A method and system for predicting the remaining service life of a lithium-ion battery

Abstract

In view of the limitations of existing numerical prediction strategies, the present invention proposes a method and system for predicting the interval of remaining useful life of lithium-ion batteries. First, fuzzy information granulation is introduced to process the battery capacity degradation time series, and the original numerical data is converted into particle level, which is the basis for realizing interval prediction. Secondly, in order to solve the problem of loss of fluctuation information caused by fuzzy information granulation when processing battery degradation data, the present invention creatively introduces a language description method to attach semantic labels to each fuzzy particle to indicate its changing state. Finally, combined with the least squares support vector machine algorithm, particles with semantic labels are used as the training input of the model, which takes into account the fluctuation characteristics of the degradation sequence while realizing the interval prediction of battery RUL. The present invention can realize the interval prediction of RUL of lithium-ion batteries, and has good prediction effect and good applicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion batteries, and in particular relates to a method and system for predicting the remaining service life interval of a lithium-ion battery based on fuzzy information granulation and language description. Background Art

[0002] As a renewable energy source, lithium-ion batteries have the advantages of high energy density, low self-discharge rate and long life, and are now widely used in many fields. From smartphones to electric vehicles to aerospace power systems, the stable operation of all of these is inseparable from the relatively stable energy supply of lithium-ion batteries. However, as the life cycle increases, the battery performance will gradually decline. When the degradation reaches a certain point (failure threshold), the possibility of battery failure will increase, which may lead to mission failure or even catastrophic events. Therefore, building a satisfactory battery remaining useful life (RUL) prediction model is of great practical significance for preventing the occurrence of dangerous accidents.

[0003] From the perspective of practical application, a well-designed prediction model should not only be able to predict the changing trend of battery capacity, but also provide a prediction interval for the remaining service life of the battery, so that users can get adequate warning before potential dangerous failures occur and make reasonable decisions to reduce unnecessary losses. In situations where safety or urgency is high, choices should be made as early as possible within the RUL prediction interval to avoid accidents, otherwise the use cycle can be appropriately extended to reduce waste. Therefore, establishing a prediction model for the remaining service life interval of lithium-ion batteries is of great significance for the practical application of lithium-ion batteries.

[0004] In recent years, researchers have studied the prediction of battery RUL from different perspectives. However, most of the existing research on lithium-ion battery health management and RUL prediction is based on the numerical level, that is, giving the exact value of the prediction result, rather than providing a prediction interval that can be used for decision-making. Some scholars have proposed new data-driven methods or hybrid methods to combine the relative advantages of different types of methods to predict battery degradation trends. Some scholars have extracted new health indicators (HIs) or performed correlation analysis on multiple HIs to identify more sensitive variables in the battery degradation state. Some literature has improved the existing optimization algorithms, and some literature has considered the capacity regeneration behavior of the battery. The prediction results of the above schemes have achieved a certain accuracy, but they are all based on the numerical prediction method of battery RUL. When it comes to practical problems, on the one hand, in some cases where human reasoning and decision-making need to rely on the prediction results, quantitative numerical results may not be absolutely necessary; on the other hand, considering that the battery degradation trajectory is complex, the measurement process is easily disturbed, and the original data is noisy, unstable and highly nonlinear, it is difficult to obtain an accurate battery RUL prediction model, and the current strategy based on numerical prediction cannot achieve a perfect prediction result, but may bring greater unreliability to the estimation of battery RUL. Summary of the invention

[0005] In view of the limitations of existing numerical prediction strategies, the present invention proposes a method and system for predicting the remaining useful life interval of a lithium-ion battery based on fuzzy information granulation and language description methods.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A method for predicting the remaining service life of a lithium-ion battery comprises the following steps:

[0008] Extract battery capacity degradation time series from battery dataset;

[0009] Discretize the battery capacity degradation time series and divide it into fixed-width, non-overlapping subsets, each of which is called a time window.

[0010] The time window is processed by using the fuzzy information granulation method to generate fuzzy particles corresponding to the time window;

[0011] Describe the fuzzy particles in language and attach semantic labels to each fuzzy particle to indicate its fluctuation state;

[0012] Using fuzzy particles with semantic labels as input, a least squares support vector machine model is adopted to perform interval prediction of the remaining battery life.

[0013] Furthermore, the method of processing the time window by using the fuzzy information granulation method to generate fuzzy particles corresponding to the time window includes:

[0014] Select the triangular membership function to generate fuzzy particles; through fuzzy information granulation, the time window T1, T2, …, T t Generate corresponding fuzzy particles G1, G2, ..., G t Each particle G i Keep three values ​​a i 、m i and b i , these three values ​​reflect the time window T i The main information of i and b i They represent the lower and upper bounds of the generated particles, respectively, and m i represents the average level of generated particles; then, the three sequences corresponding to the blurred particles {a1, a2, …, a t}, {m1,m2,…,m t} and {b1,b2,…,b t} are represented by Low, R and Up respectively, where Up and Low provide upper and lower bounds for interval prediction.

[0015] Furthermore, the language description of the fuzzy particles and adding a semantic label to each fuzzy particle to indicate its fluctuation state includes:

[0016] According to the change of time series, its fluctuation state is divided into several degrees to form a semantic set L = {L1, L2, ..., L c}, where L j Represents a language item, corresponding to a fluctuating state of data;

[0017] The fuzzy particles G1, G2, ..., G t The fluctuation state of the data in the corresponding time window is compared with the elements in the semantic set L one by one, and the language item closest to its fluctuation state is selected for each particle in turn to describe it, generating an information feature sequence The elements in this sequence are associated with the fluctuation state of the corresponding particle.

[0018] Furthermore, a semantic set consisting of five levels of linguistic items, namely lower, low, medium, high and higher, is adopted to describe the fluctuation state of particles in the battery degradation time series.

[0019] Furthermore, the method of using the fuzzy particles with semantic labels as input and adopting the least squares support vector machine model to predict the remaining service life of the battery includes:

[0020] Take the first num particles as training data and predict the particles from num+1 to t;

[0021] The final interval prediction result of the battery RUL consists of two parts. The first part is the prediction result of Up and Low of the particles, and the second part is the prediction result of the language item sequence.

[0022] When merging the prediction results, the prediction results of the second part, namely the language item sequence, are traced back and converted into fluctuation data of corresponding particles, which are superimposed with the prediction results of the first part to obtain the final prediction results.

[0023] Furthermore, the specificity criterion and the coverage criterion are used to evaluate the performance of interval prediction, where the specificity criterion is used to calculate the width of the prediction interval and the coverage criterion is used to measure the accuracy of the prediction interval.

[0024] A lithium ion battery remaining service life interval prediction system using the method of the present invention comprises:

[0025] A time series extraction module is used to extract the battery capacity degradation time series from the battery data set;

[0026] The time series discretization module is used to discretize the battery capacity degradation time series and divide the battery capacity degradation time series into fixed-width, non-overlapping subsets, each of which is called a time window;

[0027] A time window granulation module is used to process the time window using a fuzzy information granulation method to generate fuzzy particles corresponding to the time window;

[0028] The language description module is used to describe the fuzzy particles in language and add a semantic label to each fuzzy particle to indicate its fluctuation state;

[0029] The interval prediction module is used to use the fuzzy particles with semantic labels as input and adopt the least squares support vector machine model to perform interval prediction on the remaining service life of the battery.

[0030] The beneficial effects of the present invention are as follows:

[0031] In view of the limitations of existing numerical prediction strategies, a lithium-ion battery remaining service life interval prediction strategy based on fuzzy information granulation and language description method is proposed. First, fuzzy information granulation is introduced to process the battery capacity degradation time series, and the original numerical data is converted into granular level, which is the basis for realizing interval prediction. Secondly, in order to solve the problem of loss of fluctuation information caused by fuzzy information granulation when processing battery degradation data, the present invention creatively introduces a language description method to attach semantic labels to each fuzzy particle to indicate its changing state. Finally, combined with the least squares support vector machine algorithm, the particles with semantic labels are used as the training input of the model to predict the battery RUL. While realizing the battery RUL interval prediction, the fluctuation characteristics of the degradation series are taken into account.

[0032] The present invention uses four sets of battery degradation data from NASA's battery degradation test to conduct experiments. The experimental results of different battery data show that the strategy can predict the RUL interval of lithium-ion batteries and has good applicability. In addition, the present invention introduces an evaluation criterion for the interval prediction model, and evaluates the interval prediction performance of the proposed model from two aspects: interval coverage and specificity. The results show that the interval prediction model based on fuzzy information granulation and language description proposed in the present invention has a good prediction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the RUL interval prediction model based on fuzzy information granulation and language description.

[0034] Figure 2 It is the capacity degradation curve of four groups of batteries (cell 5, cell 6, cell 7 and cell 18).

[0035] Figure 3 This is the fuzzy granulation result of cell 5.

[0036] Figure 4 It is the prediction result of the language description of cell 5 from the 24th particle to the 56th particle.

[0037] Figure 5 It is the interval prediction result based on fuzzy information granulation and language description, where: (a) cell 5, (b) cell 6, (c) cell 7, (d) cell 18. DETAILED DESCRIPTION

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below through specific embodiments and drawings.

[0039] 1. Interval prediction model based on fuzzy information granulation and language description

[0040] The lithium-ion battery RUL interval prediction model based on fuzzy information granulation and language description proposed in the present invention can be summarized into five steps: Figure 1 As shown in the figure, they are respectively the extraction of time series, the discretization of time series, the granulation of time windows, the language description of granules and the establishment of a prediction model. n} represents the time series, T1, T2, …, T t represents the time window, and w represents the width of the window. The detailed description is as follows.

[0041] Step 1: Extraction of time series X. As we all know, in the field of RUL prediction of lithium-ion batteries, capacity directly reflects the health status of the battery, and modeling it as HI (health index) is usually more accurate and effective than other methods. Therefore, the capacity degradation time series is extracted from the battery data set for subsequent research. This step corresponds to Figure 1 Step 1 in

[0042] Step 2: Discretization of time series X. Discretization means dividing the original time series into fixed-width, non-overlapping subsets, each of which is called a time window. In this step, the key is to determine the width w of the window, and subsequent predictions are based on this operation. Figure 1 As shown in Step 2 in , after discretization, the time series will be divided into time windows T1, T2, …, T t .

[0043] Reasonable time window segmentation is the premise for better analysis of the original time series. When segmenting the original time series, the fixed-length segmentation method is generally used, that is, the window is segmented with a constant width, so that each subsequence formed contains the same number of sample data. Let X = {x1, x2, ..., x n} is a set of time series, and the window size after segmentation is defined as w, which ranges from 1≤w≤n. The size of the time window is related to the ability to retain the inherent information of the original time series. If w=n, it means that the entire time series is divided into a window, then the size of the particle is too large to reflect the advantage of information granulation. If w=1, each sample x i All are considered as one particle. In this case, the granulation of the time window is invalid. Therefore, choosing an appropriate time window size is crucial.

[0044] When determining the width w of the time window, the present invention also pays attention to the size of the data set used in the experiment. If the data set is small, w should be set as small as possible while ensuring that the advantages of granulation can be exercised. Otherwise, the number of particles after granulation is too small, which will affect subsequent modeling and prediction.

[0045] Step 3: Granulation of time window. After obtaining the time window, the fuzzy information granulation method is used to process the time window subset. The triangular membership function is selected to generate fuzzy particles. Figure 1 As shown in step 3, through fuzzy information granulation, the time windows T1, T2, ..., T t Generate corresponding fuzzy particles G1, G2, ..., G t After the triangular membership function processing, each particle G i Keep three values ​​a i 、m i and b i , which reflect the time window T i The main information of i and b i They represent the lower and upper bounds of the generated particles, respectively, and m i represents the average level of generated particles. Then, the three sequences corresponding to the blurred particles {a1, a2, …, a t}, {m1,m2,…,m t} and {b1,b2,…,b t} are represented by Low, R, and Up respectively. In particular, as the basis for interval prediction, Up and Low can provide upper and lower bounds for interval prediction.

[0046] Information granulation proposed by Zadeh et al. (Zadeh, LA, Fuzzy sets and information granularity. Advances in fuzzy set theory and applications, 1979: p. 3-8.) refers to dividing an overall object into multiple small parts, each of which can be regarded as an information granule. Information granulation can help us simplify complex problems and capture some details of the problem. For time series prediction problems, through information granulation, the original data can be organized into some meaningful and semantically sound information granules, thereby transforming the prediction problem from the numerical level to the granular level.

[0047] Under the concept of information granulation, the present invention mainly uses fuzzy set theory to express information granules, and this process is called fuzzy information granulation. When fuzzy information granulation is performed on time series such as battery degradation data, the original numerical series can be converted into fuzzy particles, which meets the data processing requirements of RUL interval prediction. In addition, while revealing the essence of time series, fuzzy information granulation can also reduce the computational overhead of the model to a certain extent.

[0048] After discretizing the time series, the obtained time windows are fuzzy information granulated, that is, fuzzy information particles are generated. For a time series X, in the process of fuzzy information granulation, a subset of each time window is processed to generate fuzzy information particles. Each fuzzy particle can be regarded as a representative of the corresponding time window information.

[0049] Let U be a domain, which represents a time window. The general form of fuzzy information granulation of time series is as follows:

[0050]

[0051] In the formula, g represents fuzzy particles; X represents time series; x is a variable that takes values ​​from U, G is a convex fuzzy subset of U, represented by a membership function; λ represents the probability that x belongs to G. The core of fuzzy information granulation is to determine the membership function of G. The present invention selects the triangular membership function A(x), which is mainly introduced as follows:

[0052]

[0053] Where m is the median of the subset in the time window, and a and b are the lower and upper bounds of the support for generating fuzzy particles. Then, the particle expression of the triangular membership function is constructed, that is, the above parameters are determined. The construction of fuzzy particles needs to meet two conditions. (1) The fuzzy particles can reasonably represent the original information. (2) The fuzzy particles should have a certain specificity. To meet the above conditions, the constructor Q is constructed. A as follows:

[0054]

[0055] Where M A Satisfies condition (1). N A represents the support boundary of A(x), satisfying condition (2). A When is maximum, the above conditions reach the optimal solution.

[0056] Specifically, let X' be a subset of one of the time windows. First, sort the subset in ascending order: X' = {x1, x2, ..., x M ,…,x K}. Let m = x M, where m is the median of X', determined by formula (4).

[0057]

[0058] Once the value of m is determined, X' will be divided into two subsequences, which will be processed separately to calculate the left side (sequences less than m) and the right side (sequences greater than m) of A(x). According to the left side of A(x) and formula (3), the function about a can be obtained:

[0059]

[0060] Similarly, the function of b can also be obtained:

[0061]

[0062] The derivatives of Q(a) and Q(b) are calculated as follows:

[0063]

[0064]

[0065] When Q(a) and Q(b) reach their maximum values, the values ​​of a and b are obtained respectively.

[0066]

[0067]

[0068] Step 4: Language description of particles. While fuzzy information granulation provides conditions for interval prediction, it also makes the obtained data set smoother than the original data set. This may weaken some fluctuation information in some data sets with special fluctuation processes (such as lithium-ion battery degradation data sets), thereby affecting the prediction results. To this end, the language description method is used in this step to better capture the fluctuation characteristics of battery degradation data. t After the language description is performed separately, a feature sequence of particles is generated. This sequence is used as a set of features to capture the changes in the time series in subsequent modeling, which can help better understand the characteristics of the battery degradation time series, thereby making more accurate and effective modeling for RUL prediction.

[0069] The linguistic description of particles refers to the reasonable allocation of language items to each particle to generate a set of unique information sequences, which further provides the particles with fluctuation information about their state changes. Specifically, according to the changes in the time series, its fluctuation state is divided into several degrees to form a semantic set L = {L1, L2, ..., L c}, where L jRepresents a language item, corresponding to a fluctuation state of data. After having the semantic set L, the information particles G1, G2, ..., G t The fluctuation state of the data in the corresponding time window is compared with the elements in the set one by one, and the language item closest to its fluctuation state is selected for each particle in turn to describe it. Once all particles are described in language, a sequence of language items corresponding to their fluctuation characteristics will be generated.

[0070] In order to better understand the above process, for the battery degradation time series, the fluctuation state can be simply divided into three degrees, and the language item set L = {L1, L2, L3} representing low, medium and high is used to describe the information particles after fuzzy granulation. After the fuzzy particles are generated, the fuzzy information particles are described one by one using appropriate language items according to the fluctuation information in different time windows. This operation is similar to adding semantic labels to the particle data, such as using Lowfluctuation degree, L1 and high fluctuation degree, L3 to mark particles with low and high fluctuation degrees respectively. After the language description of each particle, an information feature sequence is generated. The elements in the sequence are related to the fluctuation state of the corresponding particles. Through the above operations, each particle of the time series can have its own semantic label, so as to better understand the fluctuation characteristics of the battery degradation time series.

[0071] Step 5: Modeling and prediction. After fuzzy information granulation and language description, the sequence of Up, Low and language items will replace the original time series for modeling in this step. From numerical data to labeled granular data, the number of samples used for modeling will be reduced, so the least squares support vector machine (LSSVM) suitable for small data sets is used for prediction. Figure 1 In step 5 (Step 5), take the first num particles as training data and predict the particles from num+1 to t. The final interval prediction result of the battery RUL consists of two parts. The first part (Part 1) is the prediction result of the Up and Low of the particles, and the second part (Part 2) is the prediction result of the language item sequence. When merging the prediction results, it is necessary to trace back the prediction results of the second part, i.e., the language item, and convert them into the fluctuation data of the corresponding particles, and superimpose them with the prediction results of the first part to obtain the final prediction results.

[0072] LSSVM is a powerful tool for solving classification and regression problems developed on the basis of Support Vector Machine (SVM). SVM has good generalization performance and high processing capabilities for nonlinear problems and small samples. LSSVM inherits the above advantages well, and further transforms inequality constraint problems into equality constraint problems, and transforms quadratic programming problems into linear equations, which improves convergence accuracy while reducing time consumption. As mentioned above, LSSVM provides great convenience for the regression problem of nonlinear small-scale data sets, and its regression principle is as follows:

[0073] Given a training set {(x1,y1),(x2,y2),…,(x l ,y l )},y i ∈R, where x i is the input vector of the i-th training sample, y i is the corresponding output. The regression model can be constructed as follows:

[0074]

[0075] Where ω is the weight vector in the feature space, b is the bias, It is a nonlinear mapping function that maps the input vector x to a high-dimensional feature space. According to the principle of structural risk minimization, the minimum objective function of LSSVM can be expressed as:

[0076]

[0077] Where λ is the regularization parameter that controls the degree of error penalty, e i is the error variable and e=[e1,e2,…,e l ] T In order to solve the above optimization problem, the Lagrange multiplier method is used. Specifically, Lagrange multipliers are added to each constraint condition in equation (12), as shown in the following equation:

[0078]

[0079] Among them, α i is a Lagrange multiplier, and α=[α1,α2,…,α l ] T By setting the partial derivatives of ω, b, e, and α in equation (13) to 0, the optimal solution of the above system of equations can be obtained.

[0080]

[0081] Eliminating ω and e in equation (14), equation (13) is only related to b and α, and the following equation can be obtained:

[0082]

[0083] Where E=[1,1,…,1] T , Ω is an l×l matrix, and make And K(x i ,x j ) is the kernel function of LSSVM. Different kernel functions can generate different least squares support vector machines. Considering the serious nonlinear characteristics of lithium-ion battery degradation data, the present invention selects radial basis function (RBF) as the kernel function of LSSVM, and the expression is as follows:

[0084]

[0085] Where σ is the kernel width. By solving equation (15), we can get α and b, and establish the following least squares support vector machine regression model:

[0086]

[0087] For the LSSVM model, the kernel width σ of the regression function and the regularization parameter λ have a significant impact on the regression performance. Using the optimization algorithm to select appropriate parameters can avoid the blindness of manual selection and improve the regression performance of the model to a certain extent.

[0088] 2. Experimental steps and results analysis

[0089] 2.1 Experimental data and steps

[0090] The lithium-ion battery degradation data set used in the experiment comes from the experimental data set provided by NASA battery aging experiments, which includes data from four groups of batteries of the same model (cell 5, cell 6, cell 7, cell 18) measured under the same experimental environment. In this experiment, the capacity degradation data of the four groups of batteries were extracted. Figure 2 As shown in Figure 1, all degradation data are within 168 cycles. When the battery capacity degrades to 70% of the rated capacity, i.e. 1.38A·h, the battery is considered to have reached the failure threshold.

[0091] Before granulation, the size of the time window should be determined. Considering that the size of the degradation data set selected for the experiment is not large, in order to make the number of samples used for training appropriate and to take advantage of the granulation as much as possible, the time window size is set to 3.

[0092] Taking cell 5 as an example, we first use fuzzy information granulation to process the time series of battery capacity degradation. The obtained granular data is as follows: Figure 3As shown in the figure, Low and Up represent the lower and upper bounds of the particles, respectively, and they are used as the training input of the model. In order to achieve better prediction results, Low and Up are normalized before modeling. Then, the first 23 particles are used as the training particles of the LSSVM model to predict the Low and Up values ​​of subsequent particles.

[0093] After the fuzzy particles are generated, they are described in language. This experiment designs five levels of language items, including lower, low, medium, high, and higher, to describe the fluctuation state of particles in the battery degradation time series. In practice, the five levels of language items are sufficient to cover the fluctuation of the battery degradation series used in this paper.

[0094] Specifically, the variance is used to calculate the degree of fluctuation of the data in each particle, and appropriate semantic labels are assigned to the first 23 particles in turn. For convenience, in the experiment, these five language items are replaced by integers from 1 to 5. Then the language description sequence of the first 23 particles is used to train the LSSVM model to predict the subsequent fluctuation state. Figure 4 is the prediction result of the language item of cell 5, which represents the capacity fluctuations of cell 5 from the 24th particle to the 56th particle. Among them, 24 particles are predicted to have a lower degree of fluctuation, accounting for the vast majority, indicating that the capacity has maintained a relatively stable degradation trend overall. Further, it can be seen that fluctuations will occur after every 3 to 5 lower particles. Extending the prediction results to the cycle layer, that is, there may be certain fluctuations after every 9 to 15 degradation cycles. This degradation trend is consistent with the actual degradation state of cell 5. This shows that the language description of particles can better grasp the changing characteristics of time series to a certain extent.

[0095] Finally, the prediction results described in language are converted into corresponding fluctuation data, which are superimposed with the above prediction results of Low and Up to obtain the final prediction results. Figure 5 The experimental results of four groups of battery degradation data are shown. The results are expanded from particles to cycle periods. It can be seen that the experiment has achieved interval prediction of lithium-ion battery capacity degradation and has predicted the degradation and regeneration of capacity well. Table 1 shows the RUL interval prediction results of three groups of batteries, cell 5, cell 6, and cell 18 (cell 7 did not reach the failure threshold within the prediction period). It can be seen from the table that the RUL interval prediction results of the three groups of batteries correctly include the true value within a narrow interval. The experimental results show that in the process of battery RUL interval prediction, the model based on fuzzy information granulation and language description methods can better grasp the fluctuation characteristics of the time series and realize the true lithium-ion battery RUL interval prediction.

[0096] Table 1. RUL interval prediction results based on fuzzy information granulation and language description

[0097] Battery Prediction starting point (cycle) Real RUL(cycle) Predicted RUL interval (cycles) Cell 5 70 58 [55,67] Cell 6 70 42 [35,54] Cell 18 70 30 [26,57]

[0098] 2.2 Experimental Results Evaluation and Analysis

[0099] In order to further evaluate the performance of the proposed interval prediction model, a performance evaluation criterion for the interval prediction model is introduced based on the above experimental results. The first is the specificity criterion (denoted as V), which is used to calculate the width of the prediction interval; the second is the coverage criterion (denoted as Q), which is used to measure the accuracy of the prediction interval. In order to ensure that the measurement standard is suitable for this experiment, the order of magnitude in the formula is slightly adjusted. Assume that X i ={x1,x2,…,x w}By T i The data of (i-th time window) is composed of the prediction interval [a i ,b i ] Specificity Criteria V i and coverage criteria Q i They are defined as follows:

[0100] V i =b i -a i (18)

[0101]

[0102]

[0103] Where w is the width of the time window, represents the jth value in the i-th time window, f i (j) For judgment Whether the prediction result of the interval is accurate. A value of 1 indicates accuracy and a value of 0 indicates inaccuracy. i , when Q i When the values ​​are equal, V i The smaller the value of is, the more accurate the prediction result of the fuzzy granular model is. i When the values ​​of are equal, Q i The larger the value of is, the more accurate the prediction result of the model is. In summary, let p be the number of time windows, and V i and Q i The determined comprehensive index P is expressed as shown in formula (21), where V i The smaller the value of Q i The larger the value of , the larger the value of P, which means the more effective the model is.

[0104]

[0105] Four battery pack The and P% values ​​are shown in Table 2. V i The mean of (i=1,…,t) represents the average interval width, Q i The mean of (i=1,…,t) represents the average coverage. As shown in Table 2, the P% values ​​of the model in the four groups of batteries are all greater than 15.3%, indicating that the overall interval prediction performance of the model is good. Four battery pack The values ​​are all less than 0.066, and the prediction interval obtained by the model is narrow, indicating that under the same coverage criterion, the model proposed in this invention is highly effective in practical applications. The value of battery cell 5, cell 6 and cell 7 All of them are above 92%, indicating that the accuracy of the model is good. In summary, the RUL interval prediction model of lithium-ion batteries based on fuzzy information granulation and language description proposed in this paper can achieve good interval prediction effect.

[0106] Table 2 Evaluation results of interval prediction performance of models based on fuzzy information granulation and language description (p = 33)

[0107]

[0108] 3. Conclusion

[0109] The present invention proposes a scheme for predicting the interval of the remaining service life of lithium-ion batteries. First, the battery degradation time series is granulated using the fuzzy information granulation method, and the research object is converted from the numerical level to the particle level to provide a basis for interval prediction. Then, in the granulation process, we found that the fuzzy information granulation may cause the loss of some fluctuation information. In view of this situation, the present invention uses a language description method to describe the fluctuation characteristics of the particles. The main advantage of this method is that the fluctuation information of the original capacity degradation data is retained as much as possible. Finally, the results of granulation and language description are used as the characteristics of the particles, and the LSSVM algorithm with high computational efficiency is used to train and predict the particles.

[0110] Experiments were conducted using four sets of battery degradation data from NASA's battery aging test. The experimental results on different battery data show that the strategy can predict the RUL interval of lithium-ion batteries and has good applicability. In addition, an evaluation criterion is introduced for the interval prediction model to evaluate the interval prediction performance of the proposed model from two aspects: interval coverage and specificity. The results show that the model based on fuzzy information granulation and language description proposed in this paper has good prediction effect.

[0111] Based on the same inventive concept, another embodiment of the present invention provides a lithium-ion battery remaining service life interval prediction system implemented by the method of the present invention, which comprises:

[0112] A time series extraction module is used to extract the battery capacity degradation time series from the battery data set;

[0113] The time series discretization module is used to discretize the battery capacity degradation time series and divide the battery capacity degradation time series into fixed-width, non-overlapping subsets, each of which is called a time window;

[0114] The time window granulation module is used to granulate the time window and generate fuzzy particles corresponding to the time window using the fuzzy information granulation method;

[0115] The language description module is used to describe the fuzzy particles in language and add a semantic label to each fuzzy particle to indicate its fluctuation state;

[0116] The interval prediction module is used to make interval predictions on the remaining battery life. Fuzzy particles with semantic labels are used as inputs, and the least squares support vector machine model is used to make interval predictions on the remaining battery life.

[0117] Based on the same inventive concept, another embodiment of the present invention provides an electronic device (computer, server, smart phone, etc.), which includes a memory and a processor, the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program includes instructions for executing each step in the method of the present invention.

[0118] Based on the same inventive concept, another embodiment of the present invention provides a computer-readable storage medium (such as ROM / RAM, disk, CD), which stores a computer program. When the computer program is executed by a computer, it implements the various steps of the method of the present invention.

[0119] The specific embodiments of the present invention disclosed above are intended to help understand the content of the present invention and implement it accordingly. It can be understood by those skilled in the art that various replacements, changes and modifications are possible without departing from the spirit and scope of the present invention. The present invention should not be limited to the contents disclosed in the embodiments of this specification, and the scope of protection of the present invention shall be subject to the scope defined in the claims.

Claims

1. A method for predicting the remaining service life of a lithium-ion battery, characterized in that: The following steps are involved: Extract battery capacity degradation time series from battery dataset; Discretize the battery capacity degradation time series and divide it into fixed-width, non-overlapping subsets, each of which is called a time window. The time window is processed by using the fuzzy information granulation method to generate fuzzy particles corresponding to the time window; Describe the fuzzy particles in language and attach semantic labels to each fuzzy particle to indicate its fluctuation state; Using fuzzy particles with semantic labels as input, a least squares support vector machine model is used to perform interval prediction of the remaining battery life. The method of processing the time window by using the fuzzy information granulation method to generate fuzzy particles corresponding to the time window includes: Select the triangular membership function to generate fuzzy particles; through fuzzy information granulation, the time window T1, T2, …, T t Generate corresponding fuzzy particles G1, G2, ..., G t Each particle G i Keep three values ​​a i 、m i and b i , these three values ​​reflect the time window T i The main information of i and b i They represent the lower and upper bounds of the generated particles, respectively, and m i represents the average level of generated particles; then, the three sequences corresponding to the blurred particles {a1, a2, …, a t }, {m1,m2,…,m t } and {b1,b2,…,b t } are represented by Low, R, and Up, respectively, where Up and Low provide upper and lower bounds for interval prediction; The language description of the fuzzy particles and adding a semantic label to each fuzzy particle to indicate its fluctuation state include: According to the change of time series, its fluctuation state is divided into several degrees to form a semantic set L = {L1, L2, ..., L c }, where L j Represents a language item, corresponding to a fluctuating state of data; The fuzzy particles G1, G2, ..., G t The fluctuation state of the data in the corresponding time window is compared with the elements in the semantic set L one by one, and the language item closest to its fluctuation state is selected for each particle in turn to describe it, generating an information feature sequence The elements in this sequence are related to the fluctuation state of the corresponding particles; The method for predicting the remaining useful life interval of a lithium-ion battery adopts a specificity criterion and a coverage criterion to evaluate the performance of interval prediction, wherein the specificity criterion is used to calculate the width of the prediction interval, and the coverage criterion is used to measure the accuracy of the prediction interval; assuming that X i ={x1,x2,…,x w }By the i-th time window T i The data consists of , and the prediction interval on it is [a i ,b i ], specificity standard V i and coverage criteria Q i They are defined as follows: V i =b i -a i Where w is the width of the time window, represents the jth value in the i-th time window, f i (j) For judgment Is the prediction result of the interval accurate? A value of 1 indicates accuracy, and a value of 0 indicates inaccuracy. Let p be the number of time windows, and V i and Q i The determined comprehensive index P is: Among them, V i The smaller the value of Q i The larger the value of , the larger the value of P, which means that the interval prediction model is more effective.

2. The method according to claim 1, characterized in that: The step of selecting a triangular membership function to generate fuzzy particles comprises: The triangular membership function A(x) is selected as follows: Among them, m is the median of the subset in the time window, a and b are the lower and upper bounds of the support of the generated fuzzy particles respectively; The construction of fuzzy particles needs to meet two conditions: (1) the fuzzy particles can reasonably represent the original information; (2) the fuzzy particles have specificity. To meet these two conditions, the constructor Q A as follows: Among them, M A Satisfy condition (1); N A represents the support boundary of A(x), satisfying condition (2); when Q A When is maximum, the above conditions reach the optimal solution.

3. The method according to claim 1, characterized in that A semantic set consisting of five levels of linguistic items, namely lower, low, medium, high and higher, is adopted to describe the fluctuation state of particles in the battery degradation time series.

4. The method according to claim 1, characterized in that: The method uses fuzzy particles with semantic labels as input and adopts a least squares support vector machine model to predict the remaining service life of the battery, including: Take the first num particles as training data and predict the particles from num+1 to t; The final interval prediction result of the battery RUL consists of two parts. The first part is the prediction result of Up and Low of the particles, and the second part is the prediction result of the language item sequence. When merging the prediction results, the prediction results of the second part, namely the language item sequence, are traced back and converted into fluctuation data of corresponding particles, which are superimposed with the prediction results of the first part to obtain the final prediction results.

5. A lithium-ion battery remaining service life interval prediction system using the method according to any one of claims 1 to 4, characterized in that: include: A time series extraction module is used to extract the battery capacity degradation time series from the battery data set; The time series discretization module is used to discretize the battery capacity degradation time series and divide the battery capacity degradation time series into fixed-width, non-overlapping subsets, each of which is called a time window; A time window granulation module is used to process the time window and generate fuzzy particles corresponding to the time window using a fuzzy information granulation method; The language description module is used to describe the fuzzy particles in language and add a semantic label to each fuzzy particle to indicate its fluctuation state; The interval prediction module is used to make interval predictions on the remaining battery life. It uses fuzzy particles with semantic labels as input and adopts the least squares support vector machine model to make interval predictions on the remaining battery life.

6. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program comprises instructions for executing the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a computer, the method according to any one of claims 1 to 4 is implemented.

Images