Lithium battery health state prediction method and system

Abstract

The invention provides a lithium battery health state prediction method and system. The method comprises the following steps: acquiring a related original data sequence during charge-discharge circulation of a lithium battery; calculating a health factor feature sequence based on the related original data sequence; and constructing a prediction model, and inputting the residual component sequence, the fluctuation component sequence and the indirect health factor feature sequence into the prediction model to obtain a health state prediction result of the battery. Empirical mode decomposition is utilized to decompose a health factor feature sequence into a fluctuation sequence component representing capacity regeneration fluctuation and a trend sequence component representing an overall attenuation trend, so that the influence of lithium battery capacity regeneration on a prediction model is improved, and the prediction precision of lithium battery health state estimation is improved; by mining four indirect health factor feature sequences, the health state of the battery is predicted and estimated in a multi-dimensional manner, so that the change trend of the health state of the battery can be captured in an omnibearing manner.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and in particular to a method and system for predicting the health status of a lithium battery. Background Art

[0002] Lithium-ion batteries are widely used in energy storage due to their high energy conversion efficiency, strong safety, and environmental friendliness. Battery state of health (SOH), a core indicator of battery performance, plays a crucial role in battery management, operation, optimization, and maintenance.

[0003] Currently, there are three main methods for estimating the health status of lithium batteries: model-based methods, data-model fusion methods, and data-driven methods. The first two methods involve electrochemical mechanisms, making modeling more complex and highly dependent on modeling accuracy. Therefore, data-driven methods are still the primary prediction method. Most data-driven methods still rely on maximum discharge capacity as a single variable for prediction, failing to consider the influence of multiple variables, resulting in poor prediction results. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a lithium battery health status prediction method and system, aiming to solve the technical problem of poor prediction effect in the existing technology.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for predicting the health status of a lithium battery, comprising the following steps: Obtain relevant raw data sequence during the charge and discharge cycle of lithium batteries; An indirect health factor feature sequence and a direct health factor feature sequence are calculated based on the relevant original data sequence, wherein the indirect health factor feature sequence includes a thermal feature sequence, a power feature sequence, an energy feature sequence, and a voltage feature sequence of the battery, and the direct health factor feature sequence includes a battery health feature sequence of the battery; Performing empirical mode decomposition on the direct health factor characteristic sequence and the indirect health factor characteristic sequence to obtain a residual component sequence with a global change trend and a fluctuation component sequence representing capacity regeneration; Construct a deep learning model for battery health state prediction with multi-feature input. Use the indirect health factor feature sequence and the direct health factor feature sequence as a data set. Take part of the data set as a training set and the other part as a control set for training. By defining an optimizer and adjusting the loss function, a trained prediction model is obtained. The residual component sequence and the fluctuation component sequence are respectively input into the trend component processing module and the fluctuation component processing module of the prediction model to obtain the health status prediction result of the battery.

[0006] According to one aspect of the above technical solution, the step of obtaining a sequence of raw data related to the charge and discharge cycle of a lithium battery specifically includes: Conduct battery cycle aging experiments on the batteries to be tested. Perform multiple charge and discharge cycles on all lithium-ion batteries in the dataset. Each charge cycle includes six stages: constant voltage and constant current charging, resting, constant current discharging, resting, constant current discharging, and resting. This allows for the collection of raw data sequences containing battery data and health status data for each lithium-ion battery during each charge and discharge cycle. The relevant raw data sequence includes the temperature of the sampling point during the constant voltage charging stage, the charging current during the constant voltage charging stage, the battery storage energy during the constant voltage charging stage, the voltage during the shelf stage, the rated capacity of the lithium-ion battery when it was initially manufactured, and the current maximum discharge capacity of the lithium-ion battery.

[0007] According to one aspect of the above technical solution, the step of calculating the indirect health factor characteristic sequence and the direct health factor characteristic sequence based on the relevant original data sequence specifically includes: Extract the integral value of the temperature curve composed of the sampling point temperature in each cycle of the battery during the constant voltage charging stage of each charge and discharge cycle as the thermal feature sequence; The power curve trajectory length of the battery in the constant voltage charging stage under each charge and discharge cycle is extracted as the power feature sequence; The incremental energy of the battery before and after constant voltage charging in different cycles is extracted as the energy feature sequence; The relaxation voltage drop from the start to the end of the second rest phase in each charge-discharge cycle is extracted as a voltage characteristic sequence; Extract the battery health status under different charge and discharge cycles as the battery health feature sequence; The thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence are used as indirect health factor feature sequences, and the battery health feature sequence is used as a direct health factor feature sequence.

[0008] According to one aspect of the above technical solution, the calculation expression of the integral value of the temperature curve is: ; Where, is the integral value of the temperature curve during the cycle, The temperature variation curve of the cycle is shown as a function of time, t s Indicates the start time of the corresponding charging cycle, t c Indicates the end time of the corresponding charging cycle; The calculation expression of the power curve trajectory length of the battery is: ; ; Where, Represents the power curve function corresponding to the constant voltage stage in the charge and discharge cycle, It represents the current curve function corresponding to the constant voltage stage in the charge and discharge cycle, U is the voltage, Indicates the trajectory length of the power function curve during the cycle, Indicates the start time of the battery current time domain trajectory during this charging cycle, Indicates the end time of the battery current time domain trajectory during this charging cycle, Represents the derivative of the power curve function; The calculation expression of the incremental energy is: ; Where, is the incremental energy, and They represent the initial energy level and final energy level of the battery during a single constant voltage charging cycle; The calculation expression of the voltage drop is: ; Where, is the voltage drop, Indicates the end voltage of the second rest phase of the cycle. Indicates the starting voltage of the second rest phase of the cycle; The calculation expression of the battery health status is: ; Where, is the battery health status, Represents the rated capacity of the battery when it was initially manufactured. Represents the maximum discharge capacity of the lithium-ion battery in the current cycle.

[0009] According to one aspect of the above technical solution, the steps of obtaining a residual component sequence having a global variation trend and a fluctuation component sequence representing capacity regeneration specifically include: Step S1: Find the extreme points of the direct health factor characteristic sequence, connect the local maximum points into an upper envelope using a cubic spline curve, and connect the local minimum points into a lower envelope. The upper and lower envelopes contain all data points to obtain the maximum envelope and the minimum envelope; Step S2, averaging the maximum envelope and the minimum envelope to obtain an average value; Step S3, obtaining an intermediate signal based on a difference between a battery health feature sequence corresponding to the battery health state and the average value; Step S4, determining whether the intermediate signal satisfies an intrinsic mode function condition; Step S5: If the intermediate signal satisfies an intrinsic mode function condition, the intermediate signal is added as an intrinsic mode function to the fluctuation component sequence representing capacity regeneration, wherein the intrinsic mode function condition is: in the original feature data sequence of the battery health feature sequence, the sum of the number of local maxima and local minima is equal to or differs by at most one from the number of zero crossings, and at any time point, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero; Step S6: If the intermediate signal does not meet the intrinsic mode function condition, the intermediate signal is used as a new battery health feature sequence and steps S1 to S3 are repeated to perform secondary screening to obtain a screening signal until the intrinsic mode function condition is met. The screening signal that meets the intrinsic mode function condition is then used as the intrinsic mode function in the fluctuation component sequence. Step S7, obtaining a residual component sequence representing an overall downward trend of the battery health state based on the difference between the battery health feature sequence and the intrinsic mode function; Step S8: Using the residual component sequence as a new battery health feature sequence, and repeating steps S1 to S7 until the residual component sequence becomes a monotonic function or a constant, outputting and obtaining a final residual component sequence and a fluctuation component sequence composed of multiple intrinsic mode functions that meet the intrinsic mode function conditions; Step S9 , performing EMD decomposition on the thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence as original feature sequences, and finally obtaining a trend component feature sequence composed of trend components and a fluctuation component feature sequence composed of fluctuation components.

[0010] According to one aspect of the above technical solution, the step of inputting the residual component sequence and the fluctuation component sequence into the trend component processing module and the fluctuation component processing module of the prediction model respectively to obtain the battery health status prediction result specifically includes: All input feature data are preprocessed using a reversible instance normalization module to reduce the impact of data distribution shift, wherein the reversible instance normalization module includes a normalization layer and a denormalization layer; The trend component feature sequence and the fluctuation component feature sequence are normalized to obtain a normalized trend component feature sequence and a normalized fluctuation component feature sequence, which are respectively input into the trend feature processing module and the fluctuation feature processing module for feature sequence processing.

[0011] According to one aspect of the above technical solution, the calculation expression of the normalization layer is: ; ; ; ; Where, is the standardized data output by the normalization layer, 、 are affine parameter vectors, is the variance of the input data, is the mean of the input data, x t is the original input data, a is the length of the input time series, b is the number of variables in the input time series data, is a very small constant, j is the variable data number, is the variable data corresponding to the variable data sequence number j in a single variable sequence, and R represents the real number set of dimension; The calculation expression of the denormalization layer is: ; Where, is the time series before denormalization input, is the output after denormalization.

[0012] According to one aspect of the above technical solution, the calculation expression of the trend feature processing module is: ; ; ; ; ; ; Where, Represents the final output of the trend feature processing module, represents the subsequent T-th target variable predicted by the trend feature processing module, Indicates the i The input sequence of channels, Represents the first MLP layer, which transforms the input from the dimension l Mapping to hidden dimensions n 1, Indicates passing E The embedding representation after 1 layer, Represents the second MLP layer, which transforms the input from the dimension l Mapping to hidden dimensions n 2, represents the input embedding representation after the fully connected layer of the second layer MLP, represents regularization processing, and is the initial output of the trend feature processing module, represents element-wise addition, represents linear hysteresis processing, Represents linear projection processing; The calculation expression of the fluctuation feature processing module is: ; ; ; ; Where, Represents the sequence after input block, B Indicates the batch size, N p Indicates the number of blocks, L P represents the time step of each block, D Represents the embedding dimension, Block represents the improved temporal convolutional network module, Represents the embedding layer module in the improved temporal convolutional network module, Represents the sequence after input block, Indicates the intermediate transition amount The new intermediate transition amount obtained after the fluctuation feature processing module, It represents the intermediate transition amount obtained after the original input data is processed by the intermediate k improved temporal convolutional network modules.

[0013] According to one aspect of the above technical solution, the training process of the deep learning model includes the following steps: The first-order moment estimate and the second-order moment estimate are updated based on the following formula: ; ; Where, and represents the first-order moment estimate of the current and previous iterations, and represents the second-order moment estimate of the current and previous iterations, represents the current gradient, and represents the decay rate, and t represents the current number of iterations; The bias-corrected first-order moment estimate is calculated based on the following formula: and second-order moment estimation : ; ; Where, and represents the hyperparameter corresponding to the number of iterations; Calculate the current variance ratio based on the following formula and limiting variance ratios : ; ; when When > 4, the adaptive learning rate correction parameter is calculated based on the second-order moment of deviation correction and the correction factor according to the following formula: ; ; ; Where, is the bias-corrected second moment, is the correction factor; when When ≤4, the calculation expression of the adaptive learning rate correction parameter is: ; Where, and Represent the learning rate and smoothing factor respectively; The calculation expression of the loss function is: ; Where, RMSE represents the loss function, represents the predicted value of the mth sample, represents the true value of the mth sample, Indicates the sample size.

[0014] In a second aspect, the present application also provides a lithium battery health status prediction system, comprising: Data module, used to obtain relevant raw data sequence during the charge and discharge cycle of lithium batteries; a health feature module, configured to calculate an indirect health factor feature sequence and a direct health factor feature sequence based on the relevant raw data sequence, wherein the indirect health factor feature sequence includes a thermal feature sequence, a current feature sequence, and an electrical feature sequence of the battery, and the direct health factor feature sequence includes a battery health feature sequence of the battery; A modal decomposition module is used to perform empirical mode decomposition on the direct health factor characteristic sequence and the indirect health factor characteristic sequence to obtain a residual component sequence with a global change trend and a fluctuation component sequence representing capacity regeneration; A training module is used to build a deep learning model for predicting the health status of batteries using multiple feature inputs. The indirect health factor feature sequence and the direct health factor feature sequence are used as a data set. Part of the data set is used as a training set, and the other part is used as a control set for training. By defining an optimizer and adjusting the loss function, a trained prediction model is obtained. The prediction module is used to input the residual component sequence and the fluctuation component sequence into the trend component processing module and the fluctuation component processing module of the prediction model respectively to obtain the health status prediction result of the battery.

[0015] According to one aspect of the above technical solution, the data module is specifically used to: Conduct battery cycle aging experiments on the batteries to be tested. Perform multiple charge and discharge cycles on all lithium-ion batteries in the dataset. Each charge cycle includes six stages: constant voltage and constant current charging, resting, constant current discharging, resting, constant current discharging, and resting. This allows for the collection of raw data sequences containing battery data and health status data for each lithium-ion battery during each charge and discharge cycle. The relevant raw data sequence includes the temperature of the sampling point during the constant voltage charging stage, the charging current during the constant voltage charging stage, the battery storage energy during the constant voltage charging stage, the voltage during the shelf stage, the rated capacity of the lithium-ion battery when it was initially manufactured, and the current maximum discharge capacity of the lithium-ion battery.

[0016] According to one aspect of the above technical solution, the health feature module is specifically used to: Extract the integral value of the temperature curve composed of the sampling point temperature in each cycle of the battery during the constant voltage charging stage of each charge and discharge cycle as the thermal feature sequence; The power curve trajectory length of the battery in the constant voltage charging stage under each charge and discharge cycle is extracted as the power feature sequence; The incremental energy of the battery before and after constant voltage charging in different cycles is extracted as the energy feature sequence; The relaxation voltage drop from the start to the end of the second rest phase in each charge-discharge cycle is extracted as a voltage characteristic sequence; Extract the battery health status under different charge and discharge cycles as the battery health feature sequence; The thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence are used as indirect health factor feature sequences, and the battery health feature sequence is used as a direct health factor feature sequence.

[0017] According to one aspect of the above technical solution, the modal decomposition module is specifically used to: Step S1: Find the extreme points of the direct health factor characteristic sequence, connect the local maximum points into an upper envelope using a cubic spline curve, and connect the local minimum points into a lower envelope. The upper and lower envelopes contain all data points to obtain the maximum envelope and the minimum envelope; Step S2, averaging the maximum envelope and the minimum envelope to obtain an average value; Step S3, obtaining an intermediate signal based on a difference between a battery health feature sequence corresponding to the battery health state and the average value; Step S4, determining whether the intermediate signal satisfies an intrinsic mode function condition; Step S5: If the intermediate signal satisfies an intrinsic mode function condition, the intermediate signal is added as an intrinsic mode function to the fluctuation component sequence representing capacity regeneration, wherein the intrinsic mode function condition is: in the original feature data sequence of the battery health feature sequence, the sum of the number of local maxima and local minima is equal to or differs by at most one from the number of zero crossings, and at any time point, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero; Step S6: If the intermediate signal does not meet the intrinsic mode function condition, the intermediate signal is used as a new battery health feature sequence and steps S1 to S3 are repeated to perform secondary screening to obtain a screening signal until the intrinsic mode function condition is met. The screening signal that meets the intrinsic mode function condition is then used as the intrinsic mode function in the fluctuation component sequence. Step S7, obtaining a residual component sequence representing an overall downward trend of the battery health state based on the difference between the battery health feature sequence and the intrinsic mode function; Step S8: Using the residual component sequence as a new battery health feature sequence, and repeating steps S1 to S7 until the residual component sequence becomes a monotonic function or a constant, outputting and obtaining a final residual component sequence and a fluctuation component sequence composed of multiple intrinsic mode functions that meet the intrinsic mode function conditions; Step S9 , performing EMD decomposition on the thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence as original feature sequences, and finally obtaining a trend component feature sequence composed of trend components and a fluctuation component feature sequence composed of fluctuation components.

[0018] According to one aspect of the above technical solution, the prediction module is specifically used to: All input feature data are preprocessed using a reversible instance normalization module to reduce the impact of data distribution shift, wherein the reversible instance normalization module includes a normalization layer and a denormalization layer; The trend component feature sequence and the fluctuation component feature sequence are normalized to obtain a normalized trend component feature sequence and a normalized fluctuation component feature sequence, which are respectively input into the trend feature processing module and the fluctuation feature processing module for feature sequence processing.

[0019] Compared with the prior art, the beneficial effects of the present invention are: using empirical mode decomposition to decompose the health factor feature sequence into a fluctuation sequence component representing the capacity regeneration fluctuation and a trend sequence component representing the overall attenuation trend, thereby improving the impact of the prediction model brought by the "capacity regeneration" of the lithium battery and improving the prediction accuracy of the lithium battery health status estimation; by mining the four indirect health factor feature sequences, the battery health status is predicted and estimated in multiple dimensions, thereby being able to capture the changing trend of the battery health status in all directions; by constructing the D-DMIT algorithm model, each module is coordinated in parallel, and is continuously iterated and trained by inputting feature data, while considering the linkage relationship between the long-term and short-term features of the time series, thereby being able to form a lithium battery health status estimation model with strong generalization ability and interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a method for predicting the health status of a lithium battery in a first embodiment of the present invention; Figure 2 This is a structural block diagram of a lithium battery health status prediction system in a second embodiment of the present invention; The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0021] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate several embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0022] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] Example 1 See also Figure 1 , which shows a method for predicting the health status of a lithium battery in a first embodiment of the present invention. As shown in the figure, the method includes the following steps: Step S100, obtaining a sequence of raw data related to the charge and discharge cycle of the lithium battery. Preferably, in this embodiment, the step of obtaining a sequence of raw data related to the charge and discharge cycle of the lithium battery specifically includes: Conduct battery cycle aging experiments on the batteries to be tested. Perform multiple charge and discharge cycles on all lithium-ion batteries in the dataset. Each charge cycle includes six stages: constant voltage and constant current charging, resting, constant current discharging, resting, constant current discharging, and resting. This allows for the collection of raw data sequences containing battery data and health status data for each lithium-ion battery during each charge and discharge cycle. The relevant raw data sequence includes the temperature of the sampling point during the constant voltage charging stage, the charging current during the constant voltage charging stage, the battery storage energy during the constant voltage charging stage, the voltage during the shelf stage, the rated capacity of the lithium-ion battery when it was initially manufactured, and the current maximum discharge capacity of the lithium-ion battery. In some application scenarios of this embodiment, a battery cycle aging experiment is performed on the battery to be tested using an existing battery testing system. The BTS 4000 in the battery testing platform system is used to measure battery-related internal parameters, wherein the thermocouple sensor of the auxiliary channel is used to measure the battery temperature, and the Hall sensor inside the BTS 4000 is used to measure the current inside the battery.

[0025] Step S200, based on the relevant original data sequence, an indirect health factor feature sequence and a direct health factor feature sequence are calculated, wherein the indirect health factor feature sequence includes the thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence of the battery, and the direct health factor feature sequence includes the battery health feature sequence of the battery.

[0026] In this embodiment, the step of calculating the indirect health factor characteristic sequence and the direct health factor characteristic sequence based on the relevant original data sequence specifically includes: The integral value of the temperature curve formed by the temperature of the sampling points in each cycle of the battery in the constant voltage charging stage of each charge and discharge cycle is extracted as the thermal feature sequence. The calculation expression of the integral value of the temperature curve is: ; Where, is the integral value of the temperature curve during the cycle, The temperature variation curve of the cycle is shown as a function of time, t s Indicates the start time of the corresponding charging cycle, t c Indicates the end time of the corresponding charging cycle.

[0027] The power curve trajectory length of the battery in the constant voltage charging stage of each charge and discharge cycle is extracted as the power feature sequence. The calculation expression of the battery power curve trajectory length is: ; ; Where, Represents the power curve function corresponding to the constant voltage stage in the charge and discharge cycle, It represents the current curve function corresponding to the constant voltage stage in the charge and discharge cycle, U is the voltage, Indicates the trajectory length of the power function curve during the cycle, Indicates the start time of the battery current time domain trajectory during this charging cycle, Indicates the end time of the battery current time domain trajectory during this charging cycle, Represents the derivative of the power curve function.

[0028] The incremental energy of the battery before and after constant voltage charging in different cycles is extracted as the energy feature sequence. The calculation expression of the incremental energy is: ; Where, is the incremental energy, and They represent the initial energy level and final energy level of the battery during a single constant voltage charging cycle.

[0029] The relaxation voltage drop from the start to the end of the second rest phase in each charge-discharge cycle is extracted as a voltage feature sequence. The voltage drop is calculated as: ; Where, is the voltage drop, Indicates the end voltage of the second rest phase of the cycle. Indicates the starting voltage of the second rest phase of the cycle; The battery health status under different charge and discharge cycles is extracted as the battery health feature sequence. The calculation expression of the battery health status is: ; Where, is the battery health status, Represents the rated capacity of the battery when it was initially manufactured. Represents the maximum discharge capacity of the lithium-ion battery in the current cycle.

[0030] The thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence are used as indirect health factor feature sequences, and the battery health feature sequence is used as a direct health factor feature sequence.

[0031] Step S300 , performing empirical mode decomposition on the direct health factor characteristic sequence and the indirect health factor characteristic sequence to obtain a residual component sequence with a global change trend and a fluctuation component sequence representing capacity regeneration.

[0032] Preferably, in this embodiment, the step of obtaining a residual component sequence having a global variation trend and a fluctuation component sequence representing capacity regeneration specifically includes: Step S1, find the extreme points of the direct health factor characteristic sequence, connect the local maximum points into an upper envelope line through a cubic spline curve, and connect the local minimum points into a lower envelope line. The upper and lower envelope lines contain all data points to obtain the maximum envelope line and the minimum envelope line.

[0033] Step S2: averaging the maximum envelope and the minimum envelope to obtain an average value. The calculation formula for the average value is: ; Where, is the average value, and They are the maximum envelope and the minimum envelope respectively; Step S3, obtaining an intermediate signal based on a difference between a battery health feature sequence corresponding to the battery health state and the average value; The calculation formula of the intermediate signal is: ; Where, is the intermediate signal.

[0034] Step S4, determining whether the intermediate signal satisfies an intrinsic mode function condition; Step S5: If the intermediate signal satisfies an intrinsic mode function condition, the intermediate signal is added as an intrinsic mode function to the fluctuation component sequence representing capacity regeneration, wherein the intrinsic mode function condition is: in the original feature data sequence of the battery health feature sequence, the sum of the number of local maxima and local minima is equal to or differs by at most one from the number of zero crossings, and at any time point, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero; Step S6: If the intermediate signal does not meet the intrinsic mode function condition, the intermediate signal is used as a new battery health feature sequence and steps S1 to S3 are repeated to perform secondary screening to obtain a screening signal until the intrinsic mode function condition is met. The screening signal that meets the intrinsic mode function condition is then used as the intrinsic mode function in the fluctuation component sequence. Among them, the calculation formula for secondary screening is: ; The intermediate signal As the original feature sequence, repeat the contents of step S1 to step S3 to reconstruct the upper and lower envelopes for the second screening. 11 (cycle) is the mean of the upper and lower envelopes of A1(cycle), To filter the signal; repeat until Based on the definition of the intrinsic mode function, we can get the first-order intrinsic mode function IMF1(cycle)=B 1n (cycle), where n represents the final number of iterative screening times that meets the intrinsic mode function conditions.

[0035] Step S7: obtaining a residual component sequence representing the overall downward trend of the battery health state based on the difference between the battery health feature sequence and the intrinsic mode function; the expression of the residual component sequence is: ; Where, is the residual component sequence, is the battery health status, is the volatility component sequence.

[0036] In step S8, the residual component sequence is used as a new battery health feature sequence, and steps S1 to S7 are repeated until the residual component sequence becomes a monotonic function or a constant, and the final residual component sequence and the fluctuation component sequence composed of multiple intrinsic mode functions that meet the intrinsic mode function conditions are output and obtained.

[0037] The final sequence is: ; in, is the natural mode component, It is the residual component, representing the average trend of the feature sequence, usually a monotonic sequence or a constant sequence. Here, it is used as the trend component to represent the overall change trend of the relevant features in the feature sequence; It is the accumulation of the first-stage intrinsic mode function to the n-th-order intrinsic mode function, which can be used as the fluctuation component here to represent the characteristic sequence fluctuation caused by the battery capacity regeneration.

[0038] Step S9, the thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence are used as original feature sequences for EMD decomposition, and finally a trend component feature sequence composed of trend components and a fluctuation component feature sequence composed of fluctuation components are obtained. The four indirect health factor feature sequences are used as original feature sequences for EMD decomposition, and thus two sets of feature sequences are obtained: one set representing the trend component and the other set representing the trend component. , a set of fluctuation component characteristic sequences composed of fluctuation components representing the fluctuation situation .in, , ( =1, 2, 3, 4, 5) represents the residual component sequence of the feature sequence corresponding to the sequence number in step S300, Indicates the first T s time steps. , ( s =1, 2, 3, 4, 5) represents the fluctuation component sequence of the characteristic sequence corresponding to the sequence number in step S300, Indicates the first The above two sets of feature sequences will be used as input data for subsequent model training and testing.

[0039] Step S400 constructs a deep learning model for battery health prediction using multiple feature inputs. The indirect health factor feature sequences and the direct health factor feature sequences serve as datasets. A portion of the dataset is used as a training set, and another portion is used as a control set for training. By defining an optimizer and adjusting the loss function, a trained prediction model is obtained. In this embodiment, the model is a Decomposition-Dual Mamba Improved Temporal Convolutional Network model (D-DMIT model), which includes a feature normalization module based on a reversible instance regularization module, a trend component processing module based on the improved structured state space model of the Double Mamba, and a fluctuation component processing module based on the improved temporal convolutional network. The trend component feature sequence obtained in step S300 is normalized and input into the trend component processing module of the model. The fluctuation component feature sequence is also normalized and input into the fluctuation component processing module of the D-DMIT model. Finally, the results of the two are fused to obtain the final battery health prediction result.

[0040] Preferably, in this embodiment, the step of inputting the residual component sequence and the fluctuation component sequence into the trend component processing module and the fluctuation component processing module of the prediction model respectively to obtain the battery health status prediction result specifically includes: All input feature data are preprocessed using a reversible instance normalization module to reduce the impact of data distribution shift, wherein the reversible instance normalization module includes a normalization layer and a denormalization layer; The calculation expression of the normalization layer is: ; ; ; ; Where, is the standardized data output by the normalization layer, 、 are affine parameter vectors, is the variance of the input data, is the mean of the input data, x t is the original input data, a is the length of the input time series, b is the number of variables in the input time series data, is a very small constant, j is the variable data number, is the variable data corresponding to the variable data sequence number j in a single variable sequence, and R represents the real number set of dimension; The calculation expression of the denormalization layer is: ; Where, is the time series before denormalization input, is the output after denormalization.

[0041] The trend component processing module of DoubleMamba, based on the improved structured state-space model, exploits the unique properties of time series data and extracts the potential representation of time series at global and local scales by utilizing two parallel Mamba modules, thereby completing the long-term prediction of multivariate time series.

[0042] The calculation expression of the trend feature processing module is: ; ; ; ; ; ; Where, Represents the final output of the trend feature processing module, represents the subsequent T-th target variable predicted by the trend feature processing module, Indicates the i The input sequence of channels, Represents the first MLP layer, which transforms the input from the dimension l Mapping to hidden dimensions n 1, Indicates passing E The embedding representation after 1 layer, Represents the second MLP layer, which transforms the input from the dimension l Mapping to hidden dimensions n 2, represents the input embedding representation after the fully connected layer of the second layer MLP, represents regularization processing, and is the initial output of the trend feature processing module, represents element-wise addition, represents linear hysteresis processing, Represents linear projection processing.

[0043] Specifically, for the input feature sequence ∈ R B×b×l ,in B Indicates the batch size, l Indicates the length of the lookback window. bRepresents the number of channels, that is, the number of input variables. First, reshape the input data dimension into ( B × b )×1× l The data type of , thus ensuring the adoption of a channel-independent strategy. First, it is embedded through two layers of MLP linear fully connected layers, which can be expressed as:

[0044]

[0045] in Indicates the i The input sequence of channels, Represents the fully connected layer of the first layer MLP, which transforms the input from the dimension l Mapping to hidden dimensions n 1, Indicates passing E The embedding representation after 1 layer, Represents the input embedding representation after passing through the fully connected layer of the second MLP layer.

[0046] Specific operations include:

[0047] in is the weight matrix, is the bias matrix.

[0048] Then enter E 2 before, yes First, apply Dropout to prevent overfitting, and then pass it through the second MLP layer E 2. Further dimension adjustment is performed to obtain subsequent input .

[0049] After the input is embedded, two Mamba modules are used to capture long-term dependencies and context information. One of the two Mamba modules processes a token of length n 2 and input dimension is 1, and another module processes input with token length 1 and input dimension n 2 input. Each Mamba module adopts a dual-branch structure. Branch 1 consists of a 1D causal convolution layer, SiLU activation function and S6 model, while branch 2 consists of a simple linear mapping and SiLU activation function. The outputs of branch 1 and branch 2 are combined through element-wise multiplication to form the output of a Mamba module; the outputs of the two Mamba modules are and , then after being processed by two Mamba modules, the intermediate tensor is ,in For element-wise addition, the intermediate tensor Then through the linear projection layer P 1 get , and then through element-wise addition we get ,at last Through the linear hysteresis layer P 2 Get the final output of the trend component processing module . Represents the subsequent T The target variable SOH.

[0050] The calculation expression of the fluctuation feature processing module is: ; ; ; ; Where, Represents the sequence after input block, B Indicates the batch size, N p Indicates the number of blocks, L P represents the time step of each block, D Represents the embedding dimension, Block represents the improved temporal convolutional network module, Represents the embedding layer module in the improved temporal convolutional network module, Represents the sequence after input block, Indicates the intermediate transition amount The new intermediate transition amount obtained after the fluctuation feature processing module, It represents the intermediate transition amount obtained after the original input data is processed by the intermediate k improved temporal convolutional network modules.

[0051] Specifically, for the above-mentioned fluctuation component processing module based on the improved temporal convolutional network, for the input feature sequence ,satisfy: ∈ R B×b×l ,in, B Indicates the batch size, l Indicates the length of the lookback window. b It represents the number of channels, that is, the number of input variables. The input feature sequence is embedded into the feature model through the block embedding method. Specifically, the time series is divided into local blocks and mapped to the high-dimensional space through linear projection. It can be expressed by the formula .in Represents the sequence after input block, whereB Indicates the batch size, N p Indicates the number of blocks, L P represents the time step of each block, D represents the embedding dimension.

[0052] Afterwards, Input to the subsequent processing module, the subsequent processing module is composed of multiple improved temporal convolutional network modules stacked together, for the intermediate transition amount processed by the kth improved temporal convolutional network module , the processing method is , where Block represents the improved temporal convolutional network module, which can be expressed as .

[0053] Each improved temporal convolutional network module includes a deep large-kernel convolution to capture temporal dependencies and two consecutive point-wise group convolutional feedforward neural networks to capture cross-variable dependencies. After passing through multiple improved temporal convolutional network modules, the final output of the fluctuation component processing module, i.e., the predicted result of the target variable SOH, is obtained through a residual connection network. The prediction results of the fluctuation component processing module and the trend component processing module are added, fused, and denormalized to obtain the final prediction result.

[0054] The training process of the deep learning model includes the following steps: The first-order moment estimate and the second-order moment estimate are updated based on the following formula: ; ; Where, and represents the first-order moment estimate of the current and previous iterations, and represents the second-order moment estimate of the current and previous iterations, represents the current gradient, and represents the decay rate, and t represents the current number of iterations; The bias-corrected first-order moment estimate is calculated based on the following formula: and second-order moment estimation : ; ; Where, and represents the hyperparameter corresponding to the number of iterations; Calculate the current variance ratio based on the following formula and limiting variance ratios : ; ; when When > 4, the adaptive learning rate correction parameter is calculated based on the second-order moment of deviation correction and the correction factor according to the following formula: ; ; ; Where, is the bias-corrected second moment, is the correction factor; when When ≤4, the calculation expression of the adaptive learning rate correction parameter is: ; Where, and Represent the learning rate and smoothing factor respectively; The calculation expression of the loss function is: ; Where, RMSE represents the loss function, represents the predicted value of the mth sample, represents the true value of the mth sample, Indicates the sample size.

[0055] Step S500: Input the residual component sequence and the fluctuation component sequence into the trend component processing module and the fluctuation component processing module of the prediction model respectively to obtain the battery health status prediction result. The trained prediction model is used to output the lithium-ion battery SOH to be predicted.

[0056] In summary, the lithium battery health status prediction method in the above embodiment of the present invention uses empirical mode decomposition to decompose the health factor feature sequence into a fluctuation sequence component representing the capacity regeneration fluctuation and a trend sequence component representing the overall attenuation trend, thereby improving the prediction model impact brought by the "capacity regeneration" of the lithium battery and improving the prediction accuracy of the lithium battery health status estimation; by mining the four indirect health factor feature sequences, the battery health status is predicted and estimated in multiple dimensions, thereby being able to capture the changing trend of the battery health status in all directions; by constructing the D-DMIT algorithm model, each module is coordinated in parallel, and is continuously iteratively trained by inputting feature data, while considering the linkage relationship between the long-term characteristics and short-term characteristics of the time series, thereby being able to form a lithium battery health status estimation model with strong generalization ability and interpretability.

[0057] Example 2 The second embodiment of the present application also provides a lithium battery health status prediction system, which is used to implement the embodiments and preferred implementation methods, and will not be repeated here. As used below, the terms "module", "unit", "sub-unit", etc. can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, it is also possible and conceivable to implement it in hardware, or a combination of software and hardware.

[0058] like Figure 2 As shown, the system includes: a data module 100, a health feature module 200, a modal decomposition module 300, a training module 400 and a prediction module 500; The data module 100 is used to obtain the relevant raw data sequence during the charge and discharge cycle of the lithium battery; The health feature module 200 is used to calculate an indirect health factor feature sequence and a direct health factor feature sequence based on the relevant original data sequence, wherein the indirect health factor feature sequence includes a thermal feature sequence, a power feature sequence, an energy feature sequence, and a voltage feature sequence of the battery, and the direct health factor feature sequence includes a battery health feature sequence of the battery; The modal decomposition module 300 is used to perform empirical mode decomposition on the direct health factor characteristic sequence and the indirect health factor characteristic sequence to obtain a residual component sequence with a global change trend and a fluctuation component sequence representing capacity regeneration; The training module 400 is used to build a deep learning model for multi-feature input battery health state prediction. The indirect health factor feature sequence and the direct health factor feature sequence are used as a data set. Part of the data set is used as a training set, and the other part is used as a control set for training. By defining an optimizer and adjusting the loss function, a trained prediction model is obtained. The prediction module 500 is used to input the residual component sequence and the fluctuation component sequence into the trend component processing module and the fluctuation component processing module of the prediction model respectively to obtain the health status prediction result of the battery.

[0059] Preferably, in this embodiment, the data module 100 is specifically used for: Conduct battery cycle aging experiments on the batteries to be tested. Perform multiple charge and discharge cycles on all lithium-ion batteries in the dataset. Each charge cycle includes six stages: constant voltage and constant current charging, resting, constant current discharging, resting, constant current discharging, and resting. This allows for the collection of raw data sequences containing battery data and health status data for each lithium-ion battery during each charge and discharge cycle. The relevant raw data sequence includes the temperature of the sampling point during the constant voltage charging stage, the charging current during the constant voltage charging stage, the battery storage energy during the constant voltage charging stage, the voltage during the shelf stage, the rated capacity of the lithium-ion battery when it was initially manufactured, and the current maximum discharge capacity of the lithium-ion battery.

[0060] Preferably, in this embodiment, the health characteristic module 200 is specifically used to: Extract the integral value of the temperature curve composed of the sampling point temperature in each cycle of the battery during the constant voltage charging stage of each charge and discharge cycle as the thermal feature sequence; The power curve trajectory length of the battery in the constant voltage charging stage under each charge and discharge cycle is extracted as the power feature sequence; The incremental energy of the battery before and after constant voltage charging in different cycles is extracted as the energy feature sequence; The relaxation voltage drop from the start to the end of the second rest phase in each charge-discharge cycle is extracted as a voltage characteristic sequence; Extract the battery health status under different charge and discharge cycles as the battery health feature sequence; The thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence are used as indirect health factor feature sequences, and the battery health feature sequence is used as a direct health factor feature sequence.

[0061] Preferably, in this embodiment, the modal decomposition module 300 is specifically used to: Step S1: Find the extreme points of the direct health factor characteristic sequence, connect the local maximum points into an upper envelope using a cubic spline curve, and connect the local minimum points into a lower envelope. The upper and lower envelopes contain all data points to obtain the maximum envelope and the minimum envelope; Step S2, averaging the maximum envelope and the minimum envelope to obtain an average value; Step S3, obtaining an intermediate signal based on a difference between a battery health feature sequence corresponding to the battery health state and the average value; Step S4, determining whether the intermediate signal satisfies an intrinsic mode function condition; Step S5: If the intermediate signal satisfies an intrinsic mode function condition, the intermediate signal is added as an intrinsic mode function to the fluctuation component sequence representing capacity regeneration, wherein the intrinsic mode function condition is: in the original feature data sequence of the battery health feature sequence, the sum of the number of local maxima and local minima is equal to or differs by at most one from the number of zero crossings, and at any time point, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero; Step S6: If the intermediate signal does not meet the intrinsic mode function condition, the intermediate signal is used as a new battery health feature sequence and steps S1 to S3 are repeated to perform secondary screening to obtain a screening signal until the intrinsic mode function condition is met. The screening signal that meets the intrinsic mode function condition is then used as the intrinsic mode function in the fluctuation component sequence. Step S7, obtaining a residual component sequence representing an overall downward trend of the battery health state based on the difference between the battery health feature sequence and the intrinsic mode function; Step S8: Using the residual component sequence as a new battery health feature sequence, and repeating steps S1 to S7 until the residual component sequence becomes a monotonic function or a constant, outputting and obtaining a final residual component sequence and a fluctuation component sequence composed of multiple intrinsic mode functions that meet the intrinsic mode function conditions; Step S9 , performing EMD decomposition on the thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence as original feature sequences, and finally obtaining a trend component feature sequence composed of trend components and a fluctuation component feature sequence composed of fluctuation components.

[0062] Preferably, in this embodiment, the prediction module 500 is specifically used for: All input feature data are preprocessed using a reversible instance normalization module to reduce the impact of data distribution shift, wherein the reversible instance normalization module includes a normalization layer and a denormalization layer; The trend component feature sequence and the fluctuation component feature sequence are normalized to obtain a normalized trend component feature sequence and a normalized fluctuation component feature sequence, which are respectively input into the trend feature processing module and the fluctuation feature processing module for feature sequence processing.

[0063] It should be noted that each module can be a functional module or a program module, and can be implemented by software or hardware. For modules implemented by hardware, each module can be located in the same processor; or each module can be located in different processors in any combination.

[0064] The technical features of the above-described embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for predicting the health status of a lithium battery, characterized in that: The following steps are involved: Obtain relevant raw data sequence during the charge and discharge cycle of lithium batteries; An indirect health factor feature sequence and a direct health factor feature sequence are calculated based on the relevant original data sequence, wherein the indirect health factor feature sequence includes a thermal feature sequence, a power feature sequence, an energy feature sequence, and a voltage feature sequence of the battery, and the direct health factor feature sequence includes a battery health feature sequence of the battery; Performing empirical mode decomposition on the direct health factor characteristic sequence and the indirect health factor characteristic sequence to obtain a residual component sequence with a global change trend and a fluctuation component sequence representing capacity regeneration; Construct a deep learning model for battery health status prediction with multiple feature inputs. The indirect health factor feature sequence and the direct health factor feature sequence are used as a data set. Part of the data set is used as a training set, and the other part is used as a control set for training. By defining an optimizer and adjusting the loss function, a trained prediction model is obtained. The residual component sequence and the fluctuation component sequence are respectively input into the trend component processing module and the fluctuation component processing module of the prediction model to obtain the health status prediction result of the battery.

2. The lithium battery health status prediction method according to claim 1, characterized in that: The steps of obtaining the raw data sequence related to the charge and discharge cycle of the lithium battery specifically include: Conduct battery cycle aging experiments on the batteries to be tested. Perform multiple charge and discharge cycles on all lithium-ion batteries in the dataset. Each charge cycle includes six stages: constant voltage and constant current charging, resting, constant current discharging, resting, constant current discharging, and resting. This allows for the collection of raw data sequences containing battery data and health status data for each lithium-ion battery during each charge and discharge cycle. The relevant raw data sequence includes the temperature of the sampling point during the constant voltage charging stage, the charging current during the constant voltage charging stage, the battery storage energy during the constant voltage charging stage, the voltage during the shelf stage, the rated capacity of the lithium-ion battery when it was initially manufactured, and the current maximum discharge capacity of the lithium-ion battery.

3. The lithium battery health status prediction method according to claim 2, characterized in that: The steps of calculating the indirect health factor characteristic sequence and the direct health factor characteristic sequence based on the relevant original data sequence specifically include: Extract the integral value of the temperature curve composed of the sampling point temperature in each cycle of the battery during the constant voltage charging stage of each charge and discharge cycle as the thermal feature sequence; The power curve trajectory length of the battery in the constant voltage charging stage under each charge and discharge cycle is extracted as the power feature sequence; The incremental energy of the battery before and after constant voltage charging in different cycles is extracted as the energy feature sequence; The relaxation voltage drop from the start to the end of the second rest phase in each charge-discharge cycle is extracted as a voltage feature sequence; Extract the battery health status under different charge and discharge cycles as the battery health feature sequence; The thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence are used as indirect health factor feature sequences, and the battery health feature sequence is used as a direct health factor feature sequence.

4. The lithium battery health status prediction method according to claim 3, characterized in that: The calculation expression of the integral value of the temperature curve is: ； Where, is the integral value of the temperature curve during the cycle, The temperature variation curve of the cycle is shown as a function of time, t s Indicates the start time of the corresponding charging cycle, t c Indicates the end time of the corresponding charging cycle; The calculation expression of the power curve trajectory length of the battery is: ； ； Where, Represents the power curve function corresponding to the constant voltage stage in the charge and discharge cycle, It represents the current curve function corresponding to the constant voltage stage in the charge and discharge cycle, U is the voltage, Indicates the trajectory length of the power function curve during the cycle, Indicates the start time of the battery current time domain trajectory during this charging cycle, Indicates the end time of the battery current time domain trajectory during this charging cycle, Represents the derivative of the power curve function; The calculation expression of the incremental energy is: ； Where, is the incremental energy, and They represent the initial energy level and final energy level of the battery during a single constant voltage charging cycle; The calculation expression of the voltage drop is: ； Where, is the voltage drop, Indicates the end voltage of the second rest phase of the cycle. Indicates the starting voltage of the second rest phase of the cycle; The calculation expression of the battery health status is: ； Where, is the battery health status, Represents the rated capacity of the battery when it was initially manufactured. Represents the maximum discharge capacity of the lithium-ion battery in the current cycle.

5. The lithium battery health status prediction method according to claim 3, characterized in that: The steps of obtaining a residual component sequence with a global variation trend and a fluctuation component sequence representing capacity regeneration specifically include: Step S1: Find the extreme points of the direct health factor characteristic sequence, connect the local maximum points into an upper envelope using a cubic spline curve, and connect the local minimum points into a lower envelope. The upper and lower envelopes contain all data points to obtain the maximum envelope and the minimum envelope; Step S2, averaging the maximum envelope and the minimum envelope to obtain an average value; Step S3, obtaining an intermediate signal based on a difference between a battery health feature sequence corresponding to the battery health state and the average value; Step S4, determining whether the intermediate signal satisfies an intrinsic mode function condition; Step S5: If the intermediate signal satisfies an intrinsic mode function condition, the intermediate signal is added as an intrinsic mode function to the fluctuation component sequence representing capacity regeneration, wherein the intrinsic mode function condition is: in the original feature data sequence of the battery health feature sequence, the sum of the number of local maxima and local minima is equal to or differs by at most one from the number of zero crossings, and at any time point, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero; Step S6: If the intermediate signal does not meet the intrinsic mode function condition, the intermediate signal is used as a new battery health feature sequence and steps S1 to S3 are repeated to perform secondary screening to obtain a screening signal until the intrinsic mode function condition is met. The screening signal that meets the intrinsic mode function condition is then used as the intrinsic mode function in the fluctuation component sequence. Step S7, obtaining a residual component sequence representing an overall downward trend of the battery health state based on the difference between the battery health feature sequence and the intrinsic mode function; Step S8: Using the residual component sequence as a new battery health feature sequence, and repeating steps S1 to S7 until the residual component sequence becomes a monotonic function or a constant, outputting and obtaining a final residual component sequence and a fluctuation component sequence composed of multiple intrinsic mode functions that meet the intrinsic mode function conditions; Step S9 , performing EMD decomposition on the thermal feature sequence, power feature sequence, energy feature sequence and voltage feature sequence as original feature sequences, and finally obtaining a trend component feature sequence composed of trend components and a fluctuation component feature sequence composed of fluctuation components.

6. The lithium battery health status prediction method according to claim 5, characterized in that: The step of inputting the residual component sequence and the fluctuation component sequence into the trend component processing module and the fluctuation component processing module of the prediction model to obtain the health status prediction result of the battery specifically includes: All input feature data are preprocessed using a reversible instance normalization module to reduce the impact of data distribution shift, wherein the reversible instance normalization module includes a normalization layer and a denormalization layer; The trend component feature sequence and the fluctuation component feature sequence are normalized to obtain a normalized trend component feature sequence and a normalized fluctuation component feature sequence, which are respectively input into the trend feature processing module and the fluctuation feature processing module for feature sequence processing.

7. The lithium battery health status prediction method according to claim 6, characterized in that: The calculation expression of the normalization layer is: ； ； ； ； Where, is the standardized data output by the normalization layer, 、 are affine parameter vectors, is the variance of the input data, is the mean of the input data, x t is the original input data, a is the length of the input time series, b is the number of variables in the input time series data, is a very small constant, j is the variable data number, is the variable data corresponding to the variable data sequence number j in a single variable sequence, and R represents the real number set of dimension; The calculation expression of the denormalization layer is: ； Where, is the time series before denormalization input, is the output after denormalization.

8. The lithium battery health status prediction method according to claim 6, characterized in that: The calculation expression of the trend feature processing module is: ； ； ； ； ； ； Where, Represents the final output of the trend feature processing module, represents the subsequent T-th target variable predicted by the trend feature processing module, Indicates the i The input sequence of channels, Represents the first MLP layer, which transforms the input from the dimension l Mapping to hidden dimensions n 1, Indicates passing E The embedding representation after 1 layer, Represents the second MLP layer, which transforms the input from the dimension l Mapping to hidden dimensions n 2, represents the input embedding representation after the fully connected layer of the second layer MLP, represents regularization processing, and is the initial output of the trend feature processing module, represents element-wise addition, represents linear hysteresis processing, Represents linear projection processing; The calculation expression of the fluctuation feature processing module is: ； ； ； ； Where, Represents the sequence after input block, B Indicates the batch size, N p Indicates the number of blocks, L P represents the time step of each block, D Represents the embedding dimension, Block represents the improved temporal convolutional network module, Represents the embedding layer module in the improved temporal convolutional network module, Represents the sequence after input block, Indicates the intermediate transition amount The new intermediate transition amount obtained after the fluctuation feature processing module, It represents the intermediate transition amount obtained after the original input data is processed by the intermediate k improved temporal convolutional network modules.

9. The lithium battery health status prediction method according to claim 1, characterized in that: The training process of the deep learning model includes the following steps: The first-order moment estimate and the second-order moment estimate are updated based on the following formula: ； ； Where, and represents the first-order moment estimate of the current and previous iterations, and represents the second-order moment estimate of the current and previous iterations, represents the current gradient, and represents the decay rate, and t represents the current number of iterations; The bias-corrected first-order moment estimate is calculated based on the following formula: and second-order moment estimation : ； ； Where, and represents the hyperparameter corresponding to the number of iterations; Calculate the current variance ratio based on the following formula and limiting variance ratios : ； ； when When > 4, the adaptive learning rate correction parameter is calculated based on the second-order moment of deviation correction and the correction factor according to the following formula: ； ； ； Where, is the bias-corrected second moment, is the correction factor; when When ≤4, the calculation expression of the adaptive learning rate correction parameter is: ； Where, and Represent the learning rate and smoothing factor respectively; The calculation expression of the loss function is: ； Where, RMSE represents the loss function, represents the predicted value of the mth sample, represents the true value of the mth sample, Indicates the sample size.

10. A lithium battery health status prediction system, characterized in that: include: Data module, used to obtain relevant raw data sequence during the charge and discharge cycle of lithium batteries; a health feature module, configured to calculate an indirect health factor feature sequence and a direct health factor feature sequence based on the relevant raw data sequence, wherein the indirect health factor feature sequence includes a thermal feature sequence, a power feature sequence, an energy feature sequence, and a voltage feature sequence of the battery, and the direct health factor feature sequence includes a battery health feature sequence of the battery; A modal decomposition module is used to perform empirical mode decomposition on the direct health factor characteristic sequence and the indirect health factor characteristic sequence to obtain a residual component sequence with a global change trend and a fluctuation component sequence representing capacity regeneration; A training module is used to build a deep learning model for predicting the health status of batteries using multiple feature inputs. The indirect health factor feature sequence and the direct health factor feature sequence are used as a data set. Part of the data set is used as a training set, and the other part is used as a control set for training. By defining an optimizer and adjusting the loss function, a trained prediction model is obtained. The prediction module is used to input the residual component sequence and the fluctuation component sequence into the trend component processing module and the fluctuation component processing module of the prediction model respectively to obtain the health status prediction result of the battery.

Images