Lithium battery remaining service life prediction method based on intelligent hybrid drive model

Abstract

The invention discloses a lithium battery remaining service life prediction method based on an intelligent hybrid drive model, and the method comprises the steps: carrying out the modeling of a long-term degradation trend of a target system through a double-index degradation model (DEDM), and fitting model parameters through a nonlinear least square method; performing dynamic learning on the residual fluctuation of the actual observation data and the degradation trend through a fuzzy random configuration network (F-SCN); and finally, integrating trend prediction and residual prediction results through Gaussian process regression (GPR) to realize high-precision residual life prediction. According to the method, through collaborative optimization of the physical model and the data-driven model, the problem that trend fitting and local fluctuation are difficult to consider under limited data in a traditional method is solved, and the prediction precision and reliability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to a method for predicting the remaining service life of a lithium battery, and in particular to a method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model. Background Art

[0002] Lithium-ion batteries are widely used in portable terminals, electric vehicles, electrochemical energy storage, aerospace and other fields due to their significant advantages such as high output voltage, long cycle life, high energy density, low self-discharge rate and wide operating temperature range. However, their internal electrochemical reactions are complex and accompanied by a large number of thermal effects. If there is a lack of effective health status supervision, it is easy to cause safety hazards. Performance degradation caused by battery aging is one of the important causes of safety accidents. Therefore, accurate prediction of the remaining useful life (RUL) and degradation trajectory of lithium-ion batteries is crucial. Through RUL prediction, scientific maintenance plans can be formulated to prevent system failures, and a theoretical basis can be provided for battery recycling, reducing the capacity drop problem caused by inconsistent degradation of cells in the battery pack, thereby improving economic benefits.

[0003] Currently, lithium-ion battery RUL prediction methods fall into three main categories: Model-based methods rely on physical or chemical mechanism modeling. While they offer high accuracy, the complex side reactions within lithium-ion batteries make modeling and parameter identification difficult. Furthermore, the models are often designed for specific battery models, resulting in poor generalization. Data-driven methods use machine learning and other algorithms to extract degradation patterns from historical data. However, their performance is highly dependent on data quality and quantity, and they require significant computing resources, making them difficult to deploy on computing-constrained battery management terminals. Hybrid methods combine the advantages of model-driven and data-driven approaches. However, existing hybrid methods require more than 40% of monitoring data to predict degradation trajectories, resulting in insufficient early-stage prediction capabilities. Furthermore, commonly used methods suffer from numerous issues, including difficulty determining the initial state. Improper initial value settings can easily lead to model divergence and significant prediction errors. In subsequent iterative predictions, due to the lack of actual observations, data-driven methods are required to generate virtual observations. However, in the early stages of degradation, the virtual observations output by data-driven algorithms tend to be constant, significantly reducing prediction accuracy. In response to the above technical problems, the present invention proposes a solution based on an intelligent hybrid drive model, aiming to improve the accuracy of lithium battery remaining life prediction and early warning capabilities. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method for predicting the remaining service life of lithium batteries based on an intelligent hybrid drive model, so as to improve the accuracy of the prediction of the remaining service life of lithium batteries and the early warning capability.

[0005] Technical solution: The present invention provides a method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model, comprising:

[0006] (1) Obtain the charge and discharge cycle data of the lithium battery and preprocess it using differential calculation and Z-score normalization method;

[0007] (2) Modeling the long-term degradation trend of lithium batteries based on a double exponential degradation model, and fitting the model parameters using a nonlinear least squares method based on preprocessed lithium battery charge and discharge cycle data;

[0008] (3) Construct a fuzzy random configuration network, take the difference sequence between the predicted value and the actual value of the double exponential degradation model, the number of charge and discharge cycles of the lithium battery, the first-order residual value and the second-order residual value as input, and predict the difference between the predicted value and the actual value of the double exponential model in the next cycle;

[0009] (4) The prediction results of the double exponential degradation model, the prediction value of the fuzzy random configuration network, the average discharge voltage and discharge temperature of the battery are quantified through Gaussian process regression to output the predicted value and confidence interval of the remaining service life of the lithium battery.

[0010] Preferably, the step (1) is to perform a differential calculation on the lithium battery capacity sequence with a lag operator of l, wherein the first-order difference is expressed as:

[0011] ΔQ t =Q t -Q t-l

[0012] The second-order difference is expressed as:

[0013] Δ 2 Q t =ΔQ t -ΔQ t-l =Q t -2Q t-l +Q t-2l

[0014] Among them, Q t is the battery capacity data of the tth cycle.

[0015] Preferably, the double exponential degradation model expression in step (2) is:

[0016]

[0017] Among them, G(t) is the model output, t is the number of charge and discharge cycles, α1, α2 are the initial degradation coefficients, and β1, β2 are the degradation rate parameters.

[0018] Preferably, the fuzzy random configuration network in step (3) includes:

[0019] (31) Input 4-dimensional features, including the difference sequence between the predicted value and the actual value of the double exponential degradation model, the number of charge and discharge cycles of the lithium battery, the first-order residual value and the second-order residual value;

[0020] (32) Fuzzy C-means clustering is used to determine the membership function center and width of the input features;

[0021] (33) Generate normalized trigger intensity through TS fuzzy rules and randomly configure hidden layer node parameters;

[0022] (34) The recursive least squares method with L2 regularization is used to solve the output weights and complete the local degradation prediction.

[0023] Preferably, the Gaussian process regression in step (4) includes:

[0024] (41) Construct a hybrid feature space that integrates the predicted value of the double exponential degradation model, the fuzzy random configuration network residual correction term, and the original capacity attenuation characteristics;

[0025] (42) Design a linear combination mean function and a square exponential kernel covariance kernel function, and optimize hyperparameters through maximum likelihood estimation;

[0026] (43) Calculate the predicted mean and variance based on the posterior distribution and generate a 95% confidence interval: CI 95％ =[μ * -1.96σ * ,μ * +1.96σ * ].

[0027] Preferably, the step (4) further includes a dynamic update mechanism, which uses incremental Cholesky decomposition to reduce the covariance matrix and update complexity, and reduces memory usage by using an induced point sparsification method.

[0028] Preferably, the method further includes hybrid uncertainty quantification, combining parameterized probability box and non-parametric kernel density estimation to characterize the random uncertainty and epistemic uncertainty of lithium batteries respectively.

[0029] Preferably, the step (41) converts the long-term trend prediction value G(t) of the double exponential degradation model and the residual correction term ΔQ of the fuzzy random configuration network into FSCN , average battery discharge voltage V avg and discharge temperature T disch Fuse into multi-dimensional input vectors and construct a hybrid feature space:

[0030] X hy =[G(t),ΔQ FSCN ,V avg ,T disch ]T

[0031] Long-term trend prediction value G(t) of the double exponential degradation model and the average battery discharge voltage V avg Perform Min-Max normalization:

[0032]

[0033] Among them, G max and G min and V max and V min are the maximum and minimum values ​​in their respective data.

[0034] A computer device includes one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and when the programs are executed by the processors, implement the steps of a method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model.

[0035] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model.

[0036] Beneficial effects: Compared with the existing technology, the present invention has the following significant advantages: (1) By integrating the long-term trend modeling of the double exponential degradation model, the dynamic fluctuation learning of the fuzzy random configuration network and the integrated prediction of Gaussian process regression, the prediction accuracy of the remaining useful life (RUL) of lithium batteries is significantly improved, and the error is reduced by 48% to 60% compared with the single model, showing stronger robustness; (2) By quantifying random and cognitive uncertainties through confidence intervals (95%) and mixed probability boxes, comprehensive risk analysis support is provided, providing a reliable basis for equipment maintenance decisions; (3) The model adopts induced point sparsification and incremental calculation technology, and the training time is shortened to 20% of the traditional method, and the memory usage is reduced by 70%, meeting the real-time health management needs of complex systems such as industrial equipment and new energy vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The figure is a flow chart of the method of the present invention.

[0038] Figure 2 This is the overall model process structure diagram of the present invention.

[0039] Figure 3 This is a diagram showing the prediction effect of the remaining service life of the lithium battery of the present invention. DETAILED DESCRIPTION

[0040] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0041] like Figure 1 As shown, the present invention provides a method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model, comprising:

[0042] (1) Obtain the charge and discharge cycle data of the lithium battery and preprocess it using differential calculation and Z-score normalization method.

[0043] In the specific implementation process, the lithium battery charge and discharge cycle data provided by NASAAmes Prognostics Center was used, including parameters such as battery capacity, charge and discharge voltage, temperature, and internal resistance;

[0044] The battery capacity series is subjected to differential calculation with the lag operator being l, and its first-order difference can be expressed as:

[0045] ΔQ t =Q t -Q t-l

[0046] Among them, Q t is the battery capacity data of the tth cycle;

[0047] The second-order difference can be expressed as:

[0048] Δ 2 Q t =ΔQ t -ΔQ t-l =Q t -2Q t-l +Q t-2l

[0049] Z-score normalization is used, and the formula is:

[0050]

[0051] Among them, μ Q is the mean capacity, σ Q is the standard deviation.

[0052] (2) The long-term degradation trend of lithium batteries is modeled based on the double exponential degradation model. The model parameters are fitted using the nonlinear least squares method based on the preprocessed lithium battery charge and discharge cycle data.

[0053] In the specific implementation process, the asymmetric attenuation characteristics applicable to the positive and negative active materials of lithium-ion batteries are adopted, and the formula is:

[0054]

[0055] Where G(t) is the model output, t is the number of charge and discharge cycles, α1 and α2 are the initial degradation coefficients, β1 and β2 (both less than zero) are the degradation rate parameters, and the initial parameters are set to α1 = 0.8, α2 = 0.2, βI = -0.001, and β2 = -0.003.

[0056] The nonlinear least squares method is used to fit the nonlinear function parameters by minimizing the sum of squares of the residuals between the observed and predicted values:

[0057]

[0058] Where F(θ,X) is a nonlinear function, X is the independent variable vector, and θ is the parameter vector to be estimated. The Levenberg-Marquardt algorithm is used to iteratively optimize the parameters. The termination condition is that the change rate of the residual sum of squares is less than 1×10 -6 Or the maximum number of iterations is 100 times.

[0059] (3) A fuzzy random configuration network is constructed, and the difference sequence between the predicted value and the actual value of the double exponential degradation model is used as input, and the number of charge and discharge cycles, first-order residual value and second-order residual value of the lithium battery are used to predict the difference between the predicted value and the actual value of the double exponential model in the next cycle.

[0060] The specific process is as follows:

[0061] (31) The input feature dimension is 4-dimensional, including the difference sequence between the double exponential model predicted value and the actual value, the number of charge and discharge cycles, the first-order residual, the second-order residual, etc. Each sample uses the two differences between the double exponential model predicted value and the actual value, the number of current cycles, the first-order residual value, and the second-order residual value to predict the difference between the double exponential model predicted value and the actual value of the next cycle.

[0062] (32) Fuzzy C-means clustering is used to determine the center and width of the membership function of the input features.

[0063] The Gaussian membership function is used to calculate the membership of the input feature to the fuzzy set:

[0064]

[0065] in, and They represent the membership function center and width of the i-th input feature in the k-th fuzzy rule, respectively. They are determined after clustering the input data through fuzzy C-means clustering. The membership degree is used to calculate the triggering strength of the rule.

[0066] (33) Generate normalized trigger intensity through TS fuzzy rules and randomly configure hidden layer node parameters;

[0067] Generate K TS fuzzy rules, and the trigger strength of each rule is:

[0068]

[0069] in, is the membership of the input feature to the fuzzy set, is the i-th dimension data of the n-th sample.

[0070] Normalization gives the normalized trigger strength:

[0071]

[0072] Where K is the total number of fuzzy rules and the normalized trigger strength is used to output the rule H. f Calculation.

[0073]

[0074] consequent parameter factor is a random number that obeys the normal distribution, d is the input dimension of each sample, is the normalized trigger strength of the nth sample. The rule output is not only the input of the configuration layer, but also connected to the output layer.

[0075] Dynamically generate the jth node parameter w through a random configuration algorithm j *,b j *, the configuration layer output is:

[0076]

[0077] The output of the jth node is:

[0078]

[0079] in, and They represent the input weight and bias of the jth node in the hidden layer, and n is the number of samples:

[0080]

[0081] Where λ1 is the scale factor and rand(d,1) is a random number between 0 and 1 in dimension d. For the Sigmoid activation function, the node parameters must satisfy the improved random constraints:

[0082]

[0083] Among them, e j-1,m is the network residual of the first j-1 nodes, u j represents a sequence of non-negative real numbers, where uj ≤1-r, choose The largest candidate node joins the network.

[0084] (34) The recursive least squares method with L2 regularization is used to solve the output weights and complete the local degradation prediction.

[0085] The outputs of the fuzzy rule layer and the enhancement layer are combined to calculate the final prediction value:

[0086] Y=[H f ,H s ][VW]==H f V+H s W=Zβ

[0087] Where Z=[H f ,H s ], the output weight β is solved by recursive least squares method with L2 regularization:

[0088] β * =(Z T Z+λ2I) -1 Z T Y

[0089] (4) The prediction results of the double exponential degradation model, the prediction value of the fuzzy random configuration network, the average discharge voltage and discharge temperature of the battery are quantified through Gaussian process regression to output the predicted value and confidence interval of the remaining service life of the lithium battery.

[0090] By constructing a hybrid feature space and using Gaussian process regression (GPR) to fuse multi-source features such as long-term trend prediction and short-term fluctuation correction results, quantify uncertainty, perform nonlinear modeling and optimize calculations, we can solve problems such as lithium battery data sensitivity to noise, poor model coordination, and insufficient real-time performance, and achieve high-precision, strong robustness and explainable remaining life prediction.

[0091] The specific process is as follows:

[0092] (41) The long-term trend prediction value G(t) of the double exponential degradation model (DEDM) and the residual correction term ΔQ of the fuzzy random configuration network (F-SCN) are converted into FSCN , average battery discharge voltage V avg and discharge temperature T disch Fuse into multi-dimensional input vectors and construct a hybrid feature space:

[0093] X hy =[G(t),ΔQ FSCN ,V avg ,T disch ] T

[0094] For G(t) and V avg Perform Min-Max normalization:

[0095]

[0096] G max , G min and V max 、V min are the maximum and minimum values ​​in their respective data.

[0097] (42) Design the linear combination mean function and RBF covariance kernel function, and optimize the hyperparameters through maximum likelihood estimation;

[0098] Using linear combination to fuse physical models and data-driven features:

[0099]

[0100] in, is the normalized long-term trend forecast value G(t), ΔQ FSCN is the residual correction term of the fuzzy random configuration network, is the normalized average discharge voltage of the battery, T disch is the discharge temperature, weight coefficient α i (i=1, 2, 3, 4) is fitted by regularized least squares method to ensure the balanced utilization of multi-source features by the model.

[0101] The combined kernel function of the square exponential kernel (RBF) and the noise term is selected:

[0102]

[0103] Where: f is the signal variance, which controls the output amplitude; l is the length scale, which determines the smoothness of the function; σ n 2 is the noise variance, quantifying the observation error; δxx′ is the Kronecker delta function (1 when X=X′, otherwise 0)

[0104] Optimize the hyperparameter θ=(σ f ,l,σ n ), the objective function is the negative log marginal likelihood:

[0105]

[0106] Where K is the covariance matrix, Y is the RUL observation value, and C is the constant term. The conjugate gradient method is used for iterative optimization to ensure global convergence.

[0107] (43) Prediction and uncertainty quantification

[0108] Given a test point X * , its posterior distribution is:

[0109]

[0110] Predicted mean:

[0111]

[0112] Prediction variance:

[0113]

[0114] Among them, K * is the covariance vector of the test point and the training data, k(X * ,X * ) is the autocovariance of the test point.

[0115] Generate a 95% confidence interval based on the predicted mean and variance:

[0116] CI 95％ =[μ * -1.96σ * ,μ * +1.96σ * ]

[0117] The uncertainty of the prediction results is quantified to provide risk warning support for maintenance decisions.

[0118] (44) Model performance optimization strategy

[0119] Incremental Cholesky decomposition is used to update the covariance matrix K, which reduces the computational complexity of online update from O(N 3 ) is reduced to O(N 2 ) to achieve real-time adaptation of model parameters.

[0120] The induction point method (FITC) is introduced, and m = 50 induction points Z are selected from the training set. The approximate covariance matrix is:

[0121]

[0122] Among them, K fu is the covariance matrix between the training points and the induction points. This strategy reduces memory usage by 70% and increases training speed by 5 times.

[0123] The parameterized probability box (p-box) and non-parametric kernel density estimation are combined to characterize the random uncertainty and epistemic uncertainty in battery degradation, respectively, to achieve a more comprehensive risk analysis.

Claims

1. A method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model, characterized in that: include: (1) Obtain the charge and discharge cycle data of the lithium battery and preprocess it using differential calculation and Z-score normalization method; (2) Modeling the long-term degradation trend of lithium batteries based on a double exponential degradation model, and fitting the model parameters using a nonlinear least squares method based on preprocessed lithium battery charge and discharge cycle data; (3) Construct a fuzzy random configuration network, take the difference sequence between the predicted value and the actual value of the double exponential degradation model, the number of charge and discharge cycles of the lithium battery, the first-order residual value and the second-order residual value as input, and predict the difference between the predicted value and the actual value of the double exponential model in the next cycle; (4) The prediction results of the double exponential degradation model, the prediction value of the fuzzy random configuration network, the average discharge voltage and discharge temperature of the battery are quantified through Gaussian process regression to output the predicted value and confidence interval of the remaining service life of the lithium battery.

2. The method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model according to claim 1, characterized in that: The step (1) is to perform differential calculation on the lithium battery capacity sequence with a lag operator of l, wherein the first-order difference is expressed as: ΔQ t =Q t -Q t-l The second-order difference is expressed as: Δ 2 Q t =ΔQ t -ΔQ t-l =Q t -2Q t-l +Q t-2l Among them, Q t is the battery capacity data of the tth cycle.

3. The method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model according to claim 1, characterized in that: The double exponential degradation model expression in step (2) is: Among them, G(t) is the model output, t is the number of charge and discharge cycles, α1, α2 are the initial degradation coefficients, and β1, β2 are the degradation rate parameters.

4. The method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model according to claim 1, characterized in that: The fuzzy random configuration network in step (3) includes: (31) Input 4-dimensional features, including the difference sequence between the predicted value and the actual value of the double exponential degradation model, the number of charge and discharge cycles of the lithium battery, the first-order residual value and the second-order residual value; (32) Fuzzy C-means clustering is used to determine the membership function center and width of the input features; (33) Generate normalized trigger intensity through TS fuzzy rules and randomly configure hidden layer node parameters; (34) The recursive least squares method with L2 regularization is used to solve the output weights and complete the local degradation prediction.

5. The method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model according to claim 1, characterized in that: The Gaussian process regression in step (4) includes: (41) Construct a hybrid feature space that integrates the predicted value of the double exponential degradation model, the fuzzy random configuration network residual correction term, and the original capacity attenuation characteristics; (42) Design a linear combination mean function and a square exponential kernel covariance kernel function, and optimize hyperparameters through maximum likelihood estimation; (43) Calculate the predicted mean and variance based on the posterior distribution and generate a 95% confidence interval: CI 95％ =[μ * -1.96σ * ,μ * +1.96σ * ].

6. The method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model according to claim 1, characterized in that: The step (4) also includes a dynamic update mechanism, which uses incremental Cholesky decomposition to reduce the covariance matrix and update complexity, and reduces memory usage through an induced point sparsification method.

7. The method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model according to claim 1, characterized in that: The method also includes hybrid uncertainty quantification, which combines parameterized probability box and non-parametric kernel density estimation to characterize the random uncertainty and epistemic uncertainty of lithium batteries respectively.

8. The method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model according to claim 5, characterized in that: The step (41) converts the long-term trend prediction value G(t) of the double exponential degradation model and the residual correction term ΔQ of the fuzzy random configuration network into FSCN , average battery discharge voltage V avg and discharge temperature T disch Fuse into multi-dimensional input vectors and construct a hybrid feature space: X hy =[G(t),ΔQ FSCN ,V avg ,T disch ] T Long-term trend prediction value G(t) of the double exponential degradation model and the average battery discharge voltage V avg Perform Min-Max normalization: Among them, G max and G min and V max and V min are the maximum and minimum values ​​in their respective data.

9. A computer device, characterized in that: The method comprises one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of a method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model as described in claims 1-8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for predicting the remaining service life of a lithium battery based on an intelligent hybrid drive model as described in claims 1-8 are implemented.

Images