A method for predicting the remaining life of a lithium battery

Abstract

The application discloses a kind of lithium battery's residual life prediction method, belong to battery life prediction technical field.The application first divides lithium battery charging process into constant-current section, transition section and constant-voltage section, collects voltage and current signal in each section, obtains equivalent transient impedance spectrum by Fourier transform, and calculates interface aging factor based on charge transfer resistance, ohmic resistance and double-layer capacitance;Then respectively extract the features in each section to form the feature vector in each section, derive to obtain the feature vector between sections, splice the three feature vectors in each section corresponding to similar data and two feature vectors between sections to form feature splicing vector;Finally, using the feature weighted distribution type BP model, the feature splicing vector of equivalent transient impedance spectrum and interface aging factor is processed, and the residual life prediction value is output.The application improves the prediction accuracy significantly by charging segmented feature extraction, inter-feature extraction and weighted model optimization.

Description

Technical Field

[0001] This invention relates to the field of battery life prediction technology, and more specifically to a method for predicting the remaining life of a lithium battery. Background Technology

[0002] As the core energy system for electric vehicles, energy storage power stations, and portable electronic devices, the predicted remaining lifespan of lithium-ion batteries directly affects the safety, reliability, and maintenance costs of these devices. With increasing usage time, battery capacity decays, power decreases, and internal material aging dynamically change; therefore, accurate prediction of battery remaining lifespan has become a key issue of concern in the industry.

[0003] Existing prediction methods lack detailed analysis of the battery's operating process during feature extraction, failing to fully capture the differences in aging information at different operating stages. Furthermore, they do not make sufficient use of key aging characterization parameters such as impedance characteristics and interface aging, making it difficult to comprehensively reflect the dynamic changes in battery aging. As a result, existing technologies cannot accurately characterize the relationship between battery aging process and remaining lifespan, ultimately leading to a large deviation between the predicted and actual remaining lifespan values, which cannot meet the stringent requirements for prediction accuracy in practical application scenarios. Summary of the Invention

[0004] In view of the above-mentioned shortcomings in the prior art, the present invention provides a method for predicting the remaining life of lithium batteries, which solves the problem of low accuracy in predicting the remaining life of lithium batteries in the prior art.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is: a method for predicting the remaining life of a lithium battery, comprising the following steps:

[0006] S1. Divide the lithium battery charging process into a constant current section, a transition section, and a constant voltage section, and collect the voltage and current signals of each section.

[0007] S2. Perform Fourier transform on the voltage and current signals in each segment to obtain the equivalent transient impedance spectrum;

[0008] S3. Calculate the interface aging factor based on the charge transfer resistance, ohmic resistance, and double-layer capacitance at each moment in each segment.

[0009] S4. In each segment, extract the intra-segment features of each equivalent transient impedance spectrum and each interface aging factor to form an intra-segment feature vector.

[0010] S5. Obtain the inter-segment feature vectors based on the feature vectors within each segment;

[0011] S6. Concatenate the feature vectors within three segments and the feature vectors between two segments that belong to the same type of data to obtain the feature concatenation vector.

[0012] S7. The feature-weighted allocation BP model is used to process the feature splicing vector of the equivalent transient impedance spectrum and the feature splicing vector of the interface aging factor to obtain the predicted value of the remaining life of the lithium battery.

[0013] Furthermore, S2 includes the following sub-steps:

[0014] S21. Perform short-time Fourier transform on the voltage and current signals in each segment to obtain the voltage spectrum and current spectrum;

[0015] S22. Obtain the impedance modulus based on the ratio of the voltage spectrum to the current spectrum;

[0016] S23. Take the square of the impedance magnitude to obtain the square of the impedance magnitude;

[0017] S24. Sum the squares of the impedance magnitudes in the low-frequency range to obtain the low-frequency impedance accumulation, and sum the squares of the impedance magnitudes in the high-frequency range to obtain the high-frequency impedance accumulation.

[0018] S25. The ratio of the low-frequency impedance accumulation to the high-frequency impedance accumulation is used as the equivalent transient impedance spectrum.

[0019] Furthermore, the formula for calculating the interface aging factor in S3 is as follows:

[0020]

[0021] Where, θ t Let be the interface aging factor at time t. Let be the charge transfer resistance at time t. The standard charge transfer resistor for a battery. Let be the ohmic resistance at time t. The standard ohmic resistor for a battery, Let be the double-layer capacitance at time t. It is a standard double-layer capacitor for batteries.

[0022] Furthermore, the characteristics within the S4 mid-segment include: mean, fluctuation coefficient, aging trend coefficient, and aging growth coefficient.

[0023] Furthermore, the equivalent transient impedance spectra in each segment are sequenced, and the interface aging factors in each segment are sequenced, with the mean being the average value of each element in the sequence.

[0024] The volatility coefficient is the variance or standard deviation of the series;

[0025] The process of obtaining the aging trend coefficient includes: multiplying each time point by the corresponding element in the sequence, and summing the results of the multiplications at each time point to obtain the first term; summing the results of the time points, summing the elements at each time point in the sequence, and multiplying the two sums to obtain the second term; multiplying the first term by the total number of time points, and then subtracting the second term to obtain the first numerator; squaring each time point and summing the results to obtain the third term; summing the results of each time point and squaring the results to obtain the fourth term; multiplying the third term by the total number of time points, and then subtracting the fourth term to obtain the first denominator; and using the ratio of the first numerator to the first denominator as the aging trend coefficient.

[0026] The process of obtaining the aging growth coefficient includes: taking the natural logarithm of all elements in the sequence and calculating the average of all logarithmic results; simultaneously calculating the average of all time points in the sequence, subtracting the average of the logarithmic results from the natural logarithm of each element to obtain the first residual term; subtracting the average of all time points from each time point to obtain the second residual term; multiplying the first and second residual terms and adding the multiplication results at each time point to obtain the second numerator term; squaring the second residual term and summing them to obtain the second denominator term; and using the ratio of the second numerator term to the second denominator term as the aging growth coefficient.

[0027] Furthermore, S5 includes the following sub-steps:

[0028] S51. Subtract the intra-segment eigenvector of the transition segment from the intra-segment eigenvector of the constant current segment to obtain the difference vector between the transition segment and the constant current segment.

[0029] S52. Divide the difference vector between the transition segment and the constant current segment by the intra-segment feature vector of the constant current segment to obtain the inter-segment feature vector between the transition segment and the constant current segment.

[0030] S53. Subtract the intra-segment feature vector of the constant pressure segment from the intra-segment feature vector of the transition segment to obtain the difference vector between the constant pressure segment and the transition segment.

[0031] S54. Divide the difference vector between the constant pressure segment and the transition segment by the intra-segment feature vector of the transition segment to obtain the inter-segment feature vector between the constant pressure segment and the transition segment.

[0032] Furthermore, the feature-weighted BP model in S7 includes: a first one-dimensional feature extraction module, a second one-dimensional feature extraction module, multiplier M1, multiplier M2, a Concat layer, a weight allocation module, and a BP neural network;

[0033] The input of the first one-dimensional feature extraction module is used to input the feature splicing vector of the equivalent transient impedance spectrum, and its output is connected to the first input of the multiplier M1 and the first input of the weight allocation module, respectively.

[0034] The input of the second one-dimensional feature extraction module is used to input the feature concatenation vector of the interface aging factor, and its output is connected to the first input of the multiplier M2 and the second input of the weight allocation module, respectively.

[0035] The first output of the weight allocation module is connected to the second input of multiplier M1, and its second output is connected to the second input of multiplier M2.

[0036] The input of the Concat layer is connected to the output of multiplier M1 and multiplier M2, respectively, and its output is connected to the input of the BP neural network.

[0037] The output of the BP neural network serves as the output of the feature-weighted allocation BP model.

[0038] Furthermore, the weight allocation module includes: a first feature mapping submodule, a second feature mapping submodule, and a softmax layer;

[0039] The input of the first feature mapping submodule serves as the first input of the weight allocation module;

[0040] The input of the second feature mapping submodule serves as the second input of the weight allocation module;

[0041] The input of the softmax layer is connected to the output of the first feature mapping submodule and the output of the second feature mapping submodule, respectively. Its first output serves as the first output of the weight allocation module, and its second output serves as the second output of the weight allocation module.

[0042] Furthermore, the first one-dimensional feature extraction module and the second one-dimensional feature extraction module have the same structure, both including the following sequentially connected components: a first convolutional layer, a pooling layer, and a second convolutional layer.

[0043] The first and second feature mapping submodules have the same structure, both including a fully connected layer, a ReLU layer, and a Sigmoid layer connected in sequence.

[0044] Furthermore, the expression for the softmax layer is:

[0045]

[0046]

[0047] in, The i-th weight is output from the first output of the softmax layer. The i-th weight is output from the second output of the softmax layer. The first feature mapping submodule outputs the i-th mapped feature. The second feature mapping submodule outputs the i-th mapped feature. for The weight, for The weight, It is a natural constant.

[0048] The beneficial effects of this invention are as follows:

[0049] 1. This invention divides the lithium battery charging process into a constant current stage, a transition stage, and a constant voltage stage, and collects and analyzes the voltage and current signals of each stage. This effectively captures the differences in aging information of the battery at different working stages, and can more accurately reflect the dynamic characteristics of the battery under different charging states, thereby improving the accuracy of remaining life prediction.

[0050] 2. By performing Fourier transform on each voltage and current signal segment, this invention obtains the equivalent transient impedance spectrum. This invention can comprehensively characterize the transient electrochemical characteristics of the battery and effectively depict the complex and nonlinear coupling process of aging mechanism, thereby improving the problem that existing technologies are unable to accurately capture the aging law of the battery.

[0051] 3. This invention further extracts charge transfer resistance, ohmic resistance and double-layer capacitance to calculate the interface aging factor, enabling systematic analysis of key aging characterization parameters and effective utilization of the degradation of the battery's internal interface, thereby improving the sensitivity and reliability of the prediction model for remaining lifetime.

[0052] 4. By constructing and splicing intra-segment feature vectors and inter-segment feature vectors, this invention can achieve the fusion analysis of features at each stage of the battery charging process, comprehensively consider local and overall aging information, and enable the prediction model to fully capture the correlation between different features, thus solving the prediction bias problem caused by insufficient feature fusion in existing methods.

[0053] 5. This invention captures the impedance characteristics and key aging characteristics of the interface aging under different working stages, and constructs inter-segment feature vectors to reflect the dynamic changes in battery aging. It fully explores the aging characteristics and captures the differences in aging information at different working stages. By combining the feature splicing vector of the equivalent transient impedance spectrum and the feature splicing vector of the interface aging factor, it accurately depicts the correlation between the battery aging process and the remaining life, and improves the accuracy of life prediction.

[0054] 6. This invention uses a feature-weighted allocation BP model to process feature concatenation vectors. This invention optimizes the contribution of different features to the remaining lifetime prediction through a weighting mechanism, making the model more adaptable and robust when dealing with highly nonlinear mapping relationships, thereby significantly improving the accuracy of remaining lifetime prediction. Attached Figure Description

[0055] Figure 1 A flowchart of a method for predicting the remaining life of a lithium battery;

[0056] Figure 2 This is a schematic diagram of the feature-weighted allocation BP model.

[0057] Figure 3 A schematic diagram of the weight allocation module;

[0058] Figure 4 This is a schematic diagram of the structure of the first one-dimensional feature extraction module and the second one-dimensional feature extraction module;

[0059] Figure 5 This is a schematic diagram of the structure of the first feature mapping submodule and the second feature mapping submodule. Detailed Implementation

[0060] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0061] like Figure 1 As shown, a method for predicting the remaining life of a lithium battery includes the following steps:

[0062] S1. Divide the lithium battery charging process into a constant current section, a transition section, and a constant voltage section, and collect the voltage and current signals of each section.

[0063] S2. Perform Fourier transform on the voltage and current signals in each segment to obtain the equivalent transient impedance spectrum;

[0064] S3. Calculate the interface aging factor based on the charge transfer resistance, ohmic resistance, and double-layer capacitance at each moment in each segment.

[0065] S4. In each segment, extract the intra-segment features of each equivalent transient impedance spectrum and each interface aging factor to form an intra-segment feature vector.

[0066] S5. Based on the intra-segment feature vectors of the same type of data, obtain the inter-segment feature vectors;

[0067] S6. Concatenate the feature vectors within three segments and the feature vectors between two segments that belong to the same type of data to obtain the feature concatenation vector.

[0068] S7. The feature-weighted allocation BP model is used to process the feature splicing vector of the equivalent transient impedance spectrum and the feature splicing vector of the interface aging factor to obtain the predicted value of the remaining life of the lithium battery.

[0069] In this embodiment, after charging is started, the system uses I rated Constant current charging, with real-time acquisition of voltage and current signals.

[0070] The constant current phase ends and the transition phase begins when the following two conditions are met: 1. The real-time voltage reaches 0.95U. end ~0.98U end (The value is determined according to the battery type, such as 0.97U for ternary lithium batteries) end ), 2. The current signal begins to deviate from I rated That is, the current shows a significant decrease for the first time, U end This is the charging termination voltage.

[0071] The transition phase ends and the constant voltage phase begins when the following condition is met: the real-time voltage reaches U. end As the current continues to decrease, the transition phase is determined to be over, and the system enters the constant voltage phase, continuing to collect voltage and current signals until charging is complete.

[0072] In this embodiment, S2 includes the following sub-steps:

[0073] S21. Perform short-time Fourier transform on the voltage and current signals in each segment to obtain the voltage spectrum and current spectrum;

[0074] S22. Based on the ratio of the voltage spectrum and the current spectrum in the same time period, the impedance modulus is obtained: Z(f) = |V(f) / I(f)|, where V(f) is the voltage spectrum, I(f) is the current spectrum, Z(f) is the impedance modulus, and || is the modulus operation.

[0075] S23. Take the square of the impedance magnitude to obtain the square of the impedance magnitude;

[0076] S24. Sum the squares of the impedance magnitudes in the low-frequency range to obtain the low-frequency impedance accumulation, and sum the squares of the impedance magnitudes in the high-frequency range to obtain the high-frequency impedance accumulation.

[0077] S25. The ratio of the low-frequency impedance accumulation to the high-frequency impedance accumulation is used as the equivalent transient impedance spectrum.

[0078] This invention acquires the transient characteristics of the battery in the time and frequency domains by performing short-time Fourier transforms on the voltage and current signals of each charging segment. Furthermore, by calculating the ratio of the voltage spectrum to the current spectrum, the impedance modulus is obtained, and an equivalent transient impedance spectrum is formed by comparing the cumulative squares of the low-frequency and high-frequency impedance moduli. This allows for a comprehensive quantification of the battery's dynamic electrochemical response characteristics across different frequency ranges, thereby improving the ability to characterize impedance changes during battery aging and contributing to increased accuracy in remaining battery life prediction. By using the low-frequency / high-frequency ratio, this invention compares aging-related low-frequency changes with high-frequency background signals, making aging characteristics more prominent and suitable for remaining battery life prediction.

[0079] In this embodiment, frequencies greater than 100Hz are classified as high frequency, and frequencies less than 100Hz are classified as low frequency.

[0080] In this embodiment, the formula for calculating the interface aging factor in S3 is:

[0081]

[0082] Where, θ t Let be the interface aging factor at time t. Let be the charge transfer resistance at time t. The standard charge transfer resistor for a battery. Let be the ohmic resistance at time t. The standard ohmic resistor for a battery, Let be the double-layer capacitance at time t. It is a standard double-layer capacitor for batteries.

[0083] In this embodiment, the ohmic resistance, charge transfer resistance, and double-layer capacitance at each moment in each segment can be obtained using any existing technical means (such as impedance analysis or equivalent circuit fitting). The specific method of obtaining these parameters is not limited in this invention.

[0084] This invention combines the relative changes in charge transfer resistance and ohmic resistance (ratio to standard values) with the relative changes in double-layer capacitance, integrating multi-dimensional aging information of interface reaction dynamics (charge transfer, ohmic conduction) and interface energy storage characteristics (double-layer capacitance). It can keenly capture subtle changes in interface aging and make up for the shortcomings of existing technologies in characterizing interface aging.

[0085] In this embodiment, the equivalent transient impedance spectra in each segment are formed into a sequence, and the interface aging factors in each segment are formed into a sequence. The features within the segment in S4 include: mean, fluctuation coefficient, aging trend coefficient and aging growth coefficient. The mean, fluctuation coefficient, aging trend coefficient and aging growth coefficient constitute the feature vector within the segment.

[0086] For the equivalent transient impedance spectrum, there are three segments of equivalent transient impedance spectrum data: constant current segment, transition segment, and constant voltage segment. The characteristics within each segment are obtained. For the interface aging factor, there are three segments of interface aging factor data: constant current segment, transition segment, and constant voltage segment. The characteristics within each segment are obtained.

[0087] In this embodiment, the mean is the average value of each element in the sequence;

[0088] The volatility coefficient is the variance or standard deviation of the series;

[0089] The process of obtaining the aging trend coefficient includes: multiplying each time point by the corresponding element in the sequence, and summing the results of the multiplications at each time point to obtain the first term; summing the results of the time points, summing the elements at each time point in the sequence, and multiplying the two sums to obtain the second term; multiplying the first term by the total number of time points, and then subtracting the second term to obtain the first numerator; squaring each time point and summing the results to obtain the third term; summing the results of each time point and squaring the results to obtain the fourth term; multiplying the third term by the total number of time points, and then subtracting the fourth term to obtain the first denominator; and using the ratio of the first numerator to the first denominator as the aging trend coefficient.

[0090] The process of obtaining the aging growth coefficient includes: taking the natural logarithm of all elements in the sequence and calculating the average of all logarithmic results; simultaneously calculating the average of all time points in the sequence, subtracting the average of the logarithmic results from the natural logarithm of each element to obtain the first residual term; subtracting the average of all time points from each time point to obtain the second residual term; multiplying the first and second residual terms and adding the multiplication results at each time point to obtain the second numerator term; squaring the second residual term and summing them to obtain the second denominator term; and using the ratio of the second numerator term to the second denominator term as the aging growth coefficient.

[0091] In this embodiment, the formula for the aging trend coefficient is:

[0092]

[0093] in, Let be the aging trend coefficient of the k-th sequence. Let be the element at time t in the k-th sequence. Let t be the total time of the k-th sequence, and t be the time number. When k=1, it corresponds to the constant current segment; when k=2, it corresponds to the transition segment; and when k=3, it corresponds to the constant pressure segment. Each segment corresponds to one sequence.

[0094] The formula for calculating the aging growth factor is:

[0095] , ,

[0096] in, Let ln be the aging growth coefficient of the k-th sequence, and ln be the logarithmic function. To each Take the average. The average value is taken over all time points t.

[0097] This invention directly quantifies the overall level of aging characteristics (equivalent transient impedance spectrum / interface aging factor) within a given period using the mean value, intuitively reflecting the basic aging degree of the battery at that stage. The fluctuation coefficient captures the dispersion of characteristics within this period, corresponding to the dynamic fluctuations of the aging state during battery operation, thus overcoming the deficiency of a single mean value in reflecting state stability. Furthermore, the aging trend coefficient, based on time-series correlation calculations, accurately extracts the changing trend of aging characteristics over time (e.g., continuous increase / decrease), clearly defining the direction of aging evolution. The aging growth coefficient further focuses on the relative rate of change of aging characteristics through logarithmic transformation, keenly sensing the acceleration / deceleration of the aging process.

[0098] In this embodiment, S5 includes the following sub-steps:

[0099] S51. Subtract the intra-segment eigenvector of the transition segment from the intra-segment eigenvector of the constant current segment to obtain the difference vector between the transition segment and the constant current segment.

[0100] S52. Divide the difference vector between the transition segment and the constant current segment by the intra-segment feature vector of the constant current segment to obtain the inter-segment feature vector between the transition segment and the constant current segment.

[0101] S53. Subtract the intra-segment feature vector of the constant pressure segment from the intra-segment feature vector of the transition segment to obtain the difference vector between the constant pressure segment and the transition segment.

[0102] S54. Divide the difference vector between the constant pressure segment and the transition segment by the intra-segment feature vector of the transition segment to obtain the inter-segment feature vector between the constant pressure segment and the transition segment.

[0103] This invention first quantifies the changes in aging characteristics from the constant current stage to the transition stage and from the transition stage to the constant pressure stage by subtracting the characteristics of adjacent segments, thus intuitively reflecting the degree of transition between aging states in each stage; then, by combining the relative proportion calculation of "dividing the difference by the previous segment's characteristics", the absolute change is transformed into the relative change magnitude, eliminating the interference of differences in different initial states.

[0104] In this invention, to prevent the denominator from being 0 when taking the ratio, a very small number is set in the denominator.

[0105] In this embodiment, in S6, the three intra-segment eigenvectors and two inter-segment eigenvectors belonging to the equivalent transient impedance spectrum are concatenated to obtain the eigenvector of the equivalent transient impedance spectrum; the three intra-segment eigenvectors and two inter-segment eigenvectors belonging to the interface aging factor are concatenated to obtain the eigenvector of the interface aging factor.

[0106] like Figure 2 As shown, the feature-weighted BP model in S7 includes: a first one-dimensional feature extraction module, a second one-dimensional feature extraction module, multiplier M1, multiplier M2, a Concat layer, a weight allocation module, and a BP neural network;

[0107] The input of the first one-dimensional feature extraction module is used to input the feature splicing vector of the equivalent transient impedance spectrum, and its output is connected to the first input of the multiplier M1 and the first input of the weight allocation module, respectively.

[0108] The input of the second one-dimensional feature extraction module is used to input the feature concatenation vector of the interface aging factor, and its output is connected to the first input of the multiplier M2 and the second input of the weight allocation module, respectively.

[0109] The first output of the weight allocation module is connected to the second input of multiplier M1, and its second output is connected to the second input of multiplier M2.

[0110] The input of the Concat layer is connected to the output of multiplier M1 and multiplier M2, respectively, and its output is connected to the input of the BP neural network.

[0111] The output of the BP neural network serves as the output of the feature-weighted allocation BP model.

[0112] This invention first employs first and second one-dimensional feature extraction modules to extract features from two different types of features: equivalent transient impedance spectrum and interface aging factor, respectively, to accurately mine the aging information of each type of feature. The weight allocation module can dynamically assign appropriate weights to the two types of features based on their actual contribution to the aging process. This strengthens the influence of key features (such as the characterization effect of interface aging factor on severely aged batteries) while weakening the redundant interference of secondary features, thus solving the limitation of the traditional BP model in "indiscriminately processing" different features.

[0113] Secondly, the multiplier achieves precise weighting of weights and features, and the Concat layer then merges the two types of features into a unified input, which not only preserves the independent information of the two types of features, but also realizes the complementary relationship of information. Finally, the BP neural network performs nonlinear mapping on the fused features, which can more efficiently learn the complex relationship between features and remaining lifetime.

[0114] like Figure 3As shown, the weight allocation module includes: a first feature mapping submodule, a second feature mapping submodule, and a softmax layer;

[0115] The input of the first feature mapping submodule serves as the first input of the weight allocation module;

[0116] The input of the second feature mapping submodule serves as the second input of the weight allocation module;

[0117] The input of the softmax layer is connected to the output of the first feature mapping submodule and the output of the second feature mapping submodule, respectively. Its first output serves as the first output of the weight allocation module, and its second output serves as the second output of the weight allocation module.

[0118] like Figure 4 As shown, the first one-dimensional feature extraction module and the second one-dimensional feature extraction module have the same structure, both including a first convolutional layer, a pooling layer and a second convolutional layer connected in sequence. The kernel size of the first convolutional layer is 1×1 and the kernel size of the second convolutional layer is 1×3.

[0119] like Figure 5 As shown, the first feature mapping submodule and the second feature mapping submodule have the same structure, both including a fully connected layer, a ReLU layer and a Sigmoid layer connected in sequence.

[0120] In this embodiment, the expression for the softmax layer is:

[0121]

[0122]

[0123] in, The i-th weight is output from the first output of the softmax layer. The i-th weight is output from the second output of the softmax layer. The first feature mapping submodule outputs the i-th mapped feature. The second feature mapping submodule outputs the i-th mapped feature. for The weight, for The weight, It is a natural constant.

[0124] This invention achieves a "point-by-point competition" weight allocation method by exponentially weighting and normalizing the two mapping features of the same dimension in the softmax layer. This is different from the traditional softmax method that normalizes the entire vector at once, allowing the importance of the two types of features in the same dimension to be measured and adaptively adjusted separately.

[0125] Because weight allocation is performed independently in each dimension, the model can highlight features that are more sensitive to prediction in different dimensions while suppressing features that contribute less to that dimension, thus forming a fine-grained feature selection capability. Compared with integer vector normalization, the dimension-wise softmax weight allocation mechanism of this invention can make fuller use of the differences between multi-source features in different dimensions, improve the flexibility and accuracy of feature fusion, and ultimately enhance the stability and prediction accuracy of the remaining lifetime prediction model.

[0126] This invention divides the lithium battery charging process into a constant current stage, a transition stage, and a constant voltage stage, and collects and analyzes the voltage and current signals of each stage. This effectively captures the differences in aging information of the battery at different working stages, and can more accurately reflect the dynamic characteristics of the battery under different charging states, thereby improving the accuracy of remaining life prediction.

[0127] This invention obtains the equivalent transient impedance spectrum by performing Fourier transform on each voltage and current signal segment. This invention can comprehensively characterize the transient electrochemical characteristics of the battery and effectively depict the complex and nonlinear coupling process of aging mechanism, thereby improving the problem that existing technologies are unable to accurately capture the aging law of the battery.

[0128] This invention further extracts charge transfer resistance, ohmic resistance, and double-layer capacitance to calculate the interface aging factor, enabling systematic analysis of key aging characterization parameters and effective utilization of battery internal interface degradation, thereby improving the sensitivity and reliability of the prediction model for remaining lifetime.

[0129] This invention, through the construction and splicing of intra-segment and inter-segment feature vectors, enables the fusion analysis of features at each stage of the battery charging process. By comprehensively considering local and overall aging information, the prediction model can fully capture the correlation between different features, thus solving the prediction bias problem caused by insufficient feature fusion in existing methods.

[0130] This invention captures the impedance characteristics and key aging characteristics of the interface aging at different operating stages, and constructs inter-segment feature vectors to reflect the dynamic changes in battery aging. It fully explores the aging characteristics and captures the differences in aging information at different operating stages. By combining the feature splicing vector of the equivalent transient impedance spectrum and the feature splicing vector of the interface aging factor, it accurately characterizes the correlation between the battery aging process and the remaining life, thereby improving the accuracy of life prediction.

[0131] This invention employs a feature-weighted allocation BP model to process feature concatenation vectors. By optimizing the contribution of different features to remaining lifetime prediction through a weighting mechanism, this invention enables the model to have stronger adaptability and robustness when dealing with highly nonlinear mapping relationships, thereby significantly improving the accuracy of remaining lifetime prediction.

[0132] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining life of a lithium battery, characterized in that, Includes the following steps: S1. Divide the lithium battery charging process into a constant current section, a transition section, and a constant voltage section, and collect the voltage and current signals of each section. S2. Perform Fourier transform on the voltage and current signals in each segment to obtain the equivalent transient impedance spectrum; S3. Calculate the interface aging factor based on the charge transfer resistance, ohmic resistance, and double-layer capacitance at each moment in each segment: ; Where, θ t Let be the interface aging factor at time t. Let be the charge transfer resistance at time t. The standard charge transfer resistor for a battery. Let be the ohmic resistance at time t. The standard ohmic resistor for a battery, Let be the double-layer capacitance at time t. It is a standard double-layer capacitor for batteries; S4. In each segment, extract the intra-segment features of each equivalent transient impedance spectrum and each interface aging factor to form an intra-segment feature vector. S5. Obtain the inter-segment feature vectors based on the feature vectors within each segment; S6. Concatenate the feature vectors within three segments and the feature vectors between two segments that belong to the same type of data to obtain the feature concatenation vector. S7. The feature-weighted allocation BP model is used to process the feature splicing vector of the equivalent transient impedance spectrum and the feature splicing vector of the interface aging factor to obtain the predicted value of the remaining life of the lithium battery.

2. The method for predicting the remaining life of a lithium battery according to claim 1, characterized in that, S2 includes the following steps: S21. Perform short-time Fourier transform on the voltage and current signals in each segment to obtain the voltage spectrum and current spectrum; S22. Obtain the impedance modulus based on the ratio of the voltage spectrum to the current spectrum; S23. Take the square of the impedance magnitude to obtain the square of the impedance magnitude; S24. Sum the squares of the impedance magnitudes in the low-frequency range to obtain the low-frequency impedance accumulation, and sum the squares of the impedance magnitudes in the high-frequency range to obtain the high-frequency impedance accumulation. S25. The ratio of the low-frequency impedance accumulation to the high-frequency impedance accumulation is used as the equivalent transient impedance spectrum.

3. The method for predicting the remaining life of a lithium battery according to claim 1, characterized in that, The features within the S4 segment include: mean, fluctuation coefficient, aging trend coefficient, and aging growth coefficient.

4. The method for predicting the remaining life of a lithium battery according to claim 3, characterized in that, The equivalent transient impedance spectra in each segment are sequenced, and the interface aging factors in each segment are sequenced. The mean value is the average value of each element in the sequence. The volatility coefficient is the variance or standard deviation of the series; The process of obtaining the aging trend coefficient includes: multiplying each time point by the corresponding element in the sequence, and summing the results of the multiplications at each time point to obtain the first term; summing the results of the time points, summing the elements at each time point in the sequence, and multiplying the two sums to obtain the second term; multiplying the first term by the total number of time points, and then subtracting the second term to obtain the first numerator; squaring each time point and summing the results to obtain the third term; summing the results of each time point and squaring the results to obtain the fourth term; multiplying the third term by the total number of time points, and then subtracting the fourth term to obtain the first denominator; and using the ratio of the first numerator to the first denominator as the aging trend coefficient. The process of obtaining the aging growth coefficient includes: taking the natural logarithm of all elements in the sequence and calculating the average of all logarithmic results; simultaneously calculating the average of all time points in the sequence, subtracting the average of the logarithmic results from the natural logarithm of each element to obtain the first residual term; subtracting the average of all time points from each time point to obtain the second residual term; multiplying the first and second residual terms and adding the multiplication results at each time point to obtain the second numerator term; squaring the second residual term and summing them to obtain the second denominator term; and using the ratio of the second numerator term to the second denominator term as the aging growth coefficient.

5. The method for predicting the remaining life of a lithium battery according to claim 1, characterized in that, S5 includes the following steps: S51. Subtract the intra-segment eigenvector of the transition segment from the intra-segment eigenvector of the constant current segment to obtain the difference vector between the transition segment and the constant current segment. S52. Divide the difference vector between the transition segment and the constant current segment by the intra-segment feature vector of the constant current segment to obtain the inter-segment feature vector between the transition segment and the constant current segment. S53. Subtract the intra-segment feature vector of the constant pressure segment from the intra-segment feature vector of the transition segment to obtain the difference vector between the constant pressure segment and the transition segment. S54. Divide the difference vector between the constant pressure segment and the transition segment by the intra-segment feature vector of the transition segment to obtain the inter-segment feature vector between the constant pressure segment and the transition segment.

6. The method for predicting the remaining life of a lithium battery according to claim 1, characterized in that, The feature-weighted BP model in S7 includes: a first one-dimensional feature extraction module, a second one-dimensional feature extraction module, a multiplier M1, a multiplier M2, a Concat layer, a weight allocation module, and a BP neural network; The input of the first one-dimensional feature extraction module is used to input the feature splicing vector of the equivalent transient impedance spectrum, and its output is connected to the first input of the multiplier M1 and the first input of the weight allocation module, respectively. The input of the second one-dimensional feature extraction module is used to input the feature concatenation vector of the interface aging factor, and its output is connected to the first input of the multiplier M2 and the second input of the weight allocation module, respectively. The first output of the weight allocation module is connected to the second input of multiplier M1, and its second output is connected to the second input of multiplier M2. The input of the Concat layer is connected to the output of multiplier M1 and multiplier M2, respectively, and its output is connected to the input of the BP neural network. The output of the BP neural network serves as the output of the feature-weighted allocation BP model.

7. The method for predicting the remaining life of a lithium battery according to claim 6, characterized in that, The weight allocation module includes: a first feature mapping submodule, a second feature mapping submodule, and a softmax layer; The input of the first feature mapping submodule serves as the first input of the weight allocation module; The input of the second feature mapping submodule serves as the second input of the weight allocation module; The input of the softmax layer is connected to the output of the first feature mapping submodule and the output of the second feature mapping submodule, respectively. Its first output serves as the first output of the weight allocation module, and its second output serves as the second output of the weight allocation module.

8. The method for predicting the remaining life of a lithium battery according to claim 6, characterized in that, The first one-dimensional feature extraction module and the second one-dimensional feature extraction module have the same structure, both including the following sequentially connected components: a first convolutional layer, a pooling layer, and a second convolutional layer. The first and second feature mapping submodules have the same structure, both including a fully connected layer, a ReLU layer, and a Sigmoid layer connected in sequence.

9. The method for predicting the remaining life of a lithium battery according to claim 7, characterized in that, The expression for the softmax layer is: ； ； in, The i-th weight is output from the first output of the softmax layer. The i-th weight is output from the second output of the softmax layer. The first feature mapping submodule outputs the i-th mapped feature. The second feature mapping submodule outputs the i-th mapped feature. for The weight, for The weight, It is a natural constant.

Images