Method and system for predicting remaining service life of lithium ion battery

Abstract

The invention belongs to the technical field of battery remaining service life prediction, and aims to solve the problems that in the prior art, a lithium battery often operates under complex conditions and is easily influenced by working conditions and environment interference to generate unreliable data, so that uncertainty is introduced, and prediction precision is influenced. The invention provides a method for predicting the remaining service life of a lithium ion battery. Health indexes are reasonably selected through correlation analysis, abnormal data in the health indexes are cleaned through a linear interpolation method, attribute reliability is introduced into design in the matching degree calculation process, and parameters of a prediction model are optimized. And the Loess smoothing method is used for smoothing the prediction data, so that the stability and accuracy of prediction of the remaining service life of the lithium ion battery are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of lithium ion battery remaining useful life prediction, and particularly relates to a lithium ion battery remaining useful life prediction method and system. BACKGROUND

[0002] Due to the advantages of high energy density and long service life, lithium ion batteries have been applied and developed in many fields such as electric vehicles, ships and satellites. However, the performance of lithium ion batteries will decrease with the increase of use time, and if the lithium ion batteries are not replaced or maintained in time, it may lead to overheating, swelling and even explosion, resulting in extremely adverse consequences and significant economic losses. The remaining useful life (RUL) is a key state parameter for lithium ion battery health management and an important means to master the performance decline trend of the battery. In some cases, it may even affect the reliability and safety of the equipment equipped with these batteries, so it is necessary to predict the RUL of lithium ion batteries.

[0003] The current methods for predicting the RUL of lithium ion batteries can be divided into physical methods, data-driven methods and knowledge-based methods. Physical methods usually make decisions based on the measurement and analysis of internal parameters of lithium ion batteries. Chen et al. proposed a health state estimation method by establishing a linear relationship between ohmic resistance and capacity degradation. Experimental results show that the proposed method is consistent with the measured data of lithium ion batteries and has good accuracy. Safari et al. established an electrochemical model to reveal the impact of SEI film on battery capacity reduction. However, physical methods usually require a deep understanding of the chemical and physical processes inside the battery, which affects the accuracy of the evaluation results.

[0004] Data-driven methods can quickly process and analyze large amounts of data and have high prediction accuracy. Cheng et al. developed a SOH estimation and RUL prediction model using empirical mode decomposition (EMD) method and back propagation long short-term memory neural network. Experimental results show that the model has high robustness, precision and applicability. Liu et al. proposed a lithium ion battery capacity prediction method based on improved random forest, which has high prediction accuracy for low-capacity lithium batteries. However, data-driven methods are easily affected by the amount of data, and if there is not enough data, the prediction results may not be accurate.

[0005] Knowledge-based methods typically utilize existing knowledge and reasoning to solve problems or make decisions. Zhang et al. proposed a lithium-ion battery health analysis method based on evidence reasoning rules, and experimentally demonstrated the effectiveness of the health assessment model. Yin et al. proposed a lithium-ion battery SOH estimation method with expert knowledge credibility and interpretable BRB, and experimentally verified the effectiveness of the method. BRB is a good expert knowledge-based method that can handle the uncertainty and fuzziness contained in expert knowledge, and can make reliable inferences even with limited data. In addition, the IF-THEN rule-based belief rule base modeling method is more in line with human logical judgment, easy for experts to understand and reason, and has good interpretability. However, there are some challenges in lithium-ion battery RUL prediction. First, lithium batteries often operate under complex conditions and are easily affected by working conditions and environmental disturbances, resulting in unreliable data and introducing uncertainty, which affects prediction accuracy. Second, the high safety risk of battery systems requires the model to be interpretable to allow timely battery maintenance or replacement. Third, the initial model established based on expert knowledge may not be accurate and needs to be optimized for model parameters to improve prediction accuracy. Therefore, it is necessary to propose an accurate and reliable lithium-ion battery remaining useful life prediction method. SUMMARY

[0006] The present application aims to overcome the shortcomings of the prior art and provide a lithium-ion battery remaining useful life prediction method and system, which improves the stability and accuracy of lithium-ion battery remaining useful life prediction.

[0007] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is:

[0008] A lithium-ion battery remaining useful life prediction method, comprising the following steps:

[0009] S1, correlation analysis is performed on the health indicators to determine the key health indicators, and the data of the key health indicators is processed to obtain processed key health indicator data;

[0010] S2, an initial belief rule base is established based on the key health indicator data and expert knowledge;

[0011] S3, the attribute reliability of each key health indicator is calculated to obtain the attribute reliability of the key health indicators;

[0012] S4, a lithium battery remaining useful life prediction model is established based on the attribute reliability and the initial belief rule base; the present application specifically proposes a matching degree calculation method considering both attribute reliability and attribute weight, integrates the attribute reliability into the model, and then realizes evidence fusion by using the ER algorithm.

[0013] S5, optimizing parameters of the lithium battery remaining service life prediction model by the P-CMA-ES algorithm to obtain an optimized prediction model;

[0014] S6, inputting data of the lithium ion battery into the optimized prediction model to obtain a lithium ion battery predicted capacity curve;

[0015] S7, smoothing the lithium ion battery predicted capacity curve by a Loess smoothing method, and obtaining a lithium ion battery remaining service life prediction result according to the processed lithium ion battery predicted capacity curve.

[0016] Preferably, in step S1, the health indicators are the constant-voltage rise time in the charging process, the constant-voltage drop time in the discharging process, the constant-current drop time in the charging process, the maximum temperature in the charging process, the average temperature in the charging process, the difference between the maximum temperature and the minimum temperature in the charging process, the maximum temperature in the discharging process, the average temperature in the discharging process, and the difference between the maximum temperature and the minimum temperature in the discharging process.

[0017] Preferably, in step S1, the key health indicators are the constant-voltage rise time in the charging process, and the difference between the maximum temperature and the minimum temperature in the discharging process.

[0018] Preferably, step S3 comprises the following steps:

[0019] S31, assuming that the observation data of the ith attribute of each key health indicator is represented by x i (t), i = 1, 2,..., T k t = 1, 2,..., T, the number of observation data is T, and the fluctuation range is as follows:

[0020]

[0021] wherein, and σ i are the mean and standard deviation of x i (t), respectively, and τ is the fluctuation factor of the fluctuation range;

[0022] S32, if the observation data x i (t) is not within the fluctuation range, i.e., the observation data or indicates that the observation data is unreliable, and in this case, O i (t) = 1;

[0023] If the observation data x i (t) is within the fluctuation range, it indicates that the observation data is reliable, and in this case, O i (t) = 0;

[0024] S33, based on step S32, the total number of unreliable observation data is calculated, and the reliability of the ith attribute of each key health indicator is calculated according to the calculated total number of unreliable observation data, and the calculation formula of the reliability of the ith attribute is:

[0025]

[0026] Wherein, r i is the reliability of the ith attribute of the key health indicator, T is the number of observation data, O i represents the total number of unreliable observation data, and the calculation formula of the total number of unreliable observation data O i is as follows:

[0027]

[0028] Wherein, O i (t) is the number of unreliable observation data.

[0029] Preferably, step S4 comprises the following steps:

[0030] S41, based on the initial confidence rule base, the input data is converted into the expression form of the confidence distribution according to the reference value of the attribute, and the conversion expression is as follows:

[0031]

[0032]

[0033] Wherein, represents the matching degree of the reference value A k,i , x i is the input data of the ith attribute, A k,i and A k+1,i are the kth and k+1th reference values of the ith attribute, S(x i ) is the confidence distribution of the prediction result;

[0034] S42, the attribute reliability and the attribute weight are fused, and the fusion formula is as follows:

[0035]

[0036] Wherein, represents the relative attribute weight, and r i is the reliability of the ith attribute of the key health indicator, δ i represents the attribute weight, C i represents the parameter considering the attribute reliability r i and the attribute weight δ i ;

[0037] the matching degree of the kth rule k is:

[0038]

[0039] wherein, the matching degree of the ith result of the kth rule, C i represents the parameter considering attribute reliability r i and attribute weight δ i ;

[0040] S43, based on step S42, the activation weight of the kth rule is calculated, and the calculation formula is as follows:

[0041]

[0042] wherein, w k is the activation weight of the kth rule, and 0≤w k ≤1, θ k is the weight of the kth rule, L is the number of rules, α k is the matching degree of the kth rule, θ l is the weight of all rules, α l is the matching degree of all rules;

[0043] S44, based on step S43, the activation rules are fused by using the evidence reasoning method to obtain the confidence of the prediction result D n , and the calculation formula of the confidence is as follows:

[0044]

[0045] wherein, β n is the confidence of the nth prediction result D n , C is the confidence of the nth result of the kth rule, is the confidence of the ith result of the kth rule, N is the number of prediction results, w k is the activation weight of the kth rule, and L is the number of rules;

[0046] S45, based on step S41, and combined with the confidence of the prediction result Dn obtained in step S44, a lithium battery remaining service life prediction model is established, and the expression of the lithium battery remaining service life prediction model is as follows:

[0047]

[0048] wherein, u(S(x i )) is the utility of the prediction result confidence distribution, u(D n ) is the prediction result Dn the utility of the kth rule, y is the final utility value, β n is the confidence of the prediction result D n , and N is the number of prediction results.

[0049] Preferably, in step S45, the kth rule of the lithium battery remaining service life prediction model is expressed as follows:

[0050]

[0051] with rule weightθ k (k=1,2,...,L),

[0052] attribute weightδ i (i=1,2,...,T k ),

[0053] and attribute reliability r i (i=1,2,...,T k ),

[0054] wherein x1, x2, …, x Tk represent attributes of the lithium battery remaining service life prediction model, is the reference value of the ith attribute, θ k is the weight of the kth rule, δ i (i=1,2,...,T k ) is the weight of the ith attribute. L is the number of rules, r i represents the attribute reliability, T k is the number of attributes, D n is the prediction result, is the confidence of the prediction result D n .

[0055] Preferably, step S5 comprises the following steps:

[0056] S51, initializing parameters of the lithium battery remaining service life prediction model;

[0057] S52, executing a sampling program to generate a population;

[0058] S53, projecting the solution onto a hyperplane based on the population generated in step S52;

[0059] S54, based on step S53, performing a selection operation to calculate the average value of the updated population, the calculation formula is as follows:

[0060]

[0061] wherein, is the weight coefficient of the i-th solution, represents the i-th solution in the λ solutions of the g+1 generation, m (g+1) represents the average value of the population of the g+1 generation;

[0062] S55, performing an adaptive operation on the average value of the population obtained in step S54 to update the covariance matrix;

[0063] S56, repeating steps S52-S55 until the optimal solution Ω optimal is obtained, stopping the optimization process, and obtaining the optimized prediction model.

[0064] Preferably, in step S7, the expression of the lithium-ion battery remaining useful life prediction result is:

[0065] RUL pre =smooth(y,τ)

[0066] wherein, smooth(·) represents a Loess smoothing method, τ is a smoothing coefficient, and RUL pre represents the lithium-ion battery remaining useful life prediction result.

[0067] The present application also provides a lithium-ion battery remaining useful life prediction system for executing the above-mentioned lithium-ion battery remaining useful life prediction method, comprising:

[0068] a correlation analysis module for performing correlation analysis on the health indicators, determining the key health indicators, and processing the data of the key health indicators to obtain processed key health indicator data;

[0069] an initial confidence rule base establishment module for establishing an initial confidence rule base based on the key health indicator data and expert knowledge;

[0070] an attribute reliability calculation module for calculating the attribute reliability of each key health indicator to obtain the attribute reliability of the key health indicators;

[0071] a model construction module for establishing a lithium-ion battery remaining useful life prediction model based on the attribute reliability and the initial confidence rule base;

[0072] a model optimization module for optimizing the parameters of the lithium-ion battery remaining useful life prediction model through a P-CMA-ES algorithm to obtain an optimized prediction model,

[0073] The prediction calculation module is used for inputting data of the lithium ion battery into the optimized prediction model, obtaining a lithium ion battery prediction capacity curve, performing smoothing processing on the lithium ion battery prediction capacity curve by using a Loess smoothing method, and obtaining a lithium ion battery remaining use life prediction result according to the processed lithium ion battery prediction capacity curve.

[0074] Compared with the prior art, the present application has the beneficial effects that:

[0075] The present application reasonably selects health indexes through correlation analysis, cleanses abnormal data in the health indexes by using a linear interpolation method, improves the correlation of the data, introduces attribute reliability into the design of the matching degree calculation process, reduces the interference of unreliable data, reduces the influence of environmental interference, and thus obtains more accurate and reliable prediction capacity. BRIEF DESCRIPTION OF DRAWINGS

[0076] Figure 1 The flowchart of the lithium ion battery remaining use life prediction method provided by the present application embodiment;

[0077] Figure 2 The flowchart of the attribute reliability calculation of the health index;

[0078] Figure 3 The optimization flowchart of the lithium battery remaining use life prediction model;

[0079] Figure 4 Data cleaning of HI1 by using a linear interpolation method;

[0080] Figure 5 The data of the lithium battery capacity and the key health index;

[0081] Figure 6 The capacity prediction results of different models at SP=80 and SP=100 respectively;

[0082] Figure 7 The RUL prediction results of different models at SP=80 and SP=100 respectively. DETAILED DESCRIPTION

[0083] The specific implementation of the present application will be described below with reference to the accompanying drawings. Figures 1 to 7The technical solutions in the embodiments of the present application are described clearly and completely, obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the present application.

[0084] I. Problem formation

[0085] Problem 1: In engineering practice, the reliability of data may be disturbed by the complexity of the environment. At the same time, the noise in the actual environment also aggravates the fluctuation of data. These factors increase the uncertainty and affect the accuracy of the lithium-ion battery RUL prediction model. The reliable attribute means that its observation data is not easily affected by the external environment, and has high credibility. The expression of the reliability of the attribute is as follows:

[0086] r i = φ(x i )

[0087] Where r i represents the reliability of the ith attribute x i , and φ(·) represents the method of calculating the reliability of the attribute.

[0088] Problem 2: Although experts have rich professional knowledge, they face many difficulties in directly constructing an accurate lithium-ion battery RUL prediction model. Therefore, the lithium-ion battery RUL prediction model constructed in the initial stage is likely to be inaccurate, so it is necessary to design an optimization method to optimize the model parameters to improve the model precision. The expression of the optimized model parameters is as follows:

[0089] ε = optimize(q)

[0090] Where q represents the parameter set of the optimization algorithm, ε represents the optimized parameters, such as: confidence, rule weight and attribute weight; optimize(·) represents the optimization function.

[0091] II. Propose of the lithium-ion battery remaining useful life prediction method

[0092] Based on the above two problems, in order to improve the accuracy and reliability of the prediction, the embodiments of the present application provide a lithium-ion battery remaining useful life prediction method, which is based on BRB and considers the attribute reliability, and the prediction model expression is as follows:

[0093] y = f(x, r i , E, ε)

[0094] Where x represents the input of the method, y represents the prediction result, and E represents the expert knowledge.

[0095] Embodiment 1

[0096] As Figure 1 shown, the lithium ion battery remaining service life prediction method provided by the embodiment of the application specifically comprises the following steps:

[0097] S1, correlation analysis is performed on the health indicators, specifically, the key health indicators are determined by using Spearman and Pearson correlation coefficients, the data of the key health indicators are processed, specifically, the abnormal data in the key health indicators are cleaned by using a linear interpolation method to improve the correlation, and the processed key health indicator data are obtained; the key health indicators in the embodiment of the application are the isobaric rise time in the charging process and the difference between the maximum temperature and the minimum temperature in the discharging process.

[0098] The application extracts 9 health indicators reflecting the health state of the lithium ion battery from the voltage, current and temperature curves in the charging and discharging process, and performs correlation analysis on the 9 health indicators. The 9 health indicators are as follows:

[0099] The time interval of equal charging voltage difference (HI1) in the charging process, the maximum and minimum temperature difference during the discharge process (HI2), the time interval of equal charging current difference (HI3) in the charging process, the maximum charging temperature (HI4), the average charging temperature (HI5), the maximum and minimum temperature difference during the charge process (HI6), the maximum discharging temperature (HI7), the average discharging temperature (HI8), and the time interval of equal discharging voltage difference (HI9) in the discharging process.

[0100] Spearman's coefficient and Pearson's correlation coefficient are both used to measure the correlation between two variables. Their values range from -1 to 1. Generally, when the absolute value of Spearman's correlation coefficient and Pearson's correlation coefficient is above 0.8, it is considered that there is a strong correlation between the two variables. Using them can quantitatively evaluate the correlation degree of HIs and SOH. The correlation analysis results of the 9 health indicators (health indicators, HIs) are shown in Table 1 as follows.

[0101] Table 19 health indicators correlation analysis results

[0102]

[0103] From the analysis results in Table 1, it can be seen that health indicators HI1, HI2, HI7, HI8 and HI9 all show significant correlation with the battery state of health (SOH). In order to comprehensively characterize the aging process of lithium batteries and accurately describe its decay trend, the selection of health indicators needs to consider the following two key factors: one is the correlation with environmental factors such as temperature and time, and the other is the relevance to the dynamic characteristics of the charging and discharging process. Based on the above principles and combined with the correlation analysis results, HI1 and HI2 are finally selected as the key health indicators for evaluating the SOH of the battery. These two indicators effectively reflect the aging characteristics of the battery, fully consider temperature, time, charging process, and discharging process, and are very representative and important.

[0104] S2, an initial confidence rule base is established based on the key health indicator data and expert knowledge;

[0105] S3, the attribute reliability of each key health indicator is calculated to obtain the attribute reliability of the key health indicator. In engineering practice, observation data is affected by some disturbance factors and will fluctuate, thereby reducing its reliability and further affecting the accuracy of RUL prediction. Attribute reliability is an important indicator reflecting the influence of disturbance factors on observation data, which indicates the reliability of observation data in reflecting the true system characteristics. The present application uses statistical methods to analyze observation data and quantify the degree of fluctuation and error caused by disturbance factors; the calculation process of the attribute reliability of the health indicator is as shown in Figure 2 , which specifically includes the following steps:

[0106] S31, assuming that the observation data of the i-th attribute of each key health indicator is represented by x i (t), i = 1, 2,..., T k , t = 1, 2,..., T, the number of observation data is T, and the fluctuation range is as follows:

[0107]

[0108] wherein, and σ i are the mean and standard deviation of x i (t), respectively, and τ is the volatility factor of the volatility range;

[0109] S32, if the observation data x i (t) is not within the volatility range, i.e., the observation data or indicates that the observation data is unreliable, at this time O i (t) = 1;

[0110] If the observation data x i (t) is within the volatility range, it indicates that the observation data is reliable, at this time O i (t) = 0;

[0111] S33, based on step S32, the total number of unreliable observation data is calculated, and the reliability of the ith attribute of each key health indicator is calculated according to the calculated total number of unreliable observation data, and the calculation formula of the reliability of the ith attribute is:

[0112]

[0113] wherein, r i is the reliability of the ith attribute of the key health indicator, T is the number of observation data, O i represents the total number of unreliable observation data, and the calculation formula of the total number of unreliable observation data O i is as follows:

[0114]

[0115] wherein, O i (t) is the number of unreliable observation data.

[0116] S4, based on the attribute reliability and the initial confidence rule base, a lithium battery remaining useful life prediction model is established, specifically, a matching degree calculation method considering attribute reliability and attribute weight is proposed, the attribute reliability is integrated into the model, and then the ER algorithm is used to realize evidence fusion; comprising the following steps:

[0117] S41, based on the initial confidence rule base, according to the reference value of the attribute, the input data is converted into the expression form of the confidence distribution, and the conversion expression is as follows:

[0118]

[0119] wherein, represents the matching degree of the reference value A k,i , x i is the input data of the ith attribute, and Ak,i and A k+1,i are the kth and k+1th reference values of the ith attribute respectively, set by experts, S(x i ) is the confidence distribution of the prediction result;

[0120] S42, fuse the attribute reliability and the attribute weight, and the fusion formula is as follows:

[0121]

[0122] wherein, represents the relative attribute weight, and r i is the reliability of the ith attribute of the key health indicator, δ i represents the attribute weight, C i represents the parameter considering the attribute reliability r i and the attribute weight δ i ;

[0123] Then the matching degree α k of the kth rule is:

[0124]

[0125] wherein, is the matching degree of the ith result of the kth rule, C i represents the parameter considering the attribute reliability r i and the attribute weight δ i ;

[0126] S43, based on step S42, the activation weight of the kth rule is calculated, and the calculation formula is as follows:

[0127]

[0128] wherein, w k is the activation weight of the kth rule, and 0≤w k ≤1, θ k is the weight of the kth rule, L is the number of rules, α k is the matching degree of the kth rule, θ l is the weight of all rules, and α l is the matching degree of all rules;

[0129] S44, based on step S43, the activation rules are fused by using the evidence reasoning method. The confidence degree of the prediction result D n is obtained, and the calculation formula of the confidence degree is as follows:

[0130]

[0131] wherein, βn the confidence of the nth prediction result D n the confidence of the nth result of the kth rule, the confidence of the ith result of the kth rule, N is the number of prediction results, w k is the activation weight of the kth rule, and L is the number of rules.

[0132] S45, based on step S41, and the confidence of the prediction result D n obtained in step S44, a lithium battery remaining service life prediction model is established, and the expression of the lithium battery remaining service life prediction model is as follows:

[0133]

[0134] wherein u(S(x i )) is the utility of the prediction result confidence distribution, u(D n ) is the utility of the prediction result D n , y is the final utility value, β n is the confidence of the prediction result D n , and N is the number of prediction results.

[0135] The kth rule of the lithium battery remaining service life prediction model is expressed as follows:

[0136]

[0137] with rule weightθ k (k=1,2,...,L),

[0138] attribute weightδ i (i=1,2,...,T k ),

[0139] and attribute reliability r i (i=1,2,...,T k ),

[0140] wherein, represents the attribute of the lithium battery remaining service life prediction model, is the reference value of the ith attribute, θ k is the weight of the kth rule, δ i (i=1,2,...,T k ) is the weight of the ith attribute. L is the number of rules, r i represents the attribute reliability, T k is the number of attributes, and D​n for predicting the result, for predicting the result D n of the confidence.

[0141] S5, parameters of the lithium battery RUL remaining useful life prediction model are optimized through a P-CMA-ES algorithm, and an optimized prediction model is obtained;

[0142] As Figure 3 shown, the embodiment of the application optimizes the lithium battery RUL remaining useful life prediction model through a projection covariance matrix adaptive evolutionary strategy (P-CMA-ES) algorithm. The purpose of the optimization process is to improve the performance of the BRB by finding the optimal model parameters. In the model, the attribute weight, the rule weight and the confidence are all optimization parameters, and they should satisfy the following constraint conditions:

[0143] 1. The rule weight satisfies 0 ≤ θ k ≤ 1, (k = 1, 2,..., L);

[0144] 2. The confidence of the prediction result D n satisfies The sum of the confidences satisfies

[0145] 3. The initial attribute weight given by the expert satisfies 0 ≤ δ i ≤ 1, (i = 1, 2,..., T k ).

[0146] As Figure 3 shown, step S5 specifically includes the following steps:

[0147] S51, parameters of the lithium battery remaining useful life prediction model are initialized; the iteration number G, the population size λ, the offspring population size μ, the covariance matrix C 0 and the step size ε 0 are specifically set;

[0148] S52, a sampling program is executed to generate a population;

[0149] S53, based on the population generated in step S52, the solution is projected onto a hyperplane;

[0150] S54, based on step S53, a selection operation is performed to calculate the mean value of the updated population, and the calculation formula is as follows:

[0151]

[0152] In the formula, is the weight coefficient of the i th solution, represents the i-th solution in the g+1-th generation of the lambda solutions, m (g+1) represents the average value of the population of the g+1-th generation;

[0153] S55, performing an adaptive operation on the average value of the population obtained in step S54 to update the covariance matrix;

[0154] S56, repeating steps S52-S55 until the optimal solution Ω optimal is obtained, stopping the optimization process, and obtaining the optimized prediction model.

[0155] S6, inputting the data of the lithium ion battery into the optimized prediction model to obtain a lithium ion battery predicted capacity curve;

[0156] S7, performing smoothing processing on the lithium ion battery predicted capacity curve using a Loess smoothing method, and obtaining a lithium ion battery remaining useful life prediction result according to the processed lithium ion battery predicted capacity curve. Loess is a method of using local weighted linear regression for data smoothing, and the smooth curve generated by the Loess method is relatively smooth, which can better show the trend of data change and is easy to interpret and understand. Therefore, in order to obtain an accurate RUL, the Loess smoothing method is used to smooth the expected utility, and the expression of the lithium ion battery remaining useful life prediction result is:

[0157] RUL pre =smooth(y,τ)

[0158] Wherein, smooth(·) represents the Loess smoothing method, τ is the smoothing coefficient, and RUL pre represents the lithium ion battery remaining useful life prediction result.

[0159] Embodiment 2

[0160] The application also provides a lithium ion battery remaining useful life prediction system for executing the above-mentioned lithium ion battery remaining useful life prediction method, comprising:

[0161] A correlation analysis module is configured to perform correlation analysis on the health indicators, determine the key health indicators, and process the data of the key health indicators to obtain processed key health indicator data.

[0162] An initial confidence rule base establishment module is configured to establish an initial confidence rule base based on the key health indicator data and expert knowledge.

[0163] An attribute reliability calculation module is configured to calculate the attribute reliability of each key health indicator to obtain the attribute reliability of the key health indicators.

[0164] A model construction module is configured to establish a lithium battery remaining useful life prediction model based on attribute reliability and an initial confidence rule base.

[0165] A model optimization module is configured to optimize parameters of the lithium battery remaining useful life prediction model by a P-CMA-ES algorithm to obtain an optimized prediction model.

[0166] A prediction calculation module is configured to input data of the lithium ion battery into the optimized prediction model to obtain a lithium ion battery prediction capacity curve, perform smoothing processing on the lithium ion battery prediction capacity curve by using a Loess smoothing method, and obtain a lithium ion battery remaining useful life prediction result according to the processed lithium ion battery prediction capacity curve.

[0167] III. Prediction performance research

[0168] The prediction performance of the prediction method proposed in Embodiment 1 of the present application is evaluated by using mean square error (MSE) and absolute error (AE) respectively, and the specific MSE represents the accuracy of the predicted capacity, and the calculation formula is as follows:

[0169]

[0170] wherein, and y i represent the predicted capacity and the actual capacity respectively, and n is the number of observation data.

[0171] The AE represents the accuracy of the predicted RUL, and the calculation formula is as follows:

[0172] AE = | RUL act -RUL pre |

[0173] wherein, RUL act and RUL pre represent the actual RUL and the predicted RUL when the battery capacity reaches the failure threshold respectively.

[0174] 1. Experimental data

[0175] The NASA lithium battery data set is used to verify the effectiveness of the prediction method proposed in the present application. The detailed process of the lithium battery B0005 test is shown as follows:

[0176] The whole test was carried out at room temperature 24 degrees. The battery was charged at a constant current of 1.5A (CC mode). Once the voltage reaches 4.2V, the charging mode is switched to constant voltage (CV) mode until the charging current drops to 0.02A. The battery is discharged at a current of 2A until the voltage drops to 2.7V. The repeated charge-discharge cycle will cause the accelerated aging of the battery. When the battery reaches the end of life (EOL) standard, the experiment stops, that is, the rated capacity decreases by 30%, that is, from 2Ah to 1.4Ah.

[0177] Through the correlation analysis between the attributes and the capacity, the isochronous rise time HI1 of the charging process and the difference HI2 between the maximum temperature and the minimum temperature in the discharging process are selected as the key health indicators. The distribution of the battery capacity of the lithium ion battery and the correlation coefficient of the two key health indicators are shown in Figure 4 、 Figure 5 and Table 2 below. As can be seen from Table 2, the absolute values of the Spearman correlation coefficient and the Pearson correlation coefficient of HI1 and HI2 are greater than 0.95, indicating that the two health indicators have strong correlation with the battery capacity, and are suitable for predicting the RUL of the lithium battery.

[0178] Table 2 Spearman coefficient and Pearson coefficient of HI1 and HI2

[0179]

[0180] However, it is found in the experiment that there are zero values and abnormal values in HI1. The linear interpolation method is a simple and commonly used interpolation method, which has the advantages of strong interpretability and good stability, therefore, the linear interpolation method is used to process the data, from Figure 4 and Figure 5 It can be seen from and that after processing, HI1 has better linearity, which is conducive to improving the prediction accuracy.

[0181] 2, Constructing initial BRB

[0182] The initial BRB constructed by expert knowledge is named BRB0, and the BRB optimized by PCMAES algorithm is named BRB1. Using expert knowledge, HI1 and HI2 are set as input attributes, each attribute has three reference levels, high (H), medium (M), and low (L); the initial weight of each attribute is 1, as shown in Table 2. In the BRB0 model, in order to evaluate the health state of the lithium battery, three reference points are set: safe (S), normal (N), and bad (B), as shown in Table 3. According to the Cartesian product calculation method, the number of rules is 3x3=9. Table 4 shows the initial parameters of the rule base, the initial weight value of all rules is 1, and the initial confidence distribution is set based on expert knowledge and experience.

[0183] Initial weights and reference levels of attributes in Table 3

[0184]

[0185] Reference levels of evaluation results in Table 4

[0186]

[0187] Initial parameters of rule base in Table 5

[0188]

[0189] 3. Model optimization

[0190] In the embodiment of the present application, the range of fluctuation is selected as In the P-CMA-ES algorithm, the step size is 0.001 and the iteration number is 50. The attribute reliabilities of HI1 and HI2 are 0.4940 and 0.5060 respectively. After training of the optimized model, the weights of the two attributes are 0.9935 and 0.9963 respectively. The optimized rule weights and confidence distribution are shown in Table 6.

[0191] Table 6. Optimized rule weights and confidence distribution

[0192]

[0193] 4. Experimental results and analysis

[0194] The first 80 and 100 data of B0005 battery in NASA dataset are selected for training, and the rest of the data are used as the test part. Comparative experiments are carried out by using SVM, LR, BRB0 and BRB1. The starting point of prediction is the 80th and 100th cycle.

[0195] It is found from Figure 6 that the prediction accuracy of SVM and LR is poor, and is greatly affected by the number of training sets. The prediction result of starting point (SP) = 100 is more close to the actual capacity value compared with the prediction result of SP = 80, which shows that the SVM and LR methods strongly depend on the number of training sets. BRB0 and BRB1 are not affected by the number of training sets, and maintain high prediction accuracy. However, the prediction capacity curve of BRB0 and BRB1 fluctuates greatly, which affects the RUL prediction. Therefore, the Loess smoothing method is used to smooth BRB0 and BRB1 to obtain more accurate RUL values.

[0196] From Figure 7As can be seen, the failure threshold of the battery is 1.4 Ah, and the corresponding EOL is the 125th cycle. If the current cycle number is the 80th cycle, then their RUL is 125-80 = 45. In addition, the Loess smoothing method used can accurately capture the fluctuation characteristics of the curve and effectively retain the overall trend and change rule of the data while smoothing out irrelevant local fluctuations, so that the data presents a more stable and continuous curve, which is easy to analyze and understand, and more accurate RUL prediction results can be obtained. The prediction performance comparison results of the four models are shown in Table 7.

[0197] Table 7 Prediction performance comparison of four models based on B0005 battery data

[0198]

[0199] From the experimental results in Table 7, it can be seen that when SP = 80, the RUL prediction results of SVM and LR are poor, with AE of 26 and 30 respectively. When SP = 100, their prediction results are better, with AE of 2 and 7 respectively. This shows that the two methods have high requirements for data volume and are limited. Although SVM has AE = 2 when SP = 100, it has poor interpretability. The prediction accuracy of LR is the worst among all models. BRB0 and BRB1 have AE of 4 and 3 respectively when SP = 80. When SP = 100, AE is 2 and 0 respectively. This shows that unlike other machine learning methods that require a large amount of data to obtain reliable prediction results, the method proposed in the present application is less affected by the sample size of the training set and can produce accurate prediction results even with a small training set sample size. This is because the method proposed in the present application incorporates expert knowledge into the decision-making process, and the level reference value, rule weight, attribute weight and initial confidence are set based on the expert's domain knowledge and experience, which clearly represents the causal relationship and reasoning mechanism and has better interpretability. In addition, by introducing attribute reliability, the uncertainty caused by environmental disturbance can be better handled, and more reliable and accurate results can be obtained.

[0200] In summary, the present application selects the health indicators that are strongly correlated with capacity by Spearman and Pearson correlation analysis, and then uses linear interpolation method to clean up the abnormal data of the health indicators to improve the prediction accuracy. By introducing attribute reliability into the reasoning process, environmental disturbance is reduced, thereby obtaining more accurate and reliable prediction results. By optimizing the parameters of the model, the present application can improve the uncertainty caused by expert knowledge modeling. By using the Loess smoothing method to smooth the prediction data, the stability and accuracy of the model are improved. Experimental results show that compared with other models, the method proposed in the present application can accurately and effectively predict the RUL of lithium batteries.

[0201] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely divergences of the principles and application of the present application and that numerous modifications, changes, substitutions, and alterations can be made thereto without departing from the spirit and scope of the present application, which is defined by the following claims and their equivalents.

Claims

1. A method for predicting the remaining useful life of a lithium-ion battery, characterized by, The method comprises the following steps: S1, performing correlation analysis on the health indicators to determine the key health indicators, and processing the data of the key health indicators to obtain processed key health indicator data; S2, establishing an initial confidence rule base based on the key health indicator data and expert knowledge; S3, calculating the attribute reliability of each key health indicator to obtain the attribute reliability of the key health indicators; S4, establishing a lithium battery remaining useful life prediction model based on the attribute reliability and the initial confidence rule base; S5, optimizing the parameters of the lithium battery remaining useful life prediction model through a P-CMA-ES algorithm to obtain an optimized prediction model; S6, inputting the data of the lithium ion battery into the optimized prediction model to obtain a lithium ion battery predicted capacity curve; S7, performing smoothing processing on the lithium ion battery predicted capacity curve by using a Loess smoothing method, and obtaining a lithium ion battery remaining useful life prediction result according to the processed lithium ion battery predicted capacity curve.

2. The lithium-ion battery remaining useful lifetime prediction method of claim 1, wherein, In step S1, the health indicators are the constant voltage rising time in the charging process, the constant voltage falling time in the discharging process, the constant current falling time in the charging process, the maximum temperature in the charging process, the average temperature in the charging process, the difference between the maximum temperature and the minimum temperature in the charging process, the maximum temperature in the discharging process, the average temperature in the discharging process, and the difference between the maximum temperature and the minimum temperature in the discharging process.

3. The lithium-ion battery remaining useful lifetime prediction method of claim 1, wherein, In step S1, the key health indicators are the constant voltage rising time in the charging process and the difference between the maximum temperature and the minimum temperature in the discharging process.

4. The lithium-ion battery remaining useful lifetime prediction method of claim 1, wherein, Step S3 comprises the following steps: S31, assuming that the observation data of the i-th attribute of each key health indicator is x i (t), i = 1, 2,... T k t = 1, 2,..., T, the number of observation data is T, and the range of fluctuation is as follows: wherein and σ i are the mean and standard deviation of x i (t), respectively, and τ is the volatility factor of the volatility range. S32, if the observation data x i (t) is not in the fluctuation range, i.e. the observation data or indicates that the observation data is unreliable, at which time O i (t) = 1; If the observed data x i (t) is within the fluctuation range, indicating that the observed data is reliable, then O i (t) = 0. S33, based on step S32, calculating the total number of unreliable observation data, and calculating the reliability of the i-th attribute of each key health indicator according to the calculated total number of unreliable observation data, and the calculation formula of the reliability of the i-th attribute is as follows: where r i is the reliability of the i-th attribute of the key health indicator, T is the number of observations, O i represents the total number of unreliable observations, the total number of unreliable observations O i is calculated as follows: where O i (t) is the number of unreliable observations.

5. The lithium-ion battery remaining useful lifetime prediction method of claim 1, wherein, Step S4 comprises the following steps: S41, based on the initial confidence rule base, converting the input data into an expression form of the confidence distribution according to the reference value of the attribute, and the converted expression is as follows: wherein, represents a reference value A k,i , x i is the input data of the i-th attribute, A k,i and A k+1,i are the k-th and k+1-th reference values of the i-th attribute, respectively, S(x i ) is the confidence distribution of the prediction result; S42, fusing the attribute reliability and the attribute weight, and the fusion formula is as follows: wherein, represents a relative attribute weight, and r i is the reliability of the i-th attribute of the key health indicator, δ i represents an attribute weight, C i represents a parameter that takes into account the attribute reliability r i and the attribute weight δ i . Then the matching degree a of the kth rule is k is: wherein, is the matching degree of the i-th result of the k-th rule, C i denotes parameters taking into account the attribute reliability r i and the attribute weight δ i . S43, based on step S42, calculating the activation weight of the k-th rule, and the calculation formula is as follows: where w k is the activation weight of the kth rule, and 0≤w k ≤1, θ k is the weight of the kth rule, L is the number of rules, α k is the matching degree of the kth rule, θ l is the weight of all rules, α l is the matching degree of all rules; S44, based on step S43, the evidence reasoning method is used to fuse the activation rules. The prediction result D is obtained n The confidence of the prediction result D is calculated according to the following formula: wherein β n is the confidence of the nth prediction result D n , is the confidence of the nth result of the kth rule, is the confidence of the ith result of the kth rule, N is the number of prediction results, w k is the activation weight of the kth rule, and L is the number of rules. S45, based on step S41, and the prediction result D obtained in step S44 n The confidence level of the prediction model of the remaining service life of the lithium battery is established, and the expression of the prediction model of the remaining service life of the lithium battery is as follows: where u(S(x i )) is the utility of the predictive outcome confidence distribution, u(D n ) is the utility of the predictive outcome D n , y is the final utility value, β n is the confidence of the predictive outcome D n , and N is the number of predictive outcomes.

6. The lithium-ion battery remaining useful lifetime prediction method of claim 5, wherein, In step S45, the k-th rule of the lithium battery remaining useful life prediction model is represented as follows: with rule weight θ k (k = 1, 2,..., L), attribute weight δ i (i = 1, 2,..., T k ), and attribute reliability r i (i = 1, 2,..., T k ), in, This indicates the properties of the lithium battery remaining lifespan prediction model. θ is the reference value for the ith-th attribute. k The weight of the kth rule, δ i (i = 1, 2, ..., T) k Let r be the weight of the ith attribute. L is the number of rules, and r is the weight of the ith attribute. i T represents the reliability of the attribute. k D represents the number of attributes. n For the predicted results, For the predicted result D n The confidence level.

7. The lithium-ion battery remaining useful lifetime prediction method of claim 1, wherein, Step S5 comprises the following steps: S51, initializing the parameters of the lithium battery remaining useful life prediction model; S52, executing a sampling program to generate a population; S53, based on the population generated in step S52, projecting the solution onto a hyperplane; S54, based on step S53, performing a selection operation to calculate the average value of the updated population, and the calculation formula is as follows: wherein is the weight coefficient of the i-th solution, represents the i-th solution of the λ solutions of the g+1 generation, m (g+1) represents the average of the population of the g+1 generation; S55, performing an adaptive operation according to the average value of the population obtained in step S54 to update the covariance matrix; S56, repeat steps S52-S55 until the optimal solution Ω is obtained optimal When the optimization process is stopped, the optimized prediction model is obtained.

8. The lithium-ion battery remaining useful lifetime prediction method of claim 1, wherein, In step S7, the expression of the lithium ion battery remaining useful life prediction result is as follows: RUL pre = smooth(y, τ) wherein, smooth(·) represents the Loess smoothing method, τ is the smoothing coefficient, RUL pre represents the prediction result of the remaining useful life of the lithium-ion battery.

9. A lithium-ion battery remaining useful lifetime prediction system, characterized by, A device for performing the lithium ion battery remaining useful life prediction method of any one of claims 1-8, comprising: The correlation analysis module is configured to perform correlation analysis on the health indicators, determine key health indicators, and process data of the key health indicators to obtain processed key health indicator data. The initial confidence rule base establishment module is configured to establish an initial confidence rule base based on the key health indicator data and expert knowledge. The attribute reliability calculation module is configured to calculate attribute reliability of each key health indicator to obtain attribute reliability of the key health indicators. The model construction module is configured to establish a lithium battery remaining service life prediction model based on the attribute reliability and the initial confidence rule base. The model optimization module is configured to optimize parameters of the lithium battery remaining service life prediction model by using a P-CMA-ES algorithm to obtain an optimized prediction model. The prediction calculation module is configured to input data of the lithium ion battery into the optimized prediction model to obtain a lithium ion battery predicted capacity curve, perform smoothing processing on the lithium ion battery predicted capacity curve by using a Loess smoothing method, and obtain a lithium ion battery remaining service life prediction result according to the processed lithium ion battery predicted capacity curve.

Images