Lithium ion battery remaining useful life prediction method based on dwd-svr model

Abstract

The application discloses a lithium ion battery residual service life prediction method based on a DWD-SVR model, and comprises the following steps: 1. using a discrete wavelet decomposition (DWD) method to perform multi-scale decomposition on sampled lithium ion battery capacity time sequence data; 2. respectively calculating sample entropy and permutation entropy of each subsequence obtained after decomposition, using a K-means clustering method to cluster each scale data, and reconstructing subsequences with similar complexity and randomness according to the clustering results; 3. using a multi-scale support vector regression (SVR) model to respectively predict the clustered signal components, integrating the prediction results, and realizing lithium ion battery residual service life prediction. While ensuring the accuracy of global degradation trend prediction, the application can respond to fluctuations in a timely manner, and effectively improves the prediction accuracy of the lithium ion battery residual service life.

Description

Technical Field

[0001] This invention relates to the field of lithium battery life analysis and prediction, specifically to a method for predicting the remaining life of lithium-ion batteries based on a combination of discrete wavelet decomposition and support vector regression. Background Technology

[0002] To address the dual pressures on the environment and energy sectors brought about by a growing population, society is accelerating the development of various clean energy sources to achieve carbon peaking and carbon neutrality goals as soon as possible. Lithium-ion batteries, with their ability to achieve a good balance among various performance parameters such as cost, lifespan, and power, play a crucial role in grid energy storage systems and new energy vehicles.

[0003] During application, the degradation process of lithium-ion batteries is an extremely complex electrochemical process, exhibiting time-varying and nonlinear characteristics, making accurate state assessment difficult. Therefore, lithium-ion battery capacity time series show significant local regeneration and fluctuation phenomena, displaying multi-scale characteristics. Due to the influence of complex local regeneration and fluctuations, directly using the original capacity time series for modeling makes it difficult to accurately derive the capacity degradation process. Summary of the Invention

[0004] To overcome the shortcomings of the existing methods, this invention proposes a method for predicting the remaining lifespan of lithium-ion batteries based on the DWD-SVR model, aiming to improve the accuracy and effectiveness of the prediction, thereby enabling timely planning for the operation of energy storage facilities and improving their usage.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] The method for predicting the remaining service life of lithium-ion batteries based on the DWD-SVR model of this invention is characterized by the following steps:

[0007] Step 1: Collect the capacity degradation dataset C(t) at time t during the entire life cycle N of the lithium-ion battery and perform L-level wavelet decomposition to obtain the L-th level decomposed signal C′ at time t. L (t)=A L (t)+D L，1 (t)+D L,2 (t)+...+D L,j (t)...+D L,L (t); where A L (t) represents the low-frequency component of the Lth layer at time t, D L,j (t) represents the j-th high-frequency signal in the L-th high-frequency component at time t; j = 1, 2, ..., L; let the L-th high-frequency component D L (t)={D L，1(t),D L,2 (t),…,D L,j (t),…,D L,L (t)};

[0008] Step 2: Calculate D L,j The sample entropy and permutation entropy of (t) are used to obtain the high-frequency part D of the Lth layer. L The sample entropy and permutation entropy of (t);

[0009] Step 2.1, Calculate D L,j The sample entropy SE(D) of (t) L,j (t)), thus obtaining the high-frequency part D of the Lth layer. L The sample entropy SE(D) of (t) L (t))={SE(D L,1 (t)),SE(D L,2 (t)),…,SE(D L,j (t)),…,SE(D L,L (t))}:

[0010] Step 2.1.1: Take the j-th high-frequency signal from the high-frequency component of the L-th layer during a certain period of the lithium-ion battery's entire life cycle N to form the j-th m-dimensional high-frequency vector X of the L-th layer. L,j ={D L,j (t),D L,j (t+1),…,D L,j (t+m-1)}, 1≤t≤N-m+1; m is the vector dimension;

[0011] Step 2.1.2: Calculate any two X values ​​within the time range of 1 ≤ t ≤ N-m+1. L,j The distance between them is calculated, and the number of high-frequency signal pairs with a distance less than or equal to the given tolerance r is counted, denoted as A. L,j ;

[0012] Step 2.1.3: Calculate D using equation (1) L,j The sample entropy SE(D) of (t) L,j (t));

[0013]

[0014] Step 2.2, Calculate D L,j The permutation entropy PE(D) of (t) L,j (t)), thus obtaining the high-frequency part D of the Lth layer. L The sample entropy PE(D) of (t) L (t))={PE(D L,1 (t)),PE(D L,2 (t)),…,PE(D L,j(t)),…,PE(D L,L (t))};

[0015] Step 2.2.1: Take the j-th high-frequency signal from the high-frequency component of the L-th layer during a certain lag period in the entire life cycle N of the lithium-ion battery and form the j-th n-dimensional high-frequency lag vector ζ of the L-th layer. L,j ={D L,j (t),D L,j (t+l),…,D L,j (t+(n-1)l)}, where n is the embedding dimension and l is the hysteresis;

[0016] Step 2.2.2, for ζ L,j Sort the data and obtain n! sorted results, where! represents factorial; let α be the index of any sorted result, and ψ(α) represent the frequency of the α-th sorted result among the n! sorted results. Calculate the time series D using equation (2). L,j The permutation entropy PE(D) of (t) L,j (t)):

[0017]

[0018] Step 3, according to D L The entropy and permutation entropy of each high-frequency signal sample in (t) are analyzed using the K-means clustering method for D. L Clustering is performed on the high-frequency signals in (t) to obtain the clustering results, including: the high-frequency random component D of the Lth layer at time t. L,high (t) and the low-frequency fluctuation component D of the Lth layer at time t L,low (t); and the low-frequency component A of the Lth layer at time t. L (t) is used as the global degradation trend component, thus obtaining the high-frequency random component dataset under the entire lifecycle N. Low-frequency fluctuation component dataset and global degradation trend component dataset

[0019] Step 4: Prediction of remaining lifespan of lithium-ion batteries based on the multi-scale support vector regression (SVR) model;

[0020] Step 4.1: Convert the frequency random component dataset Low-frequency fluctuation component dataset and the global degradation trend component dataset Q AL Individual capacity values ​​are predicted by inputting each value into the SVR, and the predicted values ​​of the high-frequency random components of the lithium-ion battery are obtained accordingly. Predicted values ​​of low-frequency fluctuation components Predicted values ​​of global degradation trend components

[0021] Step 4.2: Use equation (3) to obtain the predicted remaining lifespan RUL of the lithium-ion battery at time t. t Thus, the predicted remaining useful life under the full life cycle N is obtained;

[0022]

[0023] In equation (3), D represents the high-frequency random component D of the lithium-ion battery at time t. L,high The predicted value of (t), D represents the low-frequency fluctuation component at time t. L,low The predicted value of (t), A represents the global degradation trend component at time t. L The predicted value of (t);

[0024] Step 4.3: Determine whether the predicted remaining useful life under the whole life cycle N has reached the capacity failure threshold. If so, output the predicted remaining useful life at the time corresponding to the time when the capacity failure threshold is reached; otherwise, return to step one to re-predict.

[0025] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the lithium-ion battery remaining life prediction method, and the processor is configured to execute the program stored in the memory.

[0026] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the steps of the lithium-ion battery remaining life prediction method.

[0027] Compared with existing technologies, the beneficial effects of this invention are reflected in:

[0028] 1. This invention employs discrete wavelet decomposition to decouple the capacity time series at multiple scales, thereby reducing the impact of local regeneration and fluctuation phenomena on the prediction results and improving the model's prediction performance.

[0029] 2. This invention utilizes the K-means clustering method to cluster subsequences with similar sample entropy and permutation entropy in signals at various scales. Based on the clustering results, subsequences with similar complexity and randomness are reconstructed to reduce the number of subsequences related to local regeneration and fluctuations, reduce the number of modeling steps, and improve prediction efficiency.

[0030] 3. This invention establishes a remaining useful life prediction model based on a multi-scale support vector regression model, which enables the prediction model to accurately capture the changes in capacity signals at different scales and obtain accurate remaining useful life prediction results. Attached Figure Description

[0031] Figure 1 This is a flowchart of a lithium-ion battery remaining life prediction system.

[0032] Figure 2 This is a diagram of the wavelet decomposition structure. Detailed Implementation

[0033] The technical solution of the present invention will now be described more clearly and completely with reference to the embodiments and accompanying drawings.

[0034] In this embodiment, a lithium-ion battery remaining life prediction method based on a combination of discrete wavelet decomposition and multi-scale support vector regression model is a data-driven prediction method, such as... Figure 1 As shown, it includes: 1. Using the Discrete Wavelet Decomposition (DWD) method to perform multi-scale decomposition on the sampled lithium-ion battery capacity time series data, the decomposition structure is as follows: Figure 2 As shown. 2. Calculate the sample entropy and permutation entropy for each subsequence obtained after decomposition, cluster the data at each scale using the K-means clustering method, and reconstruct the subsequences with similar complexity and randomness based on the clustering results. 3. Use a multi-scale support vector regression (SVR) model to predict the clustered signal components separately, integrate the prediction results, and realize the prediction of the remaining service life of lithium-ion batteries. Specifically, it is carried out according to the following steps:

[0035] Step 1: Collect the capacity degradation dataset C(t) at time t during the entire life cycle N of a lithium-ion battery and perform L-level wavelet decomposition. Wavelet transform is an indispensable time-frequency analysis tool that can effectively handle transient and non-stationary signals. Due to its better localization properties than Fourier transform, wavelet transform has been widely used in signal processing, financial time series analysis, and image processing. The main purpose of wavelet analysis is to determine the frequency (or scale) content of a signal and analyze the time-varying characteristics of this frequency content. In wavelet analysis, a narrow window is used to obtain accurate time resolution when analyzing small-scale signals, and a wide window is used to obtain fine frequency resolution when analyzing large-scale signals; thus, the L-th level decomposed signal C′ at time t is obtained. L (t)=A L (t)+D L，1 (t)+D L,2 (t)+...+D L,j (t)...+D L,L (t); where A L (t) represents the low-frequency component of the Lth layer at time t, D L,j (t) represents the j-th high-frequency signal in the L-th high-frequency component at time t; j = 1, 2, ..., L; let the L-th high-frequency component D L(t)={D L，1 (t),D L,2 (t),…,D L,j (t),…,D L,L (t)};Wavelet decomposition only further decomposes the low-frequency part of the signal, and can quickly obtain the trend part that reflects the change law of the original signal under stationary conditions. It is suitable for decomposing signals such as battery degradation capacity, which are mainly composed of low-frequency information.

[0036] Step 2: Calculate D L,j The sample entropy and permutation entropy of (t) are used to obtain the high-frequency part D of the Lth layer. L The sample entropy and permutation entropy of (t); To study the non-stationarity of time series, approximate entropy is generally used to measure the complexity of the time series. Approximate entropy is a system complexity measure closely related to entropy; the more regular the time signal, the smaller the approximate entropy. However, in practical applications, approximate entropy suffers from problems such as dependence on data length and insufficient consistency. Therefore, this invention uses sample entropy to replace approximate entropy by eliminating self-matching. Compared with approximate entropy, sample entropy can reduce the computation time by half, is independent of data length, and has relative consistency;

[0037] Step 2.1, Calculate D L,j The sample entropy SE(D) of (t) L,j (t)), thus obtaining the high-frequency part D of the Lth layer. L The sample entropy SE(D) of (t) L (t))={SE(D L,1 (t)),SE(D L,2 (t)),…,SE(D L,j (t)),…,SE(D L,L (t))}:

[0038] Step 2.1.1: Take the j-th high-frequency signal from the high-frequency component of the L-th layer during a certain period of the lithium-ion battery's entire life cycle N to form the j-th m-dimensional high-frequency vector X of the L-th layer. L,j ={D L,j (t),D L,j (t+1),…,D L,j (t+m-1)}, 1≤t≤N-m+1; m is the vector dimension;

[0039] Step 2.1.2: Calculate any two X values ​​within the time range of 1 ≤ t ≤ N-m+1. L,j The distance between them is calculated, and the number of high-frequency signal pairs with a distance less than or equal to the given tolerance r is counted, denoted as A. L,j ;

[0040] Step 2.1.3: Calculate D using equation (1) L,jThe sample entropy SE(D) of (t) L,j (t)); In this embodiment, m = 4, r = 0.2SD, where SD is the standard deviation of the signal sequence.

[0041]

[0042] Step 2.2, Calculate D L,j The permutation entropy PE(D) of (t) L,j (t)), thus obtaining the high-frequency part D of the Lth layer. L The sample entropy PE(D) of (t) L (t))={PE(D L,1 (t)),PE(D L,2 (t)),…,PE(D L,j (t)),…,PE(D L,L The complexity of time series can be quantified by introducing the idea of ​​sorting. Using permutation entropy to measure the complexity of a time series can characterize the complexity of local structures in complex systems, offering advantages such as simplicity and fast computation. Permutation entropy is based on comparing the values ​​of each adjacent point and mapping them to a sequential pattern. It is suitable for chaotic and complex real-world time series and is commonly used to analyze electrocardiograms, stock market time series, etc.

[0043] Step 2.2.1: Take the j-th high-frequency signal from the high-frequency component of the L-th layer during a certain lag period in the entire life cycle N of the lithium-ion battery and form the j-th n-dimensional high-frequency lag vector ζ of the L-th layer. L,j ={D L,j (t),D L,j (t+l),…,D L,j (t+(n-1)l)}, where n is the embedding dimension and l is the hysteresis;

[0044] Step 2.2.2, for ζ L,j Sort the data and obtain n! sorted results, where! represents factorial; let α be the index of any sorted result, and ψ(α) represent the frequency of the α-th sorted result among the n! sorted results. Calculate the time series D using equation (2). L,j The permutation entropy PE(D) of (t) L,j (t)):

[0045]

[0046] The smaller the permutation entropy, the more regular the time series and the lower its randomness. Permutation entropy depends on the choice of embedding dimension n and lag time l. If n is too small, the scheme will not work; if it is too large, there will be memory limitations. In this embodiment, n = 4. Meanwhile, to minimize simulation errors, the lag time l is set to 1.

[0047] Step 3: Use the K-means clustering method to cluster subsequences with similar sample entropy and permutation entropy. Reconstruct subsequences with similar complexity and randomness based on the clustering results. K-means clustering is one of the most widely used clustering methods in machine learning research, capable of handling large datasets and quickly converging to local optima. Using sequences with similar sample entropy and permutation entropy as data points, select k initial samples as initial cluster centers. Each cluster in the dataset is defined by its centroid. In the data allocation step, each data point is assigned to the centroid with the smallest distance based on the square of the Euclidean distance. The average value of all data points is assigned to the centroid cluster. The entire process is repeated with the new centroid values, and all points are reassigned to new clusters. The centroid values ​​are continuously updated after each iteration. This process continues until no cluster center changes, indicating that the clustering has converged. According to D... L The entropy and permutation entropy of each high-frequency signal sample in (t) are analyzed using the K-means clustering method for D. L Clustering is performed on the high-frequency signals in (t) to obtain the clustering results, including: the high-frequency random component D of the Lth layer at time t. L,high (t) and the low-frequency fluctuation component D of the Lth layer at time t L,low (t); and the low-frequency component A of the Lth layer at time t. L (t) is used as the global degradation trend component, thus obtaining the high-frequency random component dataset under the entire lifecycle N. Low-frequency fluctuation component dataset and global degradation trend component dataset

[0048] Step 4: Prediction of remaining lifespan of lithium-ion batteries based on the multi-scale support vector regression (SVR) model;

[0049] Step 4.1: Convert the frequency random component dataset Low-frequency fluctuation component dataset and global degradation trend component dataset Individual capacity values ​​are predicted by inputting them into the SVR, and the predicted values ​​of the high-frequency random components of the lithium-ion battery are obtained accordingly. Predicted values ​​of low-frequency fluctuation components Predicted values ​​of global degradation trend components

[0050] Step 4.2: Use equation (3) to obtain the predicted remaining lifespan RUL of the lithium-ion battery at time t. t Thus, the predicted remaining useful life under the full life cycle N is obtained;

[0051]

[0052] In equation (3), D represents the high-frequency random component D of the lithium-ion battery at time t. L,high The predicted value of (t), D represents the low-frequency fluctuation component at time t. L,low The predicted value of (t), A represents the global degradation trend component at time t. L The predicted value of (t).

[0053] Step 4.3: Determine whether the predicted remaining useful life under the whole life cycle N has reached the capacity failure threshold. If so, output the predicted remaining useful life at the time corresponding to the time when the capacity failure threshold is reached; otherwise, it means that the predicted failure has not reached the capacity failure threshold and the prediction needs to be redone according to Step 1, Step 2, and Step 3.

[0054] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0055] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A method for predicting the remaining service life of lithium-ion batteries based on the DWD-SVR model, characterized in that, The procedure is as follows: Step 1: Collect the capacity degradation dataset C(t) at time t during the entire life cycle N of the lithium-ion battery and perform L-level wavelet decomposition to obtain the L-th level decomposed signal C′ at time t. L (t)=A L (t)+D L，1 (t)+D L,2 (t)+...+D L,j (t)...+D L,L (t); Among them, A L (t) represents the low-frequency component of the Lth layer at time t, D L,j (t) represents the j-th high-frequency signal in the L-th high-frequency component at time t; j = 1, 2, ..., L; let the L-th high-frequency component D L (t)={D L，1 (t),D L,2 (t),…,D L,j (t),…,D L,L (t)}; Step 2: Calculate D L,j The sample entropy and permutation entropy of (t) are used to obtain the high-frequency part D of the Lth layer. L The sample entropy and permutation entropy of (t); Step 2.1, Calculate D L,j The sample entropy SE(D) of (t) L,j (t)), thus obtaining the high-frequency part D of the Lth layer. L The sample entropy SE(D) of (t) L (t))={SE(D L,1 (t)),SE(D L,2 (t)),…,SE(D L,j (t)),…,SE(D L,L (t))}: Step 2.1.1: Take the j-th high-frequency signal from the high-frequency component of the L-th layer during a certain period of the lithium-ion battery's entire life cycle N to form the j-th m-dimensional high-frequency vector X of the L-th layer. L,j ={D L,j (t),D L,j (t+1),…,D L,j (t+m-1)}, 1≤t≤N-m+1; m is the vector dimension; Step 2.1.2: Calculate any two X values ​​within the time range of 1 ≤ t ≤ N-m+1. L,j The distance between them is calculated, and the number of high-frequency signal pairs with a distance less than or equal to the given tolerance r is counted, denoted as A. L,j ; Step 2.1.3: Calculate D using equation (1) L,j The sample entropy SE(D) of (t) L,j (t)); Step 2.2, Calculate D L,j The permutation entropy PE(D) of (t) L,j (t)), thus obtaining the high-frequency part D of the Lth layer. L The sample entropy PE(D) of (t) L (t))={PE(D L,1 (t)),PE(D L,2 (t)),…,PE(D L,j (t)),…,PE(D L,L (t))}; Step 2.2.1: Take the j-th high-frequency signal from the high-frequency component of the L-th layer during a certain lag period in the entire life cycle N of the lithium-ion battery and form the j-th n-dimensional high-frequency lag vector ζ of the L-th layer. L,j ={D L,j (t),D L,j (t+l),…,D L,j (t+(n-1)l)}, where n is the embedding dimension and l is the hysteresis; Step 2.2.2, for ζ L,j Sort the data and obtain n! sorted results, where! represents factorial; let α be the index of any sorted result, and ψ(α) represent the frequency of the α-th sorted result among the n! sorted results. Calculate the time series D using equation (2). L,j The permutation entropy PE(D) of (t) L,j (t)): Step 3, according to D L The entropy and permutation entropy of each high-frequency signal sample in (t) are analyzed using the K-means clustering method for D. L Clustering is performed on the high-frequency signals in (t) to obtain the clustering results, including: the high-frequency random component D of the Lth layer at time t. L,high (t) and the low-frequency fluctuation component D of the Lth layer at time t L,low (t); and the low-frequency component A of the Lth layer at time t. L (t) is used as the global degradation trend component, thus obtaining the high-frequency random component dataset under the entire lifecycle N. Low-frequency fluctuation component dataset and global degradation trend component dataset Step 4: Prediction of remaining lifespan of lithium-ion batteries based on the multi-scale support vector regression (SVR) model; Step 4.1: Convert the frequency random component dataset Low-frequency fluctuation component dataset and the global degradation trend component dataset Q AL Individual capacity values ​​are predicted by inputting each value into the SVR, and the predicted values ​​of the high-frequency random components of the lithium-ion battery are obtained accordingly. Predicted values ​​of low-frequency fluctuation components Predicted values ​​of global degradation trend components Step 4.2: Use equation (3) to obtain the predicted remaining lifespan RUL of the lithium-ion battery at time t. t Thus, the predicted remaining useful life under the full life cycle N is obtained; In equation (3), D represents the high-frequency random component D of the lithium-ion battery at time t. L,high The predicted value of (t), D represents the low-frequency fluctuation component at time t. L,low The predicted value of (t), A represents the global degradation trend component at time t. L The predicted value of (t); Step 4.3: Determine whether the predicted remaining useful life under the whole life cycle N has reached the capacity failure threshold. If so, output the predicted remaining useful life at the time corresponding to the time when the capacity failure threshold is reached; otherwise, return to step one to re-predict.

2. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the lithium-ion battery remaining life prediction method of claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium storing a computer program, characterized in that, The computer program, when run by the processor, executes the steps of the lithium-ion battery remaining life prediction method according to claim 1.

Images