Lithium battery residual life prediction method fusing physical mechanism and uncertainty characterization

Abstract

The invention discloses a lithium battery residual life prediction method fusing a physical failure mechanism and uncertainty characterization, and the method comprises the steps: obtaining the capacity data of a historical lithium battery sample under different environment stress conditions through accelerating the life experiment of the historical lithium battery sample; then, calculating prior parameters of the generalized Eyring model of the target lithium battery by extracting life characteristics of historical lithium battery samples; then, adaptively fitting the degradation trend of the lithium battery through a long-short-term memory neural network, updating the mean value and variance of the drift coefficient of the model by adopting a Kalman filtering algorithm, and calculating the residual life probability density function of the target lithium battery sample at the current moment by using the updated parameters; the method has the characteristics of high prediction precision, good real-time performance, strong adaptive capability and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of reliability analysis, and more specifically, relates to a method for predicting the remaining life of a lithium battery by integrating physical failure mechanism and uncertainty characterization. Background Art

[0002] In recent years, due to the impact of environmental pollution and energy crisis, the field of electric vehicles has received widespread attention from all over the world. As the core module of electric vehicles, the reliability level of lithium batteries will greatly affect the reliability of the entire electric vehicle system. Therefore, the research on the remaining useful life (RUL) prediction technology of lithium batteries is particularly important. Its significance is reflected in the following aspects: (1) It is a key way to obtain reliability information of electric vehicles and provides a basis for online monitoring and health management of electric vehicles; (2) It helps manufacturers improve the process of lithium battery modules (including introducing new materials and optimizing packaging technology); (3) It is conducive to designing more accurate accelerated aging tests for lithium batteries, thereby obtaining more accurate aging data; (4) It can realize condition-based maintenance for electric vehicles, helping end users to obtain more information about the life of electric vehicles and reduce maintenance costs.

[0003] Existing lithium battery RUL prediction technologies are mainly divided into two categories: prediction technology driven by physical models and prediction technology driven by data. Although the prediction technology based on physical models can describe the RUL information of system modules from the perspective of materials, it requires an in-depth understanding of the characteristics and process of lithium battery manufacturing materials, and the resulting physical models are often not accurate enough, and the nonlinear relationship of parameters in the modeling process is easily lost, resulting in increased prediction errors. The data-driven prediction technology establishes a mapping relationship between input and output by learning the historical aging data of the output characteristics of lithium battery modules. It is mainly divided into probability statistics-based methods and machine learning-based methods. The machine learning-based method can use the powerful computing power of computers to accurately predict the degradation trend of lithium batteries, and shows high accuracy in RUL prediction, but the accuracy of this method is highly dependent on a large amount of training data, and it is difficult to describe the uncertainty in the prediction results; the probability statistics-based method uses statistical models to describe the degradation trend of lithium batteries, which can better reflect the uncertainty in its degradation process, but the probability density function of its remaining life is often difficult to solve. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for predicting the remaining life of a lithium battery that integrates physical failure mechanism and uncertainty characterization. The method combines the advantages of random process models, artificial intelligence models and physical failure models of lithium batteries, and adopts a generalized Eyring model to quantitatively characterize the influence of different environmental stresses on the degradation rate of lithium batteries. At the same time, a long short-term memory neural network is used to adaptively fit the complex nonlinear degradation process of lithium batteries, thereby enhancing the adaptive fitting ability of the Wiener process for lithium battery degradation modeling under different working conditions, thereby improving the accuracy of the remaining life prediction.

[0005] To achieve the above-mentioned purpose of the invention, the present invention provides a method for predicting the remaining life of a lithium battery integrating physical mechanism and uncertainty characterization, characterized in that it comprises the following steps:

[0006] (1) Obtain the capacity value of the target lithium battery at different times;

[0007] By accelerating the life experiment of M historical lithium batteries, the capacity data of each historical lithium battery at different times and its corresponding K-type working environment stress and temperature are sampled;

[0008] Among them, the capacity data of the i-th historical lithium battery at different times is recorded as The working environment stress is recorded as S i =[s i,1 ,s i,2 ,…,s i,k ,…,s i,K ] and the temperature is denoted as T i , where i = 1, 2, ... M, k = 1, 2, ..., K, s i,k represents the kth type of environmental stress of the i-th historical lithium battery, represents the capacity of the i-th historical lithium battery at the t-th moment, Indicates the number of historical lithium battery sampling moments, m i Indicates the number of capacity data collected from the i-th historical lithium battery;

[0009] (2) Construct a Wiener process model of the target lithium battery at each sampling time;

[0010] X(t)=X(0)+(η(S,T;Θ)+ε η )f(t)+σ B B(t)

[0011] Where X(0) represents the initial capacity of the target lithium battery, f(t) represents the degradation trend function fitted by the long short-term memory neural network, B(t) is the standard Brownian motion, σ B is the diffusion coefficient of the Wiener process, represents the variance of the drift coefficient of the Wiener process, η(S,T; Θ) is the generalized Eyring model, which is used to replace the mean of the drift coefficient of the Wiener process. Its mathematical expression is as follows:

[0012]

[0013] Where T represents the temperature of the target lithium battery, S = [s1, s2, ..., s k ,…,s K ] represents the set of environmental stresses of target lithium battery type K, s k represents the kth environmental stress of the target lithium battery, θ=[θ1,θ2,θ k,1 ,θ k,2 ] represents the parameter set of the generalized Eyring model, θ1, θ2, θ k,1 ,θ k,2 They represent the parameters to be determined respectively;

[0014] (3) Use historical lithium batteries to estimate the model prior parameters of the target lithium battery;

[0015] (3.1) Obtain the capacity data of the target lithium battery at the current moment;

[0016] According to the method in step (1), get the current time t n The capacity data of the target lithium battery And the working environment stress S=[s1,s2,…,s k ,…,s K ] and temperature T, judge Whether the failure threshold w is exceeded. If so, go to step (5) and the algorithm ends. If not, go to step (3.2).

[0017] (3.2), remove noise from capacity data;

[0018] The Savitzky-Golay filter is used to smooth the degradation data of M groups of historical lithium batteries and the capacity data of the target lithium battery to obtain the smoothed data of M groups of historical lithium batteries. And the smoothed capacity data of the target lithium battery in, represents the smoothed value of the degradation data of the i-th historical lithium battery at the t-th moment;

[0019] (3.3) Calculate the generalized Eying model parameter Θ using historical lithium battery samples;

[0020] (3.3.1) According to the failure threshold defined in step (3.1), calculate the life of M groups of historical lithium battery samples, denoted as Life = [Life1, Life2, …, Lifei ,…,Life M ], Life i represents the life of the i-th historical lithium battery;

[0021] (3.3.2) Select S from the M group of historical lithium battery samples lowest =min(S i ) and T lowest =min(T i ) of the historical lithium battery samples, and the mean value of its drift coefficient η lowest Defined as 1, it is expressed as:

[0022] η lowest (S lowest ,T lowest ,Θ)=1

[0023] The following formula is used to calculate the mean of the drift coefficients of the remaining M-1 historical lithium battery samples:

[0024]

[0025] Where p = 1, 2, ..., M-1, Life lowest Represents the life of the historical lithium battery sample with the lowest environmental stress level, Life p represents the life of the pth historical lithium battery sample;

[0026] (3.3.3) Use the following formula to normalize different types of environmental stress and temperature:

[0027]

[0028] Among them, s min,k and T min They represent the minimum value of the kth environmental stress and the minimum value of the temperature in the M group of historical lithium battery samples, and represent the normalized k-th environmental stress and temperature respectively;

[0029] (3.3.4) The normalized environmental stress of the M group of historical lithium battery samples temperature and degradation rate η i (S i ,T i ,Θ) into the Eyring model calculation formula in step (2) to estimate the model parameters

[0030] (3.4) Calculate the prior estimate of the mean drift coefficient of the target lithium battery

[0031] Use the normalization method in step (3.3) to normalize the environmental stress and temperature of the target lithium battery to obtain the normalized result and

[0032] Based on the estimated values ​​of the Eyring model parameters obtained in step (3.3), the normalized results and Substitute the Eyring model constructed in step (2) to calculate the prior estimate of the mean drift coefficient of the target lithium battery

[0033] (3.5) Calculate the prior estimate of the variance of the target lithium battery drift coefficient

[0034] Normalize the smoothed capacity data of M groups of historical lithium battery samples:

[0035]

[0036] in,

[0037] The normalized capacity data of the M group of historical lithium battery samples is Then, the normalized life of the historical lithium battery sample is calculated using the method in step (3.3.3) And define the mean value of its drift coefficient The inverse of lifespan:

[0038]

[0039] Finally, calculate the prior estimate of the variance of the target lithium battery drift coefficient

[0040]

[0041] (3.6) Calculate the estimated value of the diffusion coefficient of the target lithium battery

[0042]

[0043] in, represents the capacity data of M groups of historical lithium battery samples after smoothing, represents the first-order difference of the i-th group of historical lithium battery capacity data, It means that the empirical mode decomposition algorithm is used from The first-order difference of the capacity data extracted from express The covariance matrix of Indicates (m i -1)-order identity matrix, the superscript T indicates the transpose of the matrix;

[0044] Based on the above formula, the 'fmincon' function in Matlab is used to solve and obtain the estimated value of the diffusion coefficient

[0045]

[0046] (3.7) Construct the degradation trend function f(t) based on long short-term memory neural network;

[0047] Use the normalized smoothed data of historical lithium batteries of group M obtained in step (3.5) Train a long short-term memory neural network LSTM with q input nodes and 1 output node all , where t n >q>0;

[0048] Normalized smoothed data of the target lithium battery The last q data As a neural network LSTM all The input is used to get the predicted output

[0049] Then the predicted number According to the neural network LSTM all Input, get the predicted data Repeat the above steps until the predicted data is obtained. When , the prediction stops;

[0050] Finally, the data As the degradation trend function f(t) at time t1~t end The value of

[0051] (4) Real-time prediction of the remaining life of the target lithium battery;

[0052] (4.1) Update the model parameters of the target lithium battery;

[0053] The Kalman filter algorithm is used to update the mean and variance of the drift coefficient of the target lithium battery. Assuming that at t v-1 The mean and variance of the moment drift coefficient are: and Then t v The mean and variance of the moment and for:

[0054]

[0055] Among them, Δf(t v )=f(t v )-f(t v-1 ) represents the function f(t) at time t v The first-order difference at Represents the smoothed target lithium battery capacity aging data at time t v The first-order difference at ;

[0056] (4.2), calculate the current time t n The remaining life probability density function of the target lithium battery;

[0057] Based on the model parameters updated in step (4.1) and It can be calculated that at t n At the moment, the remaining life probability density function g of the target lithium battery RUL (l n ):

[0058]

[0059] in, h(l n )=f(l n +t n )-f(t n ), h(l n )'=h(l n )-h(l n-1 ), l n represents the probability density function g RUL (l n ) represents the value of the remaining life;

[0060] (4.3), select g RUL (l n ) is the time l corresponding to the maximum value n,max As the current time t n The remaining life prediction value of the target lithium battery;

[0061] (4.4), let t n =t n+1 , then return to step (4.1);

[0062] (5) When the degradation data of the target lithium battery reaches the failure threshold w, the prediction is terminated and the algorithm ends.

[0063] The object of the invention of the present invention is achieved in this way:

[0064] The present invention discloses a method for predicting the remaining life of a lithium battery that integrates physical failure mechanism and uncertainty characterization. The method accelerates the life experiment of historical lithium battery samples to obtain capacity data of the historical lithium battery samples under different environmental stress conditions. The method then calculates the prior parameters of a generalized Eyring model of a target lithium battery by extracting the life characteristics of the historical lithium battery samples. The method then adaptively fits the degradation trend of the lithium battery through a long short-term memory neural network, and uses a Kalman filter algorithm to update the mean and variance of the drift coefficient of the model. The updated parameters are used to calculate the remaining life probability density function of the target lithium battery sample at the current moment. The method has the characteristics of high prediction accuracy, good real-time performance, strong adaptability, and the like. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of a method for predicting the remaining life of a lithium battery that integrates physical failure mechanism and uncertainty characterization of the present invention;

[0066] Figure 2 These are 6 sets of lithium battery capacity degradation data under different working conditions obtained through accelerated life experiments;

[0067] Figure 3 It is the capacity degradation data of historical lithium battery samples after removing environmental stress;

[0068] Figure 4 This is a diagram of the remaining life prediction result of a lithium battery remaining life prediction method that integrates physical failure mechanism and uncertainty characterization proposed by the present invention:

[0069] Figure 5 The figures are the results of three prediction models for predicting the remaining life of lithium batteries; among them, (a) is a remaining life prediction model based on LSTM neural network; (b) is a remaining life prediction model based on traditional Wiener process; (c) is the model of the present invention for predicting the remaining life of lithium batteries. DETAILED DESCRIPTION

[0070] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0071] Example

[0072] Figure 1 It is a flow chart of a method for predicting the remaining life of a lithium battery that integrates physical failure mechanism and uncertainty characterization according to the present invention.

[0073] In this embodiment, if Figure 1As shown, the present invention provides a method for predicting the remaining life of a lithium battery integrating physical mechanism and uncertainty characterization, comprising the following steps:

[0074] S1. Obtain the capacity value of the target lithium battery at different times;

[0075] By accelerating the life experiment of M historical lithium batteries, the capacity data of each historical lithium battery at different times and its corresponding K-type working environment stress and temperature are sampled;

[0076] Among them, the capacity data of the i-th historical lithium battery at different times is recorded as The working environment stress is recorded as S i =[s i,1 ,s i,2 ,…,s i,k ,…,s i,K ] and the temperature is denoted as T i , where i = 1, 2, ... M, k = 1, 2, ..., K, s i,k represents the kth type of environmental stress of the i-th historical lithium battery, represents the capacity of the i-th historical lithium battery at the t-th moment, Indicates the number of historical lithium battery sampling moments, m i Indicates the number of capacity data collected from the i-th historical lithium battery;

[0077] S2, constructing a Wiener process model of the target lithium battery at each sampling moment;

[0078] X(t)=X(0)+(η(S,T;Θ)+ε η )f(t)+σ B B(t)

[0079] Where X(0) represents the initial capacity of the target lithium battery, f(t) represents the degradation trend function fitted by the long short-term memory neural network, B(t) is the standard Brownian motion, σ B is the diffusion coefficient of the Wiener process, represents the variance of the drift coefficient of the Wiener process, η(S,T; Θ) is the generalized Eyring model, which is used to replace the mean of the drift coefficient of the Wiener process. Its mathematical expression is as follows:

[0080]

[0081] Where T represents the temperature of the target lithium battery, S = [s1, s2, ..., s k ,…,s K ] represents the set of environmental stresses of target lithium battery type K, s k represents the kth environmental stress of the target lithium battery, θ=[θ1,θ2,θk,1 ,θ k,2 ] represents the parameter set of the generalized Eyring model, θ1, θ2, θ k,1 ,θ k,2 They represent the parameters to be determined respectively;

[0082] S3, using historical lithium batteries to estimate model prior parameters of target lithium batteries;

[0083] S3.1. Obtain the capacity data of the target lithium battery at the current moment;

[0084] Use the method in step S1 to obtain the current time t n The capacity data of the target lithium battery And the working environment stress S=[s1,s2,…,s k ,…,s K ] and temperature T, judge Whether the failure threshold w is exceeded, if exceeded, go to step S5 and the algorithm ends; if not exceeded, go to step S3.2;

[0085] S3.2, remove noise from the original capacity data;

[0086] The Savitzky-Golay filter is used to smooth the degradation data of M groups of historical lithium batteries and the capacity data of the target lithium battery to obtain the smoothed data of M groups of historical lithium batteries. And the smoothed capacity data of the target lithium battery in, represents the smoothed value of the degradation data of the i-th historical lithium battery at the t-th moment;

[0087] S3.3. Calculate the generalized Eying model parameter Θ using historical lithium battery samples;

[0088] S3.3.1. According to the failure threshold defined in step S3.1, calculate the life of the M group of historical lithium battery samples, denoted as Life = [Life1, Life2, …, Life i ,…,Life M ], Life i represents the life of the i-th historical lithium battery;

[0089] S3.3.2. Select S from the historical lithium battery samples in Group M. lowest =min(S i ) and T lowest =min(T i ) of the historical lithium battery sample, and then the one with the lowest environmental stress (S lowest ,T lowest ) is defined as 1, which is expressed as:

[0090] η lowest (S lowest ,T lowest ,Θ)=1

[0091] The following formula is used to calculate the mean of the drift coefficients of the remaining M-1 historical lithium battery samples:

[0092]

[0093] Where p = 1, 2, ..., M-1, Life lowest Represents the life of the historical lithium battery sample with the lowest environmental stress level, Life p represents the life of the pth historical lithium battery sample;

[0094] S3.3.3. Normalize different types of environmental stress and temperature using the following formula:

[0095]

[0096] Among them, s min,k and T min They represent the minimum value of the kth environmental stress and the minimum value of the temperature in the M group of historical lithium battery samples, and represent the normalized k-th environmental stress and temperature respectively;

[0097] S3.3.4. The normalized environmental stress of the historical lithium battery samples in group M temperature and degradation rate η i (S i ,T i ,Θ) into the Eyring model calculation formula in step S2 to estimate the model parameters

[0098] S3.4. Calculate the prior estimate of the mean drift coefficient of the target lithium battery

[0099] The normalization method in step S3.3 is used to normalize the environmental stress and temperature of the target lithium battery to obtain the normalized result and

[0100] Based on the estimated values ​​of the Eyring model parameters obtained in step S3.3, the normalized results and Substitute the Eyring model constructed in step S2 to calculate the prior estimate of the mean drift coefficient of the target lithium battery

[0101] S3.5. Calculate the prior estimate of the variance of the target lithium battery drift coefficient

[0102] Normalize the smoothed capacity data of M groups of historical lithium battery samples:

[0103]

[0104] in,

[0105] The normalized capacity data of the M group of historical lithium battery samples is Then, the normalized lifetime of the historical lithium battery sample is calculated using the method in step S3.3.3 And define the mean value of its drift coefficient The inverse of lifespan:

[0106]

[0107] Finally, calculate the prior estimate of the variance of the target lithium battery drift coefficient

[0108]

[0109] S3.6. Calculate the estimated value of the diffusion coefficient of the target lithium battery

[0110]

[0111] in, represents the capacity data of M groups of historical lithium battery samples after smoothing, represents the first-order difference of the i-th group of historical lithium battery capacity data, It means that the empirical mode decomposition algorithm is used from The first-order difference of the capacity data extracted from express The covariance matrix of Indicates (m i -1)-order identity matrix, the superscript T indicates the transpose of the matrix;

[0112] Based on the above formula, the 'fmincon' function in Matlab is used to solve and obtain the estimated value of the diffusion coefficient

[0113]

[0114] S3.7, construct the degradation trend function f(t) based on long short-term memory neural network;

[0115] Use the normalized smoothed data of historical lithium batteries of group M obtained in step S3.5 Train a long short-term memory neural network LSTM with q input nodes and 1 output node all , where t n >q>0;

[0116] Normalized smoothed data of the target lithium battery The last q data As a neural network LSTM all The input is used to get the predicted output

[0117] Then the predicted number According to the neural network LSTM all Input, get the predicted data Repeat the above steps until the predicted data is obtained. When , the prediction stops;

[0118] Finally, the data As the degradation trend function f(t) at time t1~t end The value of

[0119] S4. Real-time prediction of the remaining life of the target lithium battery;

[0120] S4.1. Update the model parameters of the target lithium battery;

[0121] The Kalman filter algorithm is used to update the mean and variance of the drift coefficient of the target lithium battery. Assuming that at t v-1 The mean and variance of the moment drift coefficient are: and Then t v The mean and variance of the moment and for:

[0122]

[0123] Among them, Δf(t v )=f(t v )-f(t v-1 ) represents the function f(t) at time t v The first-order difference at Represents the smoothed target lithium battery capacity aging data at time t v The first-order difference at ;

[0124] S4.2. Calculate the current time t n The remaining life probability density function of the target lithium battery;

[0125] Based on the model parameters updated in step S4.1 and It can be calculated that at t n At the moment, the remaining life probability density function g of the target lithium battery RUL (l n ):

[0126]

[0127] in, h(l n )=f(l n +t n )-f(t n ), h(l n )'=h(l n )-h(l n-1 ), l n represents the probability density function g RUL (l n ) represents the value of the remaining life;

[0128] S4.3. Select g RUL (l n ) is the time l corresponding to the maximum value n,max As the current time t n The remaining life prediction value of the target lithium battery;

[0129] S4.4, let t n =t n+1 , then return to step S4.1;

[0130] S5. The degradation data of the target lithium battery reaches the failure threshold w, the prediction is terminated, and the algorithm ends.

[0131] Simulation Verification

[0132] In order to illustrate the technical effect of the present invention, this embodiment adopts Figure 2 The six groups of lithium battery aging data with different charging current conditions provided by the University of Maryland are shown as experimental data sets. Lithium battery CS2-34 is selected as the implementation object of this patent, and the remaining five groups of lithium battery degradation data are selected as historical degradation data to simulate the remaining life prediction of lithium batteries under real-time working conditions.

[0133] In step S3, the prior parameters of the target lithium battery degradation model are estimated using the historical lithium battery. The model parameters of the CS2-34 lithium battery estimated using the remaining 5 groups of lithium battery degradation data are shown in the following table:

[0134] Table 1 Model parameters of historical lithium batteries

[0135]

[0136] The method in step S3.5 is used to remove the accelerating effect of environmental stress on the aging process of the historical lithium battery samples. The capacity degradation data of the historical lithium battery aging samples without the influence of environmental stress is obtained as follows: Figure 3 As shown. Figure 3 The data in is used to train an LSTM neural network to characterize the degradation trend function f(t) of the target lithium battery in step S2.

[0137] Then, the Kalman filter algorithm in step S4.1 is used to update the model posterior parameters of the target lithium battery, so that the model parameters can adaptively fit the degradation characteristics of the target battery as the field data of the target lithium battery increases. Based on the updated model posterior parameters, the remaining life probability density function in step S4.2 is used to obtain the probability density curve of the target lithium battery at the current moment. The remaining life curve of the lithium battery CS2-34 predicted by the present invention at different times is as follows: Figure 4 It can be clearly seen that the remaining life prediction curve obtained by the present invention can provide a probability density function curve of the remaining life at different times, which can provide more information for the maintenance and guarantee work of the lithium battery, thereby facilitating the maintenance and guarantee work of the relevant energy supply system.

[0138] In order to quantitatively compare and measure the prediction performance, Figure 5 The remaining life prediction results of lithium battery CS2-34 using the present invention, traditional nonlinear Wiener process and long short-term memory neural network model are shown. Figure 5 From the prediction results of the remaining life of lithium batteries by different types of models, it can be found that because the present invention takes into account the effects of various physical quantities on the basis of the Wiener process and adopts the long short-term memory neural network model to adaptively fit the degradation trend of the target lithium battery, it is better than the traditional Wiener process model that does not consider the environmental stress effect; at the same time, because the present invention includes the Wiener process, compared with directly using the long short-term memory neural network model for prediction, the present invention can provide the uncertainty of the prediction results at each moment. In addition, the long short-term memory neural network model is heavily dependent on a large amount of training data, but the present invention can obtain more accurate prediction results under the condition of less training data. Table 2 shows the average prediction error of each model for the remaining life of lithium battery CS2-34.

[0139] Table 2 Average absolute error and average relative error of B5 remaining life prediction by different models

[0140] Long Short-Term Memory Neural Networks Traditional Wiener Process The present invention Mean absolute error 28.0cycle 10.5cycle 8.8cycle Mean relative error 28.8％ 13.0％ 9.0％

[0141] It can be seen from the prediction results shown in Table 2 that the accuracy of the remaining life prediction results of this model is much higher than that of other models, which directly illustrates the advantages of the lithium battery remaining life prediction method proposed in the present invention that integrates physical failure mechanism and uncertainty characterization.

[0142] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A method for predicting the remaining life of a lithium battery by integrating physical mechanism and uncertainty characterization, characterized in that: The following steps are involved: (1) Obtain the capacity value of the target lithium battery at different times; By accelerating the life experiment of M historical lithium batteries, the capacity data of each historical lithium battery at different times and its corresponding K-type working environment stress and temperature are sampled; Among them, the capacity data of the i-th historical lithium battery at different times is recorded as The working environment stress is recorded as S i =[s i,1 ,s i,2 ,…,s i,k ,…,s i,K ] and the temperature is denoted as T i , where i = 1, 2, ... M, k = 1, 2, ..., K, s i,k represents the kth environmental stress of the i-th historical lithium battery, represents the capacity of the i-th historical lithium battery at the t-th moment, Indicates the number of historical lithium battery sampling moments, m i Indicates the number of capacity data collected from the i-th historical lithium battery; (2) Construct a Wiener process model of the target lithium battery at each sampling time; X(t)=X(0)+(η(S,T;Θ)+ε η )f(t)+σ B B(t) Where X(0) represents the initial capacity of the target lithium battery, f(t) represents the degradation trend function fitted by the long short-term memory neural network, B(t) is the standard Brownian motion, σ B is the diffusion coefficient of the Wiener process, represents the variance of the drift coefficient of the Wiener process, η(S,T; Θ) is the generalized Eyring model, which is used to replace the mean of the drift coefficient of the Wiener process. Its mathematical expression is as follows: Where T represents the temperature of the target lithium battery, S = [s1, s2, ..., s k ,…,s K ] represents the set of target lithium battery K environmental stress, s k represents the kth environmental stress of the target lithium battery, θ=[θ1,θ2,θ k,1 ,θ k,2 ] represents the parameter set of the generalized Eyring model, θ1, θ2, θ k,1 ,θ k,2 They represent the parameters to be determined respectively; (3) Use historical lithium batteries to estimate the model prior parameters of the target lithium battery; (3.1) Obtain the capacity data of the target lithium battery at the current moment; Obtain the current time t according to the method in step (1) n The capacity data of the target lithium battery And the working environment stress S=[s1,s2,…,s k ,…,s K ] and temperature T, judge Whether the failure threshold w is exceeded. If so, go to step (5) and the algorithm ends. If not, go to step (3.2). (3.2), remove noise from capacity data; The Savitzky-Golay filter is used to smooth the degradation data of M groups of historical lithium batteries and the capacity data of the target lithium battery to obtain the smoothed data of M groups of historical lithium batteries. And the smoothed capacity data of the target lithium battery in, represents the smoothed value of the degradation data of the i-th historical lithium battery at the t-th moment; (3.3) Calculate the generalized Eying model parameter Θ using historical lithium battery samples; (3.3.1) According to the failure threshold defined in step (3.1), calculate the life of M groups of historical lithium battery samples, denoted as Life = [Life1, Life2, …, Life i ,…,Life M ], Life i represents the life of the i-th historical lithium battery; (3.3.2) Select S from the M group of historical lithium battery samples lowest =min(S i ) and T lowest =min(T i ) of the historical lithium battery samples, and the mean value of its drift coefficient η lowest Defined as 1, it is expressed as: or lowest (S lowest ,T lowest ,Θ)=1 The following formula is used to calculate the mean of the drift coefficients of the remaining M-1 historical lithium battery samples: Where p = 1, 2, ..., M-1, Life lowest Represents the life of the historical lithium battery sample with the lowest environmental stress level, Life p represents the life of the pth historical lithium battery sample; (3.3.3) Use the following formula to normalize different types of environmental stress and temperature: Among them, s min,k and T min They represent the minimum value of the kth environmental stress and the minimum value of the temperature in the M group of historical lithium battery samples, and represent the normalized k-th environmental stress and temperature respectively; (3.3.4) The normalized environmental stress of the M group of historical lithium battery samples temperature and degradation rate η i (S i ,T i ,Θ) into the Eyring model calculation formula in step (2) to estimate the model parameters (3.4) Calculate the prior estimate of the mean drift coefficient of the target lithium battery Use the normalization method in step (3.3) to normalize the environmental stress and temperature of the target lithium battery to obtain the normalized result and Based on the estimated values ​​of the Eyring model parameters obtained in step (3.3), the normalized results and Substitute the Eyring model constructed in step (2) to calculate the prior estimate of the mean drift coefficient of the target lithium battery (3.5) Calculate the prior estimate of the variance of the target lithium battery drift coefficient Normalize the smoothed capacity data of M groups of historical lithium battery samples: in, The normalized capacity data of the M group of historical lithium battery samples is Then, the normalized life of the historical lithium battery sample is calculated using the method in step (3.3.3) And define the mean value of its drift coefficient The inverse of lifespan: Finally, calculate the prior estimate of the variance of the target lithium battery drift coefficient (3.6) Calculate the estimated value of the diffusion coefficient of the target lithium battery in, represents the capacity data of M groups of historical lithium battery samples after smoothing, represents the first-order difference of the i-th group of historical lithium battery capacity data, It means that the empirical mode decomposition algorithm is used from The first-order difference of the capacity data extracted from express The covariance matrix of Indicates (m i -1)-order identity matrix, the superscript T indicates the transpose of the matrix; Solving the above equation, we can get an estimate of the diffusion coefficient: (3.7) Construct the degradation trend function f(t) based on long short-term memory neural network; Use the normalized smoothed data of historical lithium batteries of group M obtained in step (3.5) Train a long short-term memory neural network LSTM with q input nodes and 1 output node all , where t n >q>0; Normalized smoothed data of the target lithium battery The last q data As a neural network LSTM all The input is used to get the predicted output Then the predicted number According to the neural network LSTM all Input, get the predicted data Repeat the above steps until the predicted data is obtained. When , the prediction stops; Finally, the data As the degradation trend function f(t) at time t1~t end The value of (4) Real-time prediction of the remaining life of the target lithium battery; (4.1) Update the model parameters of the target lithium battery; The Kalman filter algorithm is used to update the mean and variance of the drift coefficient of the target lithium battery. Assuming that at t v-1 The mean and variance of the moment drift coefficient are: and Then t v The mean and variance of the moment and for: Among them, Δf(t v )=f(t v )-f(t v-1 ) represents the function f(t) at time t v The first-order difference at Represents the smoothed target lithium battery capacity aging data at time t v The first-order difference at ; (4.2), calculate the current time t n The remaining life probability density function of the target lithium battery; Based on the model parameters updated in step (4.1) and It can be calculated that at t n At the moment, the remaining life probability density function g of the target lithium battery RUL (l n ): in, h(l n )=f(l n +t n )-f(t n ), h(l n )'=h(l n )-h(l n-1 );l n represents the probability density function g RUL (l n ) is an independent variable, representing the value of the remaining life; (4.3), select g RUL (l n ) is the time l corresponding to the maximum value n,max As the current time t n The remaining life prediction value of the target lithium battery; (4.4), let t n =t n+1 , then return to step (4.1); (5) When the degradation data of the target lithium battery reaches the failure threshold w, the prediction is terminated and the algorithm ends.

Images