improving the accuracy and reliability of the remaining life prediction. Specifically, the extraction of key features considering the temperature effect helps to identify early aging signs, optimize maintenance strategies, and extend the service life of the battery, thus providing more reliable data support for the battery management system.

[0171] S4. Predict the remaining life of the lithium-ion battery through the fusion prediction model;

[0172] In this embodiment, the specific process involved in predicting the remaining life of the lithium-ion battery through the fusion prediction model is:

[0173] S4.1. Combine the LSTM model, random forest model, and XGBoost model into a fusion prediction model;

[0174] S4.2. Extract the key features and train the LSTM model, random forest model, and XGBoost model;

[0175] In this embodiment, the process involved in training the LSTM model, random forest model, and XGBoost model is:

[0176] The input features X″ include static and dynamic features, expressed as:

[0177] X″ = [X static , X time_series ;

[0178] Among them, X static represents static features; X time_series represents dynamic features;

[0179] Static features include battery temperature, capacitance, and impedance;

[0180] Historical values of dynamic features in the most recent k time steps:

[0181] X time_series = {x t , x t-1 , …, x t-k+1}, x k ∈ {R 0 , R ct , C dl , T};

[0182] Among them, k represents an index variable; x t represents the feature vector at time t; t represents the time index variable; x k represents the feature vector excluding the time feature vector; x t-k+1 represents the feature vector of the first k + 1 time steps;

[0183] Input the dynamic features into the LSTM model:

[0184] Input gate:

[0185]

[0186] Forget gate:

[0187]

[0188] Cell gate state update:

[0189] c t = f t ⊙ c t-1 + i t ⊙ tanh(W xc x t + W hc h t-1 + b c );

[0190] Output gate:

[0191]

[0192] Hidden state update:

[0193] ht = o t ⊙tanh(c t );

[0194] Where, i t represents the state of the input gate; represents the activation function; f t represents the state of the forget gate; c t represents the state of the cell gate; o t represents the state of the output gate; h t represents the hidden state; W xi represents the weight matrix input to the input gate; W xf represents the weight matrix input to the forget gate; W xc represents the weight matrix input to the cell gate; W xo represents the weight matrix input to the output gate; W hi represents the weight matrix from the hidden state to the input gate; W hf represents the weight matrix from the hidden state to the forget gate; W hc represents the weight matrix from the hidden state to the cell gate; W ho represents the weight matrix from the hidden state to the output gate; b i represents the bias term of the input gate; b f represents the bias term of the forget gate; b c represents the bias term of the cell gate; b o represents the bias term of the output gate; tanh represents the hyperbolic tangent activation function; h t represents the hidden state; h t-1 represents the hidden state at the previous time step t - 1; c t-1 represents the cell gate state at the previous time step t - 1; ⊙ represents element-wise multiplication;

[0195] Considering that it is not easy to extract key information on the change of battery health state from static and dynamic features, a health index variable is introduced to optimize the LSTM model;

[0196] Then the updated input gate:

[0197]

[0198] The updated forget gate:

[0199]

[0200] The update of the updated cell gate state:

[0201] c t ' = f t ⊙c t-1 '+ i t '⊙tanh(Wxc [x i ; hi t +W hc h t-1 +b c )

[0202] Updated output gate:

[0203]

[0204] Updated hidden state update:

[0205] h t ' = o t ' ⊙ tanh(c t ')

[0206] where i t ' represents the state of the updated input gate; f t ' represents the state of the updated forget gate; c t ' represents the state of the updated cell gate; o t ' represents the state of the updated output gate; h t ' represents the updated hidden state; c t-1 ' represents the state of the cell gate at the previous time step t - 1

[0207] The LSTM output is calculated through a fully connected layer for the remaining life:

[0208]

[0209] where represents the predicted remaining service life; W out represents the weight matrix of the final output; h T represents the hidden state vector at the last time step of the LSTM model; b out represents the bias term of the final output

[0210] Input the static features into the random forest model:

[0211] The random forest consists of multiple decision trees;

[0212] Each tree generates a predicted value based on the input features;

[0213] The prediction of a single decision tree is:

[0214]

[0215] where represents the prediction result of a single decision tree; f tree represents the prediction function of the decision tree

[0216] Specifically, f tree An example of the prediction function is:

[0217]

[0218] Among them, X 1 , X 2 , X 3 represent assumed features; y 1 -y 6 represents the predicted value of the leaf node;

[0219] Considering that the temperature and humidity in the environment also affect the battery life, temperature and humidity variables are introduced to optimize the random forest model;

[0220]

[0221] Among them, represents the prediction result of a single decision tree after introducing temperature and humidity variables; E represents the temperature and humidity variable;

[0222] The predicted value of the random forest is the average of the predicted values of all decision trees:

[0223]

[0224] Among them, represents the overall prediction result of the random forest; N represents the number of decision trees in the random forest model; represents the predicted value of the r-th decision tree; r represents the r-th decision tree;

[0225] Input the static features into the XGBoost model:

[0226] XGBoost uses a weighted tree model and iteratively trains to generate M weighted decision trees;

[0227] The prediction of the m-th tree is:

[0228]

[0229] Among them, represents the predicted value of the m-th decision tree; f m represents the prediction function of the m-th decision tree;

[0230] Specifically, f m The prediction function is:

[0231]

[0232] Among them, represents the predicted value of this leaf node; X 1 -X k represents random features;

[0233] The model prediction is the weighted sum of all trees:

[0234]

[0235] where represents the overall prediction result of the XGBoost model; α m represents the weight of the m-th decision tree; M represents the number of decision trees in the XGBoost model; m represents the m-th decision tree.

[0236] Specifically, in the method for predicting the remaining life of a lithium-ion battery based on the fusion of intelligent algorithms, the LSTM model, the random forest model, and the XGBoost model each play unique roles to improve the accuracy and reliability of the prediction. Specifically, the LSTM model processes dynamic features (such as time series data) to capture the changing trend of the battery health state over time, and is particularly suitable for processing data with time dependence; the random forest model is good at processing static features (such as battery temperature, capacitance, and impedance), and enhances the stability and accuracy of the model by integrating multiple decision trees, while considering the impact of environmental factors on the battery life; the XGBoost model uses a weighted tree model for iterative training, can efficiently process complex non-linear relationships, and further improves the prediction performance by optimizing the weight allocation. The combination of the three forms a powerful hybrid model that can more comprehensively and accurately predict the remaining service life of lithium-ion batteries.

[0237] S4.3. Perform model fusion through weighted averaging to obtain the final prediction result.

[0238] In this embodiment, the specific process of performing model fusion through weighted averaging to obtain the final prediction result is as follows:

[0239] Use the method of weighted averaging to synthesize the prediction results of different models:

[0240]

[0241] w LSTM +w RF +w XGB = 1;

[0242] where w LSTM represents the weight of the LSTM model; w RF represents the weight of the random forest model; w XGB represents the weight of the XGBoost model; represents the prediction result after fusion.

[0243] Specifically, in the method for predicting the remaining life of lithium-ion batteries based on the fusion of intelligent algorithms, model fusion through weighted averaging can integrate the advantages of three models, namely LSTM, random forest, and XGBoost, to obtain more accurate and reliable prediction results. Specifically, this method assigns corresponding weights according to the performance of each model to ensure that the model with better performance plays a greater role in the final prediction, while satisfying the constraint that the sum of the weights is 1. This can not only make up for the deficiencies of a single model but also effectively improve the stability and accuracy of the prediction results, thereby more accurately evaluating the remaining service life of lithium-ion batteries.

[0244] Each of the three models receives and processes different types of input features, and at the same time, they are divided in terms of the modeling capabilities for static features, dynamic features, and non-linear relationships:

[0245] The LSTM model processes dynamic features and inputs dynamic time series features X time-series ={x t ,x t-1 ,…,x t-k+1}, capturing the time dependence of the battery health state.

[0246] The LSTM model further combines health indicator variables to optimize the output using an improved gating mechanism

[0247] The output is the prediction result of the time series trend

[0248] The random forest model processes static features and inputs static features X static (including temperature, capacitance, impedance, etc.), combines with environmental variables E for modeling, and uses the integrated decision tree method to predict the current state of the battery.

[0249] The output is the prediction result optimized based on static features and environmental variables

[0250] The XGBoost model processes the complex non-linear relationships of static features, learns the complex non-linear relationships in X static through a weighted tree model, combines with feature importance evaluation to optimize the decision tree structure, and outputs the prediction result

[0251] Among them, w LSTM gives priority to the prediction ability of time series features, and its weight is proportional to the time length k of dynamic features and the strength of the time correlation of the prediction. For example, when the time series k is long and the historical trend has a significant impact, a larger weight (w LSTM =0.5) is assigned. w RFThe modeling ability that combines static features and environmental variables, and its weight is proportional to the importance of static features and the influence intensity of environmental factors. If the environmental variable has a significant impact on battery life, a medium weight (w RF = 0.3) can be assigned. w XGB is mainly used to capture complex non-linear relationships in static features. The weight assignment is related to the feature complexity and is applicable to the case where the non-linear feature relationship is strong (w XGB = 0.2).

[0252] S5. Conduct error analysis based on the prediction results and perform health status assessment.

[0253] In this embodiment, the specific steps involved in conducting error analysis based on the prediction results and performing health status assessment are as follows:

[0254] S5.1. Calculate the error index based on the prediction results;

[0255] In this embodiment, the specific process of calculating the error index based on the prediction results is as follows:

[0256] Mean Square Error:

[0257]

[0258] where MSE represents the Mean Square Error; represents the actual remaining life;

[0259] Mean Absolute Error:

[0260]

[0261] where MAE represents the Mean Absolute Error;

[0262] Relative Error:

[0263]

[0264] where RE represents the Relative Error.

[0265] S5.2. Define the health status index;

[0266] Among them, the defined health status index is:

[0267]

[0268] where SOH represents the health status score; I Z represents the Zth health index; w Z represents the weight of the Zth health index; Z represents the index variable; z represents the total number of health indices.

[0269] Specifically, by comparing the prediction results with the actual data, a quantitative error assessment is provided, which helps to confirm the accuracy and reliability of the model and provides a direction for further optimizing the model. Through multi-dimensional health parameter assessment, a comprehensive health status score is generated, which not only reflects the current health status of the battery but also provides key information for formulating effective maintenance and management strategies. These two steps together ensure the comprehensiveness and reliability of the prediction method, making the remaining life prediction of the battery more accurate and practical.

[0270] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion, characterized in that: The following steps are involved: S1. Collecting the operating data of lithium-ion batteries under different working conditions through a data monitoring system; S2. Clean and standardize the operating data to eliminate the influence of noise and outliers; S3, using electrochemical impedance spectroscopy to extract key features that reflect the health status of the battery from the operating data; S4. Predicting the remaining life of the lithium-ion battery by using a fusion prediction model, wherein the fusion prediction model is a fusion model obtained by combining an LSTM model, a random forest model, and an XGBoost model; S5. Perform error analysis based on the prediction results and conduct health status assessment.

2. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 1 is characterized in that: In S1, the operating data under different working conditions include voltage, current, battery temperature, capacitance and impedance at different temperatures and different charge and discharge rates.

3. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 1 is characterized in that: In S2, the specific process of cleaning and standardizing the operation data is as follows: Outlier detection and removal: Calculate the mean μ and standard deviation σ of the data set: Where n is the total number of data points; d i represents the i-th data point; i represents the index variable; Determine the range of outliers: The range of outliers is [μ_3σ,μ+3σ]; When |d i -μ|>3σ, then it is an outlier, and the abnormal data point d i Remove; Use linear interpolation to fill missing values: For two known data points d i-1 and d i+1 , the middle missing value data point d i Use the following formula to estimate: Among them, d i-1 Indicates d i Previous data point; d i+1 Indicates d i The next data point; t i Indicates timestamp; t i-1 Indicates the previous timestamp; t i+1 Indicates the next timestamp; Use Min-Max scaling; Where X represents the original data point; X min Represents the minimum value in the data set; X max Represents the maximum value in the data set; X′ means converting the original data X into a value between 0 and 1 through the Min-Max scaling method; Use Z-score normalization: Among them, Z represents the standardized value of Z-score.

4. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 1 is characterized in that: In S3, the specific process of extracting key features that can reflect the health status of the battery from the operating data using electrochemical impedance spectroscopy analysis is as follows: Connecting the lithium-ion battery to an electrochemical impedance spectroscopy test device; Perform frequency range selection; Use electrochemical impedance spectroscopy to analyze the test equipment and apply a small amplitude sine wave voltage and current signal to measure the impedance; Where Z(ω) represents the total impedance; R0 represents the ohmic resistance; R ct represents the charge transfer resistance; C dl represents double-layer capacitance; ψ represents a coefficient related to impedance; ω represents an angular frequency; and j represents an imaginary unit.

5. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 4 is characterized in that: Considering that battery temperature also affects the health status of the battery, battery temperature optimization electrochemical impedance spectroscopy analysis is introduced; The impedance parameters of the battery are ohmic resistance R0, charge transfer resistance R ct , double layer capacitance C dl It will change with temperature; Where R′0(T) represents the ohmic resistance after considering the influence of battery temperature; R′ ct (T) represents the charge transfer resistance after considering the influence of battery temperature; C′ dl (T) represents the double-layer capacitance after considering the influence of battery temperature; represents the activation energy of the ohmic resistor; represents the activation energy of the charge transfer resistance; represents the activation energy of the double-layer capacitor; T represents the current operating temperature of the battery; T0 represents the reference temperature used for calibration; R represents the gas constant; R0(T0) represents the ohmic resistance of the battery measured at the reference temperature T0; R ct (T0) represents the charge transfer resistance measured at the reference temperature T0; C dl (T0) represents the double layer capacitance measured at the reference temperature T0; exp represents the exponential function e x abbreviation of; The optimized total impedance formula is: Where Z(ω, T) represents the total complex impedance of the battery at angular frequency ω and battery temperature T.

6. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 1 is characterized in that: In S4, the specific process involved in predicting the remaining life of the lithium-ion battery by using the fusion prediction model is as follows: S4.

1. Combine the LSTM model, random forest model and XGBoost model into a fusion prediction model. S4.

2. Extract key features and train the LSTM model, random forest model and XGBoost model; S4.

3. Perform model fusion through weighted averaging to obtain the final prediction result.

7. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 6 is characterized in that: In S4.2, the process involved in training the LSTM model, the random forest model and the XGBoost model is: The input features X″ include static and dynamic features, expressed as: X″=[X static ,X time_series ]; Among them, X static Indicates static characteristics; X time_series Represents dynamic features; Static characteristics include battery temperature, capacitance, and impedance; The historical values ​​of dynamic features in the last k time steps: X time_series ={x t ,x t-1 ,...,x t-k+1 }),x k ∈{R0,R ct ,C dl ,T}; Where k represents the index variable; x t represents the feature vector at time t; t represents the time index variable; x k represents the eigenvector except the time eigenvector; x t-k+1 Represents the feature vector of the first k+1 time steps; Input dynamic features into the LSTM model: Input Gate: Forget Gate: Unit door status update: c t =f t ⊙c t-1 +i t ⊙tanh(W xc x t +W hc h t-1 +b c ); Output Gate: Hide status updates: h t =o t ⊙tanh(c t ); Among them, i t Indicates the state of the input gate; represents the activation function; f t Indicates the state of the forget gate; c t Indicates the unit door status; o t Indicates the state of the output gate; h t Indicates hidden state; W xi Represents the weight matrix input to the input gate; W xf Represents the weight matrix input to the forget gate; W xc Represents the weight matrix input to the unit gate; W xo Represents the weight matrix from input to output gate; W hi Represents the weight matrix from hidden state to input gate; W hf Represents the weight matrix from hidden state to forget gate; W hc Represents the weight matrix from hidden state to unit gate; W ho Represents the weight matrix from hidden state to output gate; b i represents the bias term of the input gate; b f represents the bias term of the forget gate; b c represents the bias term of the unit gate; b o represents the bias term of the output gate; tanh represents the hyperbolic tangent activation function; h t Indicates hidden state; h t-1 represents the hidden state of the previous time step t-1; c t-1 represents the state of the unit gate at the previous time step t-1; ⊙ represents element-by-element multiplication; Considering that it is not easy to extract key information about battery health status changes from static and dynamic features, health indicator variables are introduced to optimize the LSTM model; Then the updated input gate is: Updated forget gate: Updated unit door status update: c t ′=f t ⊙c t-1 ′+i t ′⊙tanh(W xc [x t ;hi t ]+W hc h t-1 +b c ); Updated output gate: Updated hidden status update: h t ′=o t ′⊙tanh(c t ′); Among them, i t ′ represents the updated state of the input gate; f t ′ represents the state of the updated forget gate; c t ′ represents the updated unit gate state; o t ′ represents the state of the updated output gate; h t ′ represents the updated hidden state; c t-1 ′ represents the updated unit gate state of the previous time step t-1; The LSTM output passes through the fully connected layer to calculate the remaining lifespan: in, Represents the predicted remaining useful life; W out Represents the weight matrix of the final output; h T Represents the hidden state vector of the last time step of the LSTM model; b out Represents the bias term of the final output; Input static features to the random forest model: Random forests consist of multiple decision trees; Each tree generates a prediction value based on the input features; The prediction of a single decision tree is: in, Represents the prediction result of a single decision tree; f tree Represents the prediction function of the decision tree; Considering that the temperature and humidity in the environment also affect the battery life, the temperature and humidity variables are introduced to optimize the random forest model; in, It represents the prediction result of a single decision tree after introducing temperature and humidity variables; E represents temperature and humidity variables; The prediction value of the random forest is the average of the prediction values ​​of all decision trees: in, Represents the overall prediction result of the random forest; N represents the number of decision trees of the random forest model; represents the predicted value of the r-th decision tree; r represents the r-th decision tree; Input static features to the XGBoost model: XGBoost uses a weighted tree model and iterative training to generate M weighted decision trees; The prediction of the mth tree is: in, represents the predicted value of the mth decision tree; f m Represents the prediction function of the mth decision tree; The model prediction is the weighted sum of all trees: in, Represents the overall prediction result of the XGBoost model; α m represents the weight of the mth decision tree; M represents the number of decision trees in the XGBoost model; m represents the mth decision tree.

8. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 7 is characterized in that: In S4.3, the specific process of performing model fusion by weighted averaging to obtain the final prediction result is as follows: Use the weighted average method to combine the prediction results of different models: IN LSTM +W RF +W XGB =1; Among them, w LSTM Represents the weight of the LSTM model; W RF Represents the weight of the random forest model; W XGB Represents the weight of the XGBoost model; Represents the prediction result after fusion.

9. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 1 is characterized in that: In S5, the specific steps involved in performing error analysis based on the prediction results and health status assessment are: S5.

1. Calculate the error index based on the prediction results; S5.

2. Define health status indicators; Among them, the health status indicators are defined as: Among them, SOH represents the health status score; I Z represents the Zth health indicator; w Z represents the weight of the Zth health indicator; Z represents the index variable; z represents the total number of health indicators.

10. The method for predicting the remaining life of a lithium-ion battery based on intelligent algorithm fusion according to claim 9, characterized in that: In S5.1, the specific process of calculating the error index according to the prediction result is: Mean Square Error: Among them, MSE represents mean square error; Indicates the actual remaining life; Mean absolute error: Among them, MAE means mean absolute error; Relative error: Among them, RE represents relative error.

Images