Battery state of health prediction method based on two-stage neural wiener process

Abstract

The application discloses a battery state of health prediction method based on a two-stage neural Wiener process, comprising the following steps: step one: change point identification: 11) learning the time sequence characteristics in the battery degradation process through a Bi-GRU model, using the forward and backward propagation hidden layer state splicing to predict the state of health in the battery degradation process; 12) introducing a learnable adaptive threshold to judge whether a change point occurs; step two: two-stage Wiener process parameter estimation: 21) taking the change point time τ as a demarcation point to establish a two-stage Wiener degradation model including a drift coefficient and a diffusion coefficient; 22) adopting an EM algorithm to estimate the drift coefficient and the diffusion coefficient; 23) updating the drift coefficient; step three: deriving the life distribution function of the two-stage Wiener process; step four: battery state of health prediction: 41) deriving the state transition probability density function of the neural Wiener process; 42) taking the mathematical expectation of the state transition probability density function as the battery state of health of the current cycle.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery management and monitoring, and in particular to a battery state of health prediction method based on a two-stage neural Wiener process. BACKGROUND

[0002] Lithium-ion batteries have been widely used in important fields of the national economy. However, as the performance of the batteries gradually deteriorates, safety accidents occur from time to time, which has attracted people's high attention to battery management and monitoring. Therefore, accurately estimating the state of health (SOH) of the battery and its future degradation trajectory is of great significance to ensure the safe operation of the battery, which not only helps to improve the stability of the battery operation, but also effectively reduces the potential safety risks. In this context, the Wiener process has become an effective tool for quantifying the non-monotonic degradation process of battery performance due to its excellent mathematical properties. This process can capture and describe the complexity of the change of battery performance over time in the form of probability.

[0003] Most existing researches use single-stage degradation models with a single degradation rate to predict the SOH of lithium-ion batteries. Such models can reasonably explain the actual degradation process of the battery in a specific simplified scenario, including methods based on the Wiener process (WP), methods based on the Gamma process, and methods based on the inverse Gaussian process. Among these methods, the WP model has become an important tool for battery random degradation modeling due to its efficiency and applicability.

[0004] However, the actual degradation process of lithium-ion batteries often exhibits significant nonlinear characteristics, and is influenced by age and usage state. This means that the degradation trajectories of different batteries under similar conditions may have individual differences, and their change rules may not be consistent. The capacity of the battery decreases slowly at first, but the speed of decrease accelerates after the change point, which means that the degradation mechanism of the battery changes before and after the change point. At this time, the single-stage degradation model cannot describe the complex two-stage degradation process due to its assumption of a single degradation rate. Therefore, it is necessary to study degradation models that reflect the degradation characteristics of different stages to achieve SOH prediction of batteries with different degradation rates. In the actual degradation process of the battery, the position of the change point varies due to individual differences, and the distribution of the change point has great flexibility in time. At the same time, due to individual differences, batteries of the same batch have different operating conditions and health states, and the degradation processes of different batteries are different, which brings challenges to prediction and management. SUMMARY

[0005] Therefore, the battery state of health prediction method based on two-stage neural Wiener process has full consideration on individual difference of change point position, distribution form is not limited initially, and can effectively predict the battery SOH.

[0006] To achieve the above object, the present application provides the following technical scheme.

[0007] A battery state of health prediction method based on two-stage neural Wiener process comprises the following steps.

[0008] Step one: change point identification

[0009] 11) The Bi-GRU model is used to learn the time sequence characteristics in the battery degradation process, the hidden layer state of forward and backward propagation is spliced to predict the state of health in the battery degradation process;

[0010] 12) A learnable adaptive threshold is introduced, if the residual error between the predicted value and the observed value of the battery state of health exceeds the set adaptive threshold, the change point occurs;

[0011] Step two: two-stage Wiener process parameter estimation

[0012] 21) The change point time is taken as a dividing point to establish a two-stage Wiener degradation model including drift coefficient and diffusion coefficient;

[0013] 22) The EM algorithm is used to estimate the drift coefficient and diffusion coefficient in the two-stage Wiener degradation model;

[0014] 23) The adaptive gated double attention unit is used to dynamically update the drift coefficient;

[0015] Step three: derivation of life distribution function of two-stage Wiener process

[0016] The nonlinear degradation process is converted into standard Brown motion, the distribution of remaining life is derived by using the properties of standard Brown motion, and the life distribution function is derived according to the independent increment property of Wiener process;

[0017] Step four: battery state of health prediction

[0018] 41) The state transition probability density function of neural Wiener process is derived according to the Ito principle;

[0019] 42) The mathematical expectation of the state transition probability density function is taken as the battery state of health of the current cycle.

[0020] Further, in the step 11), the hidden layer state of forward GRU output is:

[0021]

[0022] The hidden layer state output by the inverse GRU is:

[0023]

[0024] By concatenating the hidden state from the forward and backward propagation, we obtain:

[0025]

[0026] in: This represents the hidden layer state obtained by splicing. The hidden layer state is the output of the positive GRU; The hidden layer state output by the inverse GRU; This is the input data from the previous time step; This is the embedding vector.

[0027] Furthermore, in step 12), a comprehensive loss function is constructed to optimize the adaptive threshold. The comprehensive loss function is:

[0028]

[0029] in: This is the overall loss function; The loss function for identifying change points; Optimize the loss function for the threshold; The equilibrium coefficient is:

[0030]

[0031] in: This is a real label; The probability of a change point predicted by the Bi-GRU model; The current moment;

[0032]

[0033] in: An adaptive threshold; This is an indicator function.

[0034] Furthermore, in step 21), the two-stage Wiener degradation model is expressed as:

[0035]

[0036] in: Indicates that lithium-ion batteries are The amount of degradation at any given time; For lithium-ion batteries The amount of degradation at time t; t is the change point time; is the degradation amount of lithium-ion battery at the change point; and are the drift coefficient functions at the two degradation stages, respectively; is the first stage time; is the first order drift function parameter; is the second stage time; is the second order drift function parameter; and are the diffusion coefficients before and after the change point, respectively; B(t) is the standard Brownian motion.

[0037] Further, in the step 22), the method for estimating the drift coefficient and the diffusion coefficient in the two-stage Wiener degradation model by using the EM algorithm is as follows: in the E step, the posterior distribution of the hidden variable is calculated by Kalman filtering, and the optimal degradation state estimation is provided; in the M step, the drift coefficient and the diffusion coefficient are estimated by maximizing the log-likelihood function based on the degradation state estimation.

[0038] Through the iteration of the EM algorithm, the estimated value of the drift coefficient is obtained and the estimated value of the diffusion coefficient is obtained :

[0039]

[0040]

[0041] wherein: is the degradation state estimation calculated in the E step; is the current time; is the variance of the observation noise.

[0042] Further, in the step 23), the adaptive gated double attention unit comprises a reset gate, an update gate, an attention gate 1 and an attention gate 2.

[0043] The reset gate is used to combine the state at the previous time and the current drift coefficient input , and is expressed as:

[0044]

[0045] wherein: is a sigmoid activation function; is a weight matrix of the reset gate; is a bias matrix;

[0046] The update gate is used to control the fusion proportion of the state information at the previous time and the state information at the current time, and is expressed as:

[0047]

[0048] wherein: is the weight matrix of the update gate, is the bias matrix of the update gate;

[0049] The attention gate 1 is used to control the attention degree of the input at the current time and the hidden state information at the previous time, denoted as:

[0050]

[0051] wherein: denotes the attention gate 1; is the weight matrix of the attention gate 1; is the weight converted by the softmax function score; denotes the dot product operation;

[0052] The attention gate 2 is used to focus on the output state of the reset gate and the update gate, denoted as:

[0053]

[0054] wherein: denotes the attention gate 2; and is the weight matrix learned by the attention gate 2; is the variable point position;

[0055] The update formula of the hidden state of the adaptive gated double attention unit is:

[0056]

[0057]

[0058] wherein: and are weight matrices; is a bias matrix; is obtained by function, denoting the candidate hidden state at time t; is the current time state.

[0059] Further, the output of the adaptive gated double attention unit is denoted as:

[0060]

[0061]

[0062]

[0063]

[0064] wherein: is the drift coefficient function updated by adaptive gated double attention unit; is the original hidden state with Dropout regularization; is the deep feature after ReLU+Dropout; is the drift coefficient related feature after full connection; , and are the weight matrices of the output layer; , and are the bias vectors; is the neural network model overfitting regularization technique, is the activation function.

[0065] Further, in the first stage of the nonlinear degradation process in step three:

[0066]

[0067] in the second stage of the nonlinear degradation process:

[0068]

[0069] After the first stage, the degradation amount is transferred from to in time, and the transition probability is expressed as:

[0070]

[0071] wherein: represents the posterior probability;

[0072] Based on the definition of FHT, the two-stage life distribution function is obtained:

[0073]

[0074] wherein: is the failure threshold; is the transition probability.

[0075] Further, in step 41), the neural Wiener process is expressed as:

[0076]

[0077] wherein: represents the drift coefficient function; represents the diffusion coefficient function; standard Brown motion;

[0078] The state transition probability density function is:

[0079]

[0080] wherein: is represented as historical observation data.

[0081] Further, in the step 42), the battery health state of the current cycle period is:

[0082]

[0083] wherein: is the mathematical expectation of the state transition probability density function.

[0084] The beneficial effects of the present application are:

[0085] The present application utilizes lithium ion battery charging and discharging history monitoring degradation data, and proposes a battery health state prediction method based on a two-stage neural Wiener process, aiming to realize more accurate lithium ion battery degradation trajectory prediction. The method of the present application estimates the drift coefficient and diffusion coefficient in the battery degradation process through the expectation maximization algorithm (EM algorithm), effectively overcoming the limitations of the traditional maximum likelihood estimation method in the battery degradation model, and can better handle the randomness and unpredictability in the battery degradation process. The present application method fully considers the influence of randomness and change points in the degradation process on the degradation trajectory, studies the update model of the drift coefficient based on the adaptive gated double attention unit (AGDAU) neural network, and the AGDAU neural network model successfully copes with the short-term fluctuations and change points in the battery degradation process by adaptively adjusting the attention weight, so that the model can accurately capture the complex nonlinear relationship between the degradation state and the degradation rate.

[0086] Therefore, the method of the present application has unique advantages, not only solves the problem that the single-stage degradation model cannot describe the complex two-stage degradation process due to its assumed single degradation rate, but also accurately reflects the degradation model of different stage degradation characteristics, realizes accurate prediction of battery SOH with different degradation rates, and has high popularization value and application value. BRIEF DESCRIPTION OF DRAWINGS

[0087] In order to make the purpose, technical scheme and beneficial effects of the present application clearer, the present application provides the following drawings for illustration:

[0088] Figure 1 is the flow chart of the battery health state prediction method based on the two-stage neural Wiener process of the present application;

[0089] Figure 2A schematic diagram of the adaptive gated double attention unit. DETAILED DESCRIPTION

[0090] The application will be further described below in conjunction with the drawings and specific embodiments so that those skilled in the art can better understand the application and implement it. The embodiments are not intended to limit the application.

[0091] The embodiment is based on the idea of fusing random degradation model and data-driven model, and proposes a new battery state of health prediction method based on two-stage neural Wiener process. The method fully considers the individual differences of lithium-ion batteries in the same batch at each stage. In order to solve the problem that the change point time is not subject to the initial distribution form, the randomness in the degradation process and the short-term change point fluctuation are considered, and a change point identification model is constructed by combining Bi-GRU network and adaptive threshold adjustment mechanism, which can adaptively find the change point and optimize the accuracy of change point detection, thereby providing more accurate prediction and decision support for battery health management. By designing an adaptive gated double attention mechanism, the context information at the historical, future and change point is obtained. Attention 1 controls the attention degree of the current time input and the hidden state information of the previous time, and attention 2 responds to the influence of the change point. The drift coefficient function is updated in real time according to the few failure sample data, which enhances the learning ability and accuracy of the neural Wiener process (WP) model. The model estimates the degradation coefficient based on the EM algorithm, fully considers the uncertainty of the two-stage degradation amount, and uses a neural network to describe the degradation process of each stage at the change point. A more accurate mapping relationship between the drift coefficient and the degradation state is established, which overcomes the limitations of existing neural network, WP model and single-stage neural WP model in estimating SOH. Compared with other traditional methods, the embodiment fully considers the individual differences of the change point position, the distribution form is not limited by the initial value, and the battery SOH can be more effectively predicted. The SOH prediction obtained by the embodiment is closer to the true value.

[0092] Specifically, as shown in Figure 1 The battery state of health prediction method based on two-stage neural Wiener process includes the following steps.

[0093] Step 1: Change point identification

[0094] 11) Learn the time sequence characteristics of the battery degradation process through the Bi-GRU model, and use the forward and backward propagation of the hidden layer state to splice the health state of the battery degradation process.

[0095] Traditional two-stage wigner process change point estimation method usually assumes that the change point obeys a specific probability distribution. However, considering the working conditions and individual differences of the battery, the occurrence time of the change point has strong randomness. Therefore, the embodiment learns the time sequence characteristics in the battery degradation process through the Bi-GRU model, and uses the splicing of the forward and backward propagation hidden layer states to better model the change point in the battery degradation process. The Bi-GRU network is trained using time series data, and the model learns the time dependence of battery degradation.

[0096] The hidden layer state output by the forward GRU is:

[0097]

[0098] The hidden layer state output by the reverse GRU is:

[0099]

[0100] The spliced hidden layer state of the forward and backward propagation is:

[0101]

[0102] Wherein: is the spliced hidden layer state; is the hidden layer state output by the forward GRU; is the hidden layer state output by the reverse GRU; is the input data at the previous time; is the embedding vector.

[0103] After training, the GRU can predict the battery health state at each time step.

[0104] 12) A learnable adaptive threshold is introduced. If the residual error between the predicted value and the observed value of the battery health state exceeds the set adaptive threshold, a change point occurs.

[0105] Specifically, to improve the flexibility of change point detection, a learnable adaptive threshold is introduced , and the threshold is adaptively adjusted when optimizing the loss function. By considering the identification of change points and the optimization of thresholds, the model can dynamically adjust according to different degradation modes.

[0106] In this embodiment, a comprehensive loss function is constructed to optimize the adaptive threshold, and the comprehensive loss function is:

[0107]

[0108] Wherein: is the comprehensive loss function; is the change point identification loss function; The threshold value is optimized for the loss function. The balance coefficient is optimized for the loss function.

[0109] The performance of the network in change point detection is measured by the cross-entropy loss, and the change point identification loss can be defined as the binary cross-entropy loss, which is:

[0110]

[0111] Wherein: is the true label; is the probability of the Bi-GRU model predicting a change point; is the current time.

[0112] By adjusting the threshold value , the network can better adapt to different degradation modes. The goal of the threshold value optimization loss is to minimize the error of change point detection, and the adaptive optimization is realized by introducing the threshold value, which is represented as:

[0113]

[0114] Wherein: is the adaptive threshold value; is the indicator function, when is greater than the current threshold value , it will affect the loss function.

[0115] In this embodiment, by combining the Bi-GRU network and the adaptive threshold adjustment mechanism to build a change point identification model, the change point in the degradation trajectory of the lithium ion battery can be accurately detected, thereby providing more accurate prediction and decision support for the health management of the battery.

[0116] Step two: two-stage Wiener process parameter estimation

[0117] 21) Take the change point time as the dividing point, and establish a two-stage Wiener degradation model including the drift coefficient and the diffusion coefficient.

[0118] The degradation rate of lithium ion batteries changes in two different stages, and the degradation rate of the battery is different in each stage. In order to facilitate research, we assume that the two degradation stages occur in non-overlapping time periods, and each stage follows the definition and characteristics of the Wiener continuous time random process, specifically, both stages have the property of stationary independent increment. Based on this assumption, the two-stage Wiener degradation model is constructed as:

[0119]

[0120] where: represents the degradation amount of the lithium-ion battery at time ; is the degradation amount of the lithium-ion battery at time ; 𝜏 is the change point time; is the degradation amount of the lithium-ion battery at the change point; and are the drift coefficient functions at the two degradation stages, respectively; is the first stage time; is the first-order drift function parameter; is the second stage time; is the second-order drift coefficient function; and are the diffusion coefficients before and after the change point, respectively; 𝐵(𝑡) is the standard Brownian motion.

[0121] 22) Estimate the drift coefficient and diffusion coefficient in the two-stage Wiener degradation model using the EM algorithm.

[0122] After extracting the lithium battery degradation data, the expectation-maximization algorithm is used to estimate the unknown parameters. In the case of missing data, the expectation-maximization (EM) algorithm is an effective method for parameter estimation. For the Wiener process degradation model, there are some hidden variables, such as the true degradation state of the battery at each time , which cannot be directly observed and can only be indirectly inferred through the battery's performance indicators. The core idea of the EM algorithm is to maximize the log-likelihood function of the complete data by iteratively calculating the expectation of the "latent data". Assuming that the degradation process of the lithium-ion battery consists of two stages, each stage can be considered as an independent Wiener process, and the basic steps of the EM algorithm for estimating the coefficients of the two-stage Wiener degradation model are as follows.

[0123] 1. E-step (Expectation Step): Calculate the posterior distribution of the hidden variables based on the current parameter estimates.

[0124] 2. M-step (Maximization Step): Maximize the log-likelihood function based on the expectation of the hidden variables calculated in the E-step, thereby obtaining new parameter estimates.

[0125] Suppose we have a set of lithium-ion battery observation data , the relationship between these data and the hidden variables (i.e., the state of the degradation process) can be modeled by the following equations.

[0126] State model:

[0127]

[0128] where: is the drift coefficient; is the diffusion coefficient; is the time difference between the last time and the current time; is the standard normal distribution random noise.

[0129] Observation model: observation data and the hidden state The relationship between them can be expressed by the following formula:

[0130]

[0131] Where: is the observation noise, usually assumed to be normally distributed , R is the variance of the observation noise.

[0132] Specifically, the method for estimating the drift coefficient and the diffusion coefficient in the two-stage Wiener degradation model using the EM algorithm is as follows: the E step calculates the posterior distribution of the hidden variable by Kalman filtering and provides the optimal degradation state estimate; the M step estimates the drift coefficient and the diffusion coefficient by maximizing the log-likelihood function based on the degradation state estimate.

[0133] Through the iteration of the EM algorithm, I get the final estimate of the drift coefficient and the estimate of the diffusion coefficient :

[0134]

[0135]

[0136] Where: is the degradation state estimate calculated in the E step; is the current time; is the variance of the observation noise.

[0137] 23) Use adaptive gated double attention unit to dynamically update the drift coefficient.

[0138] In order to enhance the context capture ability, so that the model can more effectively process long-distance dependence relationship in sequence data, by increasing two attention gates, in the presence of variable points, so that the model can more flexibly pay attention to the influence of historical information on the current state.

[0139] As shown in Figure 2 , in this embodiment, the adaptive gated double attention unit (AGDAU) includes a reset gate, an update gate, an attention gate 1 and an attention gate 2.

[0140] The reset gate is used to combine the state of the previous time and the current drift coefficient input is denoted as:

[0141]

[0142] wherein: is a sigmoid activation function; is a weight matrix of the reset gate; is a bias matrix.

[0143] the update gate for controlling the fusion proportion of the state information at the previous moment and the state information at the current moment, denoted as:

[0144]

[0145] wherein: is a weight matrix of the update gate, is a bias matrix of the update gate.

[0146] the attention gate 1 is used for controlling the attention degree of the input at the current moment and the hidden state information at the previous moment, denoted as:

[0147]

[0148] wherein: denotes the attention gate 1; is an attention gate 1 weight matrix; is a weight converted by a softmax function score; denotes a dot product operation.

[0149] the attention gate 2 is used for paying attention to the output state of the reset gate and the update gate, denoted as:

[0150]

[0151] wherein: denotes the attention gate 2; and are weight matrices learned by the attention gate 2; is a variable point position.

[0152] the update formula of the hidden state of the adaptive gated double attention unit is:

[0153]

[0154]

[0155] wherein: and are weight matrices; is a bias matrix; by The function is obtained, which represents the candidate hidden state at time t; The current time state.

[0156] Through the reset gate, update gate, and the outputs of attention gate 1 and attention gate 2 in the AGDAU, a weighted output for calculating the drift coefficient function is finally obtained. The output of the adaptive gated dual attention unit is represented as:

[0157]

[0158]

[0159]

[0160]

[0161] Wherein: is the drift coefficient function updated by the adaptive gated dual attention unit; is the original hidden state after Dropout regularization; is the deep feature after ReLU+Dropout; is the drift coefficient related feature after full connection; , and are the weight matrices of the output layer; , and are the bias vectors; is a neural network model overfitting regularization technique, is an activation function.

[0162] Step three: derive the life distribution function of the two-stage Wiener process

[0163] The nonlinear degradation process is converted into a standard Brownian motion, and the distribution of the remaining life is derived by using the properties of the standard Brownian motion. According to the independent increment property of the Wiener process, the life distribution function is derived.

[0164] In order to derive the residual life distribution, it is assumed that through appropriate time-space transformation, the nonlinear degradation process can be converted into a standard Brownian motion B(t). On this basis, the distribution of the residual life can be derived by using the properties of the standard Brownian motion, and the life distribution function is derived according to the independent increment property of the Wiener process.

[0165] In the first stage of the nonlinear degradation process:

[0166]

[0167] In the second stage of the nonlinear degradation process:

[0168]

[0169] After the first stage Time, the degradation amount from to Transition probability is expressed as:

[0170]

[0171] where: represents the posterior probability.

[0172] Based on the definition of FHT (Fast Hadamard Transform), the two-stage life distribution function is obtained:

[0173]

[0174] where: is the failure threshold; is the transition probability.

[0175] When the degradation amount of lithium-ion batteries before the inflection point (i.e., the moment when the degradation mechanism changes significantly) exceeds the failure threshold w, it means that the battery has failed before reaching the stage where its degradation mechanism changes, so there is no need to consider the degradation behavior after the inflection point, and the life T < τ at this time; When the lithium-ion battery fails after the inflection point, that is, the battery life T > τ, it means that the degradation process of the battery includes two stages: the degradation stage before the inflection point and the degradation stage after the inflection point. Once the unknown parameters of the two-stage degradation model are determined, the life distribution function (PDF) can be obtained.

[0176] Step four: battery state of health prediction

[0177] 41) According to the Ito principle, the state transition probability density function of the neural Wiener process is derived.

[0178] The neural Wiener process is represented as:

[0179]

[0180] where: represents the drift coefficient function; is the diffusion coefficient function; is the standard Brownian motion.

[0181] When given a historical observation data , the state transition probability density function (TPDF) about can be obtained as:

[0182]

[0183] wherein: is represented as historical observation data.

[0184] 42) the mathematical expectation of the state transition probability density function as the battery health state of the current cycle period.

[0185] Specifically, the battery health state of the current cycle period is:

[0186]

[0187] wherein: is the mathematical expectation of the state transition probability density function.

[0188] The embodiment utilizes lithium ion battery charging and discharging historical monitoring degradation data, and proposes a battery health state prediction method based on a two-stage neural Wiener process, aiming to realize more accurate lithium ion battery degradation trajectory prediction. The method estimates the drift coefficient and diffusion coefficient in the battery degradation process through the expectation maximization algorithm (EM algorithm), effectively overcoming the limitations of the traditional maximum likelihood estimation method in the battery degradation model, and can better handle the randomness and unpredictability in the battery degradation process. The embodiment method fully considers the influence of randomness and change points in the degradation process on the degradation trajectory, studies the update model of the drift coefficient based on the adaptive gated double attention unit (AGDAU) neural network, and successfully copes with the short-term fluctuations and change points in the battery degradation process by adaptively adjusting the attention weight, so that the model can accurately capture the complex nonlinear relationship between the degradation state and the degradation rate.

[0189] Therefore, the embodiment method has unique advantages, not only solves the problem that the single-stage degradation model cannot describe the complex two-stage degradation process due to its assumed single degradation rate, but also accurately reflects the degradation model of different stage degradation characteristics, realizes accurate prediction of the battery SOH with different degradation rates, and has high popularization value and application value.

[0190] The above-described embodiments are only preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Any equivalent replacement or transformation made by those skilled in the art based on the present application is within the protection scope of the present application. The protection scope of the present application is subject to the claims.

Claims

1. A method for battery state of health prediction based on two-stage neural Wiener process, characterized in that: The method comprises the following steps: Step one: change point identification 11) learning the time sequence characteristics in the battery degradation process through a Bi-GRU model, using the splicing of the forward and backward propagation hidden layer states to predict the health state in the battery degradation process; 12) introducing a learnable adaptive threshold, if the residual error between the predicted value and the observed value of the battery health state exceeds the set adaptive threshold, a change point occurs; Step two: two-stage Wiener process parameter estimation 21) with the turning point time as the demarcation point, a two-stage Wiener degradation model including drift coefficient and diffusion coefficient is established; 22) using an EM algorithm to estimate the drift coefficient and diffusion coefficient in the two-stage Wiener degradation model; 23) using an adaptive gated double attention unit to dynamically update the drift coefficient; Step three: deriving the life distribution function of the two-stage Wiener process The nonlinear degradation process is converted into a standard Brown motion, the distribution of the remaining life is derived by using the properties of the standard Brown motion, and the life distribution function is derived according to the independent increment property of the Wiener process; Step four: battery health state prediction 41) deriving the state transition probability density function of the neural Wiener process according to the It principle; 42) taking the mathematical expectation of the state transition probability density function as the battery health state in the current cycle; In the step 21), the two-stage Wiener degradation model is represented as: in: Indicates that lithium-ion batteries are The amount of degradation at any given time; For lithium-ion batteries The amount of degradation at time t; t is the change point time; This represents the degradation of a lithium-ion battery at the degradation point. and These are the drift coefficient functions for the two degradation stages, respectively; This is the first phase of the time; These are the parameters of the first-order drift function; This is the second phase of the time; This is the drift coefficient function for the second stage; and , respectively, represent the diffusion coefficients before and after the change point; 𝐵(𝑡) represents standard Brownian motion; In the step 22), the method for estimating the drift coefficient and diffusion coefficient in the two-stage Wiener degradation model by using an EM algorithm is as follows: the E step calculates the posterior distribution of the hidden variable by using Kalman filtering, and provides the optimal degradation state estimation; the M step estimates the drift coefficient and diffusion coefficient by maximizing the log-likelihood function based on the degradation state estimation; By iteration of the EM algorithm, estimates of the drift coefficients are obtained and estimates of the diffusion coefficients : wherein: is the degradation state estimate calculated in the E step; is the current time instant; is the variance of the observation noise; In the first stage of the nonlinear degradation process in the step three: In the second stage of the nonlinear degradation process: After the first stage elapsed time, the amount of degradation from to transition probability is expressed as: wherein: denotes the posterior probability; Based on the FHT definition, the two-stage life distribution function is obtained: wherein: is a failure threshold; is a transition probability.

2. The method of claim 1, wherein: In the step 11), the hidden layer state output by the forward GRU is: The hidden layer state output by the backward GRU is: The spliced hidden layer states of the forward and backward propagation are: wherein: is the concatenated hidden state; is the forward GRU output hidden state; is the backward GRU output hidden state; is the input data at the previous time step; is the embedding vector.

3. The method of claim 1, wherein: In the step 12), a comprehensive loss function is constructed to optimize the adaptive threshold, and the comprehensive loss function is: wherein: is a comprehensive loss function; is a change point identification loss function; is a threshold optimization loss function; is a balance coefficient; and: wherein: is the true label; is the probability predicted by the Bi-GRU model that the time point is a change point; is the current time; wherein: is an adaptive threshold value; is an indicator function.

4. The method of claim 1, wherein: In the step 23), the adaptive gated double attention unit comprises a reset gate, an update gate, an attention gate 1, and an attention gate 2; Reset gate for incorporating the state of the previous time with the current drift coefficient input is expressed as: wherein: is a sigmoid activation function; is a weight matrix of the reset gate; is a bias matrix; Updating gate The fusion ratio of the state information of the previous time and the state information of the current time is represented as: wherein: is an updated weight matrix for the gate, is a bias matrix for the The attention gate 1 is used to control the attention degree of the input at the current moment and the hidden state information at the previous moment, and is represented as: where: denotes attention gate 1; is the attention gate 1 weight matrix; is the weight converted by the softmax function score; denotes the dot product operation; The attention gate 2 is used to focus on the output states of the reset gate and the update gate, and is represented as: wherein: denotes the attention gate 2; and is a weight matrix learned by the attention gate 2; is a variable point position; The update formula of the hidden state of the adaptive gated double attention unit is: wherein: and is a weight matrix; is a bias matrix; is obtained by function, representing the candidate hidden state at time t; is the current time state.

5. The method of claim 4, wherein: The output of the adaptive gated double attention unit is represented as: wherein: is the drift coefficient function updated by adaptive gated double attention unit; is the original hidden state with Dropout regularization; is the deep feature with ReLU+Dropout; is the drift coefficient related feature after full connection; , and are the weight matrices of the output layer; , and are the bias vectors; is the neural network model overfitting regularization technique, is the activation function.

6. The method of claim 1, wherein: In the step 41), the neural Wiener process is represented as: wherein: represents a drift coefficient function; represents a diffusion coefficient function; represents a standard Brownian motion; The state transition probability density function is: wherein: represents historical observation data.

7. The method of claim 6, wherein: In the step 42), the battery health state in the current cycle is: where: is the mathematical expectation of the state transition probability density function.

Images