Lithium battery cycle life prediction method and device based on aging state

Abstract

The invention provides a lithium battery cycle life prediction method and device based on an aging state, and belongs to the technical field of batteries, the life prediction method comprises the following steps: collecting lithium battery cycle life data, and identifying an aging curve inflection point, i.e., an inflection point of significant change of a battery capacity decline rate; based on the inflection point, dividing the loop data into a front state sequence and a rear state sequence, constructing a state transition matrix model, and identifying an initial state and a termination state; extracting regression features and classification features from the early cycle life data, calculating the duration of each aging state, and respectively training a state duration prediction model and a state classification model; and integrating the state transition matrix, the duration prediction model and the state classification model, and constructing a cycle life prediction model. Starting from the mechanism of the aging state of the lithium battery, the aging state of inflection point segmentation is introduced, quantitative description of the aging mode of the lithium battery and accurate prediction of the cycle life are realized, and the physical significance of the prediction result is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of batteries, and particularly to a method and device for predicting the cycle life of a lithium battery based on the aging state. Background Art

[0002] With the wide application of lithium batteries in fields such as electric vehicles, renewable energy storage, and portable electronic devices, accurately predicting the cycle life of lithium batteries is of great significance for strategic management and risk mitigation. However, the aging behavior of lithium batteries is complex and non-linear, and traditional data-driven methods often lack interpretability and are difficult to meet the actual application requirements.

[0003] Existing methods mostly focus on directly predicting the cycle life through machine learning models. For example, the invention patent application with the publication number CN116449240 discloses a method, system, device, and medium for analyzing the aging and predicting the life of a lithium-ion battery. The battery aging analysis method and life prediction method include constructing a battery aging mathematical model, which conducts aging analysis based on electrochemical reactions, mechanical stress, and material fatigue effects; calculating the remaining capacity of the battery after any number of charge-discharge cycles based on the battery aging mathematical model; and the remaining capacity of the battery is used to reflect the aging degree of the lithium-ion battery. Although this method can effectively analyze the aging degree of lithium-ion batteries, constructing a battery aging mathematical model in the form of numerical modeling is usually a black-box operation, making it difficult to reveal the causal relationship between the aging mechanism and the life, which limits its popularization in practical applications.

[0004] During the aging process of the battery under a given protocol, the capacity decay process can be roughly divided into two stages. In the first stage, the capacity decay is approximately linear with time or the number of cycles; in the second stage, the capacity decay rate suddenly accelerates, and the battery performance decays rapidly. The turning point between the two stages is called the capacity dip point or the inflection point in the battery aging trajectory (i.e., the critical point where the capacity drops sharply), which is a key sign of the aging mode change. When the capacity dip point or the inflection point in the battery aging trajectory appears during the aging process of the battery, the decay rate accelerates, which greatly affects normal use and usually means that the life is about to end; therefore, early determination and prediction are crucial in the use process of the whole life cycle of lithium-ion batteries.

[0005] Recent research has attempted to predict the lifespan through the capacity dive point or the inflection point in the battery aging trajectory. For example, the invention patent application with the publication number CN109596983A discloses a method for predicting the capacity dive during the battery aging process. The prediction method includes: 1) Arbitrarily select a new battery, and obtain the OCV-discharge capacity curve and the corresponding slope inflection point of the new battery; 2) Obtain the internal resistance of the new battery selected in step 1); 3) Obtain the relationship between the internal resistance and the number of cycles during the aging process of the battery of the same model as the battery selected in step 1); 4) Obtain the OCV-discharge capacity curve and the corresponding slope inflection point during the aging process of the battery of the same model as the battery selected in step 1); 5) Obtain the discharge curve and the corresponding slope inflection point under the given discharge regulation during the aging process of the battery of the same model as the battery selected in step 1); 6) Determine whether the slope inflection point of the discharge curve obtained in step 5) reaches the discharge lower cut-off voltage: If so, it is considered that the capacity of the battery under the given discharge regulation will experience a dive during the subsequent aging process. Based on this method, it is possible to predict early whether the capacity will dive during the battery aging process, and better use the battery throughout its entire life cycle.

[0006] Although recent research has attempted to predict the lifespan through inflection points, there are still limitations. Some methods rely on explicit mathematical expressions to fit the aging trajectory and cannot adapt to complex non-linear aging processes. And the differences in the aging patterns before and after the inflection point have not been deeply analyzed, resulting in the lack of physical meaning support for the prediction results. Summary of the Invention

[0007] The object of the present invention is to provide a method and device for predicting the cycle life of lithium batteries based on the aging state. Starting from the mechanism of the aging state of lithium batteries, the aging state divided by inflection points is introduced, and the aging trajectory is divided into an initial state (before the inflection point) and a termination state (after the inflection point) according to the inflection point. The aging patterns of each are modeled respectively to obtain the aging law of lithium batteries, and the lifespan of lithium batteries is predicted in a data-driven manner through the aging law, realizing the quantitative description of the aging mode of lithium batteries and the accurate prediction of the cycle life, and enhancing the interpretability of the model prediction.

[0008] A method for predicting the cycle life of lithium batteries based on the aging state provided by the embodiment includes the following steps:

[0009] Collect the cycle life data of lithium batteries, and identify the inflection points of the aging curve through the cycle life data. The inflection point is the turning point where the battery capacity decline rate changes significantly;

[0010] Based on the inflection points, divide the cycle life data into two state sequences before and after, establish a state transition matrix model, and identify the initial state and the termination state;

[0011] Extract regression features and classification features from the early cycle life data of lithium batteries, calculate the duration of each aging state, use the duration of the aging state and the regression features as inputs, and train a state duration prediction model; extract classification features from the regression features, and use the initial state identified by the state transition matrix model and the classification features as inputs to train a state classification model;

[0012] Integrate the state transition matrix model, the state duration prediction model and the state classification model to construct a lithium battery cycle life prediction model, and predict the cycle life of the lithium battery.

[0013] In one embodiment, the cycle life data includes battery cycle data under different charging protocols, temperatures and internal resistance conditions.

[0014] In one embodiment, identifying the inflection point of the aging curve from the cycle life data includes:

[0015] Preprocess the cycle life data to remove outliers and noise data;

[0016] Adopt an offline inflection point identification algorithm, use the aging curve drawn from the cycle life data as input, and detect the inflection point of the lithium battery aging trajectory.

[0017] In one embodiment, use a clustering algorithm to cluster the segments before and after the inflection point. The segments with similar aging states are grouped into one category. Each aging state is described by aging characteristics and duration. The cycle life data within the segment has similar aging characteristics within the state.

[0018] In one embodiment, establishing the state transition matrix model includes: according to the aging state sequence, count the occurrence times of state pairs, calculate the state transition frequency matrix, normalize the state transition frequency matrix to obtain the state transition probability matrix, and establish the state transition matrix model.

[0019] In one embodiment, the regression features are extracted based on the early cycle life data and are used to describe the aging state of the lithium battery.

[0020] In one embodiment, the classification features are selected from the regression features by the recursive feature elimination method and are used to classify the aging state of the lithium battery.

[0021] In one embodiment, the objective function of the lithium battery cycle life prediction model is where y i is the true cycle life of the lithium battery, is the predicted cycle life of the lithium battery; θ is the set of state duration prediction model parameters and state classification model parameters; α is the weight factor used to balance the contributions of state duration prediction and state transition.

[0022] In one embodiment, the prediction of the cycle life of the lithium battery includes:

[0023] Taking the early cycle data of the lithium battery as input and using the state classification model to predict the initial state s 0 and the duration d of the initial state i，1 ;

[0024] Based on the predicted initial state s 0 , through the cumulative distribution function calculate the termination state s * ;

[0025] Based on the predicted initial state s 0 and the termination state s * , use the lithium battery cycle life prediction model to predict the duration of the termination state, and linearly sum the duration of the initial state and the duration of the termination state to obtain the lithium battery cycle life.

[0026] In one example, the predicted initial state s 0 is the state with the maximum probability in the probability distribution predicted by the state classification model;

[0027] The said termination state s * is the smallest state that satisfies Φ(s)≥p * , and is selected from the cumulative distribution function Φ(s) by Monte Carlo sampling through the random number p * ∈[0, 1].

[0028] To clearly show the lithium battery cycle life prediction method based on the aging state, a lithium battery cycle life prediction device based on the aging state is provided, including a memory and a processor. The memory is used to store a computer program, and the processor is used to implement the lithium battery cycle life prediction method based on the aging state when executing the computer program.

[0029] Compared with the prior art, the beneficial effects of the present invention at least include:

[0030] (1) The aging state segmented by the inflection point can capture the key features in the aging process of the lithium battery and improve the prediction accuracy of the model for the cycle life of the lithium battery.

[0031] (2) The aging state provides an in-depth understanding of the aging process of the lithium battery, making the prediction results of the model more interpretable. The state transition matrix is correlated with the microscopic aging mechanism (such as lithium ion loss, active material loss), which can enhance the physical meaning of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art.

[0033] Figure 1 It is a flowchart of the lithium battery cycle life prediction method based on the aging state in the present invention.

[0034] Figure 2 It is the prediction result of the present invention based on the MIT-Stanford-Toyota dataset. Specific embodiments

[0035] To make the objectives, technical solutions and advantages of the present invention more clearly understood, the following further details the present invention with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and do not limit the protection scope of the present invention.

[0036] The solution of the embodiment of the present invention is as Figure 1 shown. A lithium battery cycle life prediction method based on the aging state includes the following steps:

[0037] S1. Collect the cycle life data of the lithium battery, and identify the inflection point of the aging curve through the cycle life data. The inflection point is the turning point where the battery capacity degradation rate changes significantly.

[0038] In the embodiment, collect the cycle life data of the lithium battery, including the battery cycle data under different charging protocols, temperatures and internal resistances, etc., denoted as representing the data of lithium battery i at cycle t. Clean and preprocess the collected data to remove outliers and noise data to ensure the accuracy and integrity of the data;

[0039] Calculate the average capacity of each cycle in the cycle life data for capacity extraction;

[0040] The detection of the inflection point is based on the process of gradual decline of the battery capacity. When the battery capacity degradation rate changes significantly, it is considered that an inflection point appears. The offline inflection point identification algorithm Bacon-Watts method is used to detect the inflection point of the battery aging trajectory. For the capacity c of the cycle data i , identify the inflection point t in the aging trajectory i . The so-called offline algorithm refers to an algorithm whose input is a complete aging curve.

[0041] S2. Based on the inflection point, divide the cycle life data into two segments of state sequences before and after, establish a state transition matrix model, and identify the initial state and the termination state.

[0042] In the embodiment, based on the inflection point t identified from each aging trajectory curve in S1 i , the cycle life data is divided into X i initial segments and termination segments Using the K-Means++ algorithm to cluster the segmented segments, removing clusters with significantly fewer internal segment numbers than other clusters. Similar aging segments will be grouped into one category, forming different aging states. Each state is mainly described by its aging characteristics and duration. The data within the segment has similar aging characteristics within the state. Each clustered segment is labeled as state f i，k ∈s x , obtaining the state set S = {s 1 , s 2 ,..., s x ,...} and the state sequence [s x , s y .

[0043] Based on the state sequence in the state set, count the occurrence times of state pairs and calculate the state transition frequency matrix By row-normalizing the state transition frequency matrix to obtain the state transition probability matrix A = {a ij}, Establish a state transition matrix model to describe the transition relationship and probability between different states, where a ij = P(s j |s i ) represents the probability of transitioning from state s i to s j , and n = |S| is the number of states. Identify the initial state S 0 = {q i，1 |q i ∈q} and the termination state where q represents the state sequence, q i represents the state sequence of the i-th aging data, and q i，1 represents the first state of the i-th aging data, indicating that a certain row in the state probability transition matrix being 0 is the termination state.

[0044] S3. Extract regression features and classification features from the early cycle life data of the lithium battery, calculate the duration of each aging state, and use the duration of the aging state and the regression features as inputs to train a state duration prediction model; extract classification features from the regression features, and use the initial state identified by the state transition matrix model and the classification features as inputs to train a state classification model.

[0045] In the embodiment, regression feature V is extracted from the early cycle data of the lithium battery r and classification feature V c . The regression feature is extracted from data such as the capacity-voltage curve based on domain knowledge and is used to describe the aging state of the battery. The classification feature is selected from the regression features by the recursive feature elimination method and is used for state classification.

[0046] Next, the average duration of each aging state is calculated where f i，k is the data segment, q i，k is the state of the data segment, s x is a state in the state set, and |·| represents length or quantity.

[0047] The state duration d is obtained based on the average duration D i，j , and the state duration prediction model g is trained using the regression feature to obtain the model parameter θ 1 of the state duration prediction model g. This model is used to predict the duration of the termination state Based on the initial state set S 0 , the state classification model h is trained using the classification feature to obtain the model parameter θ 2 of the state classification model h. This model is used to predict the initial state and obtain the probability distribution p i，1 = h(V i c ).

[0048] S4. An integrated state transition matrix model, state duration prediction model, and state classification model are used to construct a lithium battery cycle life prediction model to predict the cycle life of the lithium battery.

[0049] In the embodiment, a machine learning or deep learning model (such as an elastic net, decision tree, and neural network, etc.) is selected as the benchmark state duration prediction model and state classification model. Combined with the state transition matrix model, the early cycle data is used to train the model, and the model hyperparameters are optimized through five-fold cross-validation to construct a lithium battery cycle life prediction model. The objective function of the model is:

[0050]

[0051] where α is the weight factor, and θ is the set of the state duration prediction model parameter θ 1 and the state classification model parameter θ 2 ; is the root mean square error, y i is the true cycle life of the lithium battery, is the predicted cycle life of the lithium battery. Based on the true lithium battery life yi Combined with the trained state duration prediction model, state classification model, and state transition matrix model, a weight factor α is obtained to balance the contributions of state duration prediction and state transition.

[0052] In the prediction stage, the model predicts the initial state and its duration of the battery based on the input early cycle data, then predicts the subsequent termination state and its duration through the state transition matrix and regression features, and finally obtains the cycle life of the battery through linear summation. The specific process is as follows:

[0053] Using the state classification model h, based on the early cycle features, predict the initial state s of the battery 0 and its duration d i，1 = h(V i c )D, where the initial state s 0 ∈ S 0 is the state with the maximum probability in the probability distribution predicted by the classification model.

[0054] Based on the state transition probability matrix A, generate the termination state s* through the Monte Carlo algorithm. The termination state is obtained by the cumulative distribution function and represents the cumulative probability of transferring from the initial state s 0 to all states less than or equal to s. Here, s is a variable in the state space, arranged in a certain order. According to the Monte Carlo algorithm, the computer generates a random number p * ∈ [0, 1], and the termination state satisfies argmin s s.t. Φ(s) ≥ p * .

[0055] Use the lithium battery cycle life prediction model to predict the duration of the termination state where α is the weight factor;

[0056] Linearly sum the initial state duration d i，1 and the termination state duration d i，2 to obtain the total life Realize the prediction of the lithium battery cycle life.

[0057] To verify the effectiveness of the present invention, verification was carried out. This method uses the publicly available MIT-Stanford-Toyota lithium-ion battery dataset for verification. The dataset contains cycle-to-failure data of 124 lithium iron phosphate / graphite lithium-ion batteries in three batches. These datasets cover battery cycle data under different charging strategies and are appended with measurement data of temperature and internal resistance.

[0058] In the experiment, the Elastic Net (EN), Decision Tree (DT), and Neural Network (NN) were selected as the base regressors. The hyperparameters were optimized through grid search. The experiment adopted a five-fold cross-validation method, repeated 20 times, and the Root Mean Square Error (RMSE) and Coefficient of Determination (R 2 were used as the validation metrics. The results are as Figure 2 shown. The "S-" model is the regressor based on the aging state proposed in the present invention.

[0059] As Figure 2 shown by the results, the regressor based on the aging state provided by the present invention is superior to the base regressors in all early-cycle data. Specifically, among the base regressors, the neural network performs poorly on different early-cycle data due to its high demand for data. The average RMSE value is about 200, and the average R2 value is about 0.3. As a linear model, the Elastic Net has good interpretability but cannot effectively capture the non-linear degradation pattern. In contrast, the Decision Tree can effectively capture the non-linear pattern and performs best among the base regressors.

[0060] The regressor based on the aging state provided by the present invention is superior to the base regressors in terms of both RMSE and R 2 metrics, and the result distribution is more concentrated, indicating that the aging state significantly improves the robustness of the regressor. By transforming the complex non-linear degradation pattern into a state transition, this method not only improves the prediction accuracy but also enhances the interpretability of the model.

[0061] Based on the analysis results, the lithium battery cycle life prediction model based on the aging state provided by the present invention realizes the quantitative description of the lithium battery aging mode and the accurate prediction of the cycle life. By segmenting the inflection points of the aging curve, the key features in the battery aging process can be captured. In addition, the aging state provides an in-depth understanding of the battery aging process, making the prediction results of the model more interpretable. The state transition matrix is correlated with the microscopic aging mechanisms (such as lithium ion loss and active material loss), which can enhance the physical meaning of the prediction results.

[0062] The above specific embodiments have detailed the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, supplements, equivalent replacements, etc. made within the scope of the principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the cycle life of a lithium battery based on aging status, characterized in that: The following steps are involved: Collecting cycle life data of lithium batteries, and identifying the inflection point of the aging curve through the cycle life data, wherein the inflection point is the turning point at which the rate of battery capacity decrease changes significantly; Based on the inflection point, the cycle life data is divided into two state sequences, a state transfer matrix model is established, and the initial state and the terminal state are identified; Regression features and classification features are extracted from the early cycle life data of lithium batteries, the duration of each aging state is calculated, and the state duration prediction model is trained with the duration of the aging state and the regression features as input; classification features are extracted from the regression features, and the initial state identified by the state transfer matrix model and the classification features are used as input to train the state classification model; The state transfer matrix model, state duration prediction model and state classification model are integrated to construct a lithium battery cycle life prediction model to predict the cycle life of lithium batteries.

2. The lithium battery cycle life prediction method according to claim 1, characterized in that: The aging data includes battery cycling data under different charging protocols, temperatures, and internal resistance conditions.

3. The lithium battery cycle life prediction method according to claim 1, characterized in that: The method of identifying the inflection point of the aging curve through the cycle life data includes: Preprocess the cycle life data to remove outliers and noise data; An offline inflection point recognition algorithm is used to detect the inflection point of the lithium battery aging trajectory using the aging curve drawn from the cycle life data as input.

4. The lithium battery cycle life prediction method according to claim 1, characterized in that: A clustering algorithm is used to cluster the segments before and after the inflection point. Segments with similar aging states are grouped together. Each aging state is described by aging characteristics and duration. The cycle life data within the segment have similar aging characteristics within the state.

5. The lithium battery cycle life prediction method according to claim 1, characterized in that: The state transfer matrix model is established by counting the number of occurrences of state pairs according to the aging state sequence, calculating the state transfer frequency matrix, normalizing the state transfer frequency matrix to obtain the state transfer probability matrix, and establishing the state transfer matrix model.

6. The lithium battery cycle life prediction method according to claim 1, characterized in that: The regression features are extracted based on early cycle life data and are used to describe the aging state of lithium batteries; The classification features are selected from the regression features by a recursive feature elimination method and are used to classify the aging state of the lithium battery.

7. The lithium battery cycle life prediction method according to claim 1, characterized in that: The objective function of the lithium battery cycle life prediction model is Among them, y i is the real cycle life of lithium battery, is the predicted cycle life of lithium battery; θ is the set of state duration prediction model parameters and state classification model parameters; α is the weight factor used to balance the contribution of state duration prediction and state transition.

8. The lithium battery cycle life prediction method according to claim 1, characterized in that: The prediction of the cycle life of the lithium battery includes: Using the early cycle data of lithium batteries as input, the state classification model is used to predict the initial state of lithium batteries. 0 and initial state duration d i，1 ; Based on the predicted initial state s 0 , through the cumulative distribution function Calculate the terminal state s * ; Based on the predicted initial state s 0 and the terminal state s * , the lithium battery cycle life prediction model is used to predict the duration of the terminal state, and the initial state duration and the terminal state duration are linearly summed to obtain the lithium battery cycle life.

9. The lithium battery cycle life prediction method according to claim 8, characterized in that: The predicted initial state s 0 is the state with the maximum probability in the probability distribution predicted by the state classification model; The terminal state s * is such that Φ(s)≥p * The minimum state is obtained by Monte Carlo sampling through random numbers p * ∈[0, 1] is selected from the cumulative distribution function Φ(s).

10. A lithium battery cycle life prediction device based on aging status, comprising a memory and a processor, wherein the memory is used to store a computer program, characterized in that: The processor is used to implement the method for predicting the cycle life of a lithium battery based on aging status as described in any one of claims 1 to 9 when executing the computer program.

Images