Lithium ion battery health state estimation method based on TSLANet network

Abstract

The invention provides a lithium ion battery health state estimation method based on a TSLANet network, and belongs to the technical field of power batteries, and the method comprises the steps: obtaining multi-dimensional time sequence data of a lithium ion battery in an actual operation process, and carrying out the preprocessing of the multi-dimensional time sequence data; constructing a battery health state estimation model by using the preprocessed multi-dimensional time sequence data based on a TSLANet network and in combination with a neural network; inputting the operation condition of the current lithium ion battery system into a battery health state estimation model, and estimating the health state of the current lithium ion battery; the method is based on the TSLANet network, combines the Kalman filtering algorithm, the IC curve, the health factor and the neural network, constructs the battery health state estimation model to estimate the health state of the lithium ion battery, correctly reflects the long-term change trend of the battery performance, is excellent in noise data processing performance, and reduces the calculation complexity and the resource consumption.

Description

Technical Field

[0001] The present invention belongs to the technical field of power batteries, and in particular relates to a method for estimating the health status of lithium-ion batteries based on a TSLANet network. Background Art

[0002] As the number of charge and discharge cycles increases, the active substances in lithium-ion batteries will be consumed, leading to unstable chemical reactions, which in turn cause irreversible changes in the internal characteristics of the battery. This change may cause safety issues such as battery fires or even explosions. The state of health (SOH) of the power battery of an electric vehicle is an important parameter for evaluating its operating status, and is also a key state variable for related control of the power battery in the battery management system (BMS). SOH reflects the aging state of the battery throughout its life cycle. Therefore, accurately estimating the SOH of a lithium battery is important for measuring its performance degradation.

[0003] SOH is usually expressed as the ratio of the current available capacity of the battery to the initial capacity, which is used to describe the degree of battery degradation under different cycles. Its accurate estimation helps in battery health diagnosis and timely replacement of deteriorated batteries.

[0004] Among the existing SOH estimation methods, the estimation method based on data model can achieve higher accuracy. Commonly used data-driven methods include neural networks, Gaussian process regression, support vector machines and relevance vector machines. Among them, neural networks are widely used because of their strong nonlinear fitting and model generalization capabilities. The Transformer model has shown great potential in tasks such as lithium-ion battery state estimation due to its excellent long-distance dependency modeling capabilities and efficient parallel computing capabilities. However, its high computational cost, inefficiency in processing long sequence data, and sensitivity to noise limit its effectiveness in practical applications. In order to overcome these limitations, this patent proposes a new type of time series lightweight adaptive network TSLANet network. On the basis of inheriting the advantages of the Transformer model, this network effectively reduces the computational complexity, improves data processing efficiency, and significantly enhances the robustness of the model to noisy time series data through innovative means such as optimizing the network structure, introducing adaptive spectrum analysis technology, and designing interactive convolution feature extraction modules, thereby achieving a more accurate SOH estimation effect.

[0005] After research, the above prior art has the following disadvantages:

[0006] 1. Many traditional SOH estimation methods often have difficulty effectively capturing the long-term trend of battery performance degradation over time when processing time series data. This is because they are usually based on empirical formulas or simple statistical models. When processing time series data, these methods may ignore key information in the battery operation history, such as changes in battery usage patterns, the number of charge and discharge cycles, and the impact of ambient temperature. The lack of this information may lead to inaccurate estimates of the battery health status and fail to reflect the long-term trend of battery performance changes.

[0007] 2. In practical applications, the operating data of the battery often contains various noises, which may come from measurement errors, ambient temperature fluctuations, changes in battery usage conditions, etc. These noisy data may have a significant impact on the SOH estimation results, resulting in instability and inaccuracy of the estimation results. Some traditional SOH estimation methods, such as those based on fixed thresholds or simple filtering, may perform poorly when processing noisy data. They may not be able to effectively distinguish between noise and useful information, resulting in deviations in the estimation results due to interference from noise.

[0008] 3. Although some advanced SOH estimation methods can provide higher estimation accuracy, their computational complexity and resource consumption are also relatively high. These methods may require large-scale training data and powerful computing resources to support the model training and reasoning process. For example, deep learning-based methods usually require a lot of computing resources for model training and parameter optimization, and also require high computational complexity during the reasoning process. This may be difficult to implement in some resource-constrained application scenarios, such as electric vehicles, portable devices, etc. Summary of the invention

[0009] In view of the above-mentioned deficiencies in the prior art, the present invention provides a lithium-ion battery health status estimation method based on TSLANet network, which solves the problems of inaccurate battery health status estimation, large influence of noise data, high complexity and large resource consumption.

[0010] In order to achieve the above objectives, the technical solution adopted by the present invention is as follows: In the first aspect, the present invention provides a method for estimating the health status of a lithium-ion battery based on a TSLANet network, comprising the following steps:

[0011] S1. Acquire multidimensional time series data of lithium-ion batteries during actual operation, and preprocess the multidimensional time series data;

[0012] S2. Using the pre-processed multi-dimensional time series data, based on the TSLANet network and combined with the neural network, a battery health status estimation model is constructed;

[0013] S3. Input the current operating status of the lithium-ion battery system into a battery health status estimation model to estimate the current health status of the lithium-ion battery.

[0014] The beneficial effects of the present invention are as follows: the present invention proposes a lithium-ion battery health status estimation method based on the TSLANet network, based on the TSLANet network and combined with a neural network, a battery health status estimation model is constructed, and the battery health status estimation model is used to estimate the current lithium-ion battery health status, thereby improving the accuracy of the battery health status estimation, reducing the influence of noise data, reducing the complexity of calculation, and reducing the consumption of computing resources.

[0015] Furthermore, the S1 comprises the following steps:

[0016] S101, obtaining a lithium-ion battery aging data set, merging the original data in the lithium-ion battery aging data set by charging and discharging, and obtaining multidimensional time series data of the lithium-ion battery during actual operation;

[0017] S102, cleaning the multidimensional time series data in the actual operation process, wherein the cleaning process includes removing abnormal values ​​and missing values, and calculating the health status in the charging and discharging process;

[0018] S103, obtaining incremental capacity and drawing an IC curve according to the cleaned multi-dimensional time series data;

[0019] S104, processing the IC curve using a Kalman filter algorithm, and analyzing the battery aging characteristics contained in the processed IC curve;

[0020] S105, extracting the charging and discharging characteristics of the lithium-ion battery, and using the extracted charging and discharging characteristics of the lithium-ion battery and the battery aging characteristics as health factors;

[0021] S106. Use the Pearson correlation coefficient to evaluate the correlation between the health factors and the SOH values, filter out the health factors with the lowest correlation, and define the filtered health factors as the preprocessed time series data.

[0022] The beneficial effects of the above further scheme are as follows: the present invention adopts pre-processing and cleaning to obtain the battery operation data, thereby ensuring the integrity and consistency of the battery operation data; using incremental capacity to draw the IC curve, obtaining many battery aging characteristics contained in the IC curve, making the acquired time series data more comprehensive, and using the Kalman filter algorithm to process the IC curve, reducing the noise interference of the battery operation data, making the data more suitable for TSLANet network processing and analysis; using the health factor and Perason correlation coefficient, the accuracy of the time series data is improved, which can correctly reflect the long-term change trend of battery performance.

[0023] Furthermore, the S103 is specifically as follows:

[0024] According to the cleaned multi-dimensional time series data, the relationship between the capacity increment and voltage change of the lithium-ion battery during the constant current charging or discharging stage is analyzed, the incremental capacity is obtained and the IC curve is drawn.

[0025] Furthermore, the S104 is specifically as follows:

[0026] Determine the state equation and observation equation of the battery system, and use the state estimation and input of the battery system at the previous moment to predict the state of the battery system at the current moment;

[0027] The state of the battery system at the current moment is used to update the Kalman gain, the state of the battery system and the covariance matrix to obtain a processed IC curve, and the battery aging characteristics contained in the processed IC curve are analyzed.

[0028] The beneficial effect of the above further scheme is that the present invention draws the IC curve, which can more intuitively analyze the characteristics in the operating data of the lithium-ion battery, and uses the Kalman filter algorithm to process the IC curve data to improve the accuracy of the IC curve characteristics.

[0029] Furthermore, the S2 comprises the following steps:

[0030] S201, constructing a TSLANet network including an adaptive spectrum block and an interactive convolution block, and training and verifying the TSLANet network;

[0031] S202, inputting the preprocessed time series data into the trained and verified TSLANet network, and obtaining a feature vector by fusing the extracted features;

[0032] S203, obtaining a prediction result using a neural network according to the feature vector, estimating the battery health status of the prediction result, and optimizing the neural network using the estimation result;

[0033] S204: construct a battery health status estimation model based on the trained and verified TSLANet network and the optimized neural network.

[0034] The beneficial effects of the above further scheme are as follows: the present invention constructs a TSLANet network and performs training and verification, thereby improving the computing speed and accuracy of the TSLANet network for the operating data of lithium-ion batteries, and constructs a battery health status estimation model in combination with a neural network, thereby achieving accurate estimation of the battery health status, reducing the impact of noise data, reducing the complexity of calculations, and reducing the consumption of computing resources.

[0035] Furthermore, the S201 includes the following steps:

[0036] S2011, build a TSLANet network containing adaptive spectrum blocks and interactive convolution blocks;

[0037] S2012. Use battery operation data to train the TSLANet network, and use the health status during the charging and discharging process as the true value of the TSLANet network training;

[0038] S2013. Use cross-validation method to verify the TSLANet network.

[0039] The beneficial effects of the above further scheme are as follows: the present invention utilizes the health status during the charging and discharging process as the true value for TSLANet network training, thereby improving the performance of the TSLANet network, reducing the error between the model estimation value and the true value, and utilizing the cross-validation method to verify the TSLANet network, thereby improving the generalization ability and estimation accuracy of the TSLANet network.

[0040] Furthermore, the S202 includes the following steps:

[0041] S2021, input the preprocessed time series data into the trained TSLANet network;

[0042] S2022, using an adaptive spectrum block to perform Fourier analysis on the preprocessed time series data to obtain an enhanced time series;

[0043] S2023, extracting features in the enhanced time series using an interactive convolution block to obtain output features;

[0044] S2024. Perform fusion processing on the output features to obtain a feature vector.

[0045] Furthermore, the expression of the output feature is as follows:

[0046] O ICB =Conv3(A1+A2)

[0047]

[0048] Among them, Conv1(·) represents the first convolutional layer, Conv2(·) represents the second convolutional layer, Conv3(·) represents the third convolutional layer, φ represents the Gaussian nonlinear unit activation function, represents element multiplication, S′ represents the time domain signal, A1 represents the adjusted first convolutional layer feature, A2 represents the adjusted second convolutional layer feature, O ICB Represents the output features.

[0049] In a second aspect, the present invention provides a computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein when the processor executes the computer program, any or all steps of the method for estimating the health status of a lithium-ion battery based on a TSLANet network are implemented.

[0050] In a third aspect, the present invention provides a computer storage medium having a computer program stored thereon, and when the computer program is executed by a processor, any or all steps of the method for estimating the health status of a lithium-ion battery based on a TSLANet network are implemented.

[0051] The beneficial effects of the above further scheme are as follows: the present invention utilizes the trained TSLANet network, uses ASB and ICB components, enriches the TSLANet network feature extraction capability, and enables the TSLANet network to have excellent performance for processing battery time series data, reduces computational complexity and reduces the consumption of computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The figure is a flow chart of the method of the present invention.

[0053] Figure 2 The flowchart of the battery health status estimation model is constructed in the present invention.

[0054] Figure 3 This is a flow chart of the overall steps of this embodiment.

[0055] Figure 4 This is the TSLANet network structure diagram of this embodiment. DETAILED DESCRIPTION

[0056] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0057] Before describing this embodiment, the following terms are explained:

[0058] TSLANet: Time Series Lightweight Adaptive Network;

[0059] SOH: battery health status;

[0060] BMS: Battery Management System;

[0061] ASB: Adaptive Spectrum Block;

[0062] ICB: Interactive Convolutional Block;

[0063] IC: Incremental Capacity;

[0064] HF: health factor;

[0065] FFT: Fast Fourier Transform;

[0066] IFFT: Inverse Fast Fourier Transform;

[0067] GELU: Gaussian nonlinear unit.

[0068] Example 1

[0069] like Figure 1 As shown, the present invention provides a method for estimating the health status of a lithium-ion battery based on a TSLANet network, and the implementation method thereof is as follows:

[0070] S1. Obtain multidimensional time series data of lithium-ion batteries during actual operation, and preprocess the multidimensional time series data. The implementation method is as follows:

[0071] S101, obtaining a lithium-ion battery aging data set, merging the original data in the lithium-ion battery aging data set by charging and discharging, and obtaining multidimensional time series data of the lithium-ion battery during actual operation;

[0072] S102, cleaning the multidimensional time series data in the actual operation process, wherein the cleaning process includes removing abnormal values ​​and missing values, and calculating the health status in the charging and discharging process;

[0073] In this embodiment, the present invention preferentially adopts the battery aging data set of the Center for Advanced Lifecycle Engineering at the University of Maryland. The batteries used in the data set all undergo the same charging process, a standard constant current / constant voltage protocol, with a constant current rate of 0.5C until the voltage reaches 4.2V and maintains 4.2V until the charging current drops below 0.05A.

[0074] S103. Obtain incremental capacity and draw an IC curve based on the cleaned multi-dimensional time series data, specifically:

[0075] According to the cleaned multi-dimensional time series data, the relationship between the capacity increment and voltage change of the lithium-ion battery during the constant current charging or discharging stage is analyzed, the incremental capacity is obtained and the IC curve is drawn;

[0076] S104, using the Kalman filter algorithm to process the IC curve, and analyzing the battery aging characteristics contained in the processed IC curve, specifically:

[0077] Determine the state equation and observation equation of the battery system, and use the state estimation and input of the battery system at the previous moment to predict the state of the battery system at the current moment;

[0078] The state of the battery system at the current moment is used to update the Kalman gain, the state of the battery system and the covariance matrix to obtain a processed IC curve, and the battery aging characteristics contained in the processed IC curve are analyzed.

[0079] In this embodiment, the original data is first processed: according to the acquired lithium-ion battery aging data set, the charging and discharging are classified and merged, and the abnormal data is eliminated, and the health status of the charging and discharging process is finally obtained by calculation; then the incremental capacity data is extracted, and the incremental capacity data is obtained by analyzing the relationship between the capacity increment and the voltage change of the battery in the constant current charging or discharging stage. The IC curve is composed of two peaks of different heights on the image, which contains many characteristics of battery aging. Analyzing the characteristics contained in the IC curve can further reflect the aging mechanism of the battery. The extracted incremental capacity data contains a lot of noise, and direct use will affect the accuracy of SOH estimation, so the Kalman filter algorithm is used to process the IC curve data.

[0080] IC represents the battery capacity increment measured on continuous voltage steps, that is, the capacity change corresponding to the unit voltage change during the charging or discharging process of the lithium-ion battery.

[0081] In this embodiment, the specific steps of Kalman filtering are as follows: determine the dynamic model of the lithium-ion battery system, that is, the state equation and observation equation of the lithium-ion battery system; for the IC curve data, the expressions of the state equation and observation equation are as follows:

[0082] x k+1 =A k x k +B k u k +w k

[0083] y k =H k x k +v k

[0084] Among them, x k+1 =A k x k +B k u k +w k represents the state equation, y k =H k x k +v k represents the observation equation, x k+1represents the state vector of the battery system at time k+1, x k represents the state vector of the battery system at time k, A k represents the state transfer matrix, B k represents the control input matrix, u k represents input, w k represents the system process noise, y k represents the observed value of the battery system at time k, H k represents the observation matrix, v k represents the observation noise.

[0085] The state at the current moment is predicted using the state estimation and input at the previous moment; the expression of the state of the battery system at the current moment is as follows:

[0086]

[0087] in, represents the prior estimate of the battery system at time k, represents the posterior estimate of the battery system at time k-1, A k represents the state transfer matrix, B k represents the control input matrix, represents the prior error covariance matrix at time k, represents the posterior error covariance matrix at time k-1, Q k-1 represents process noise;

[0088] And update the Kalman gain, the state of the battery system and the covariance matrix, and repeat the prediction and update steps for each time point in the IC curve data until all the data are processed and the filtering of the IC curve data is completed;

[0089] The expressions for updating the Kalman gain, state and covariance matrix are as follows:

[0090]

[0091] Among them, K k represents the Kalman gain, R k represents the covariance matrix of the observation noise, I represents the identity matrix, C k represents the state observation matrix.

[0092] S105, extracting the charging and discharging characteristics of the lithium-ion battery, and using the extracted charging and discharging characteristics of the lithium-ion battery and the battery aging characteristics as health factors;

[0093] In this embodiment, the characteristics of the charging and discharging process are extracted as health factors, and the following six HF health factors are extracted:

[0094] HF1 is the charging voltage rise time, and the threshold voltage selected in this embodiment is 3.8V-4.0V;

[0095] HF2 is the constant current charging time;

[0096] HF3 is the constant voltage charging time;

[0097] HF4 is the peak value of the IC curve;

[0098] HF5 is the area of ​​the first peak of the IC curve, i.e., the amount of battery power added during this period, and the voltage threshold is 3.5V-3.75V;

[0099] HF6 is the area of ​​the second peak of the IC curve, and the voltage threshold is 3.75V-4.0V;

[0100] S106. Use the Pearson correlation coefficient to evaluate the correlation between the health factors and the SOH values, filter out the health factors with the lowest correlation, and define the filtered health factors as the preprocessed time series data.

[0101] In this embodiment, the Pearson correlation coefficient is used to evaluate the correlation between the six features and the SOH value. The closer the result is to 1, the higher the correlation is, which makes the estimated SOH value closer to the true value.

[0102] The formula for the Pearson correlation coefficient is as follows:

[0103]

[0104] Among them, X i Represents the data of a single health factor HF, Y i represents the SOH value obtained by processing the original data, represents the mean value of a single health factor HF, Represents the mean value of SOH.

[0105] S2. Using the pre-processed multi-dimensional time series data, based on the TSLANet network and combined with the neural network, a battery health status estimation model is constructed, such as Figure 2 As shown, the implementation method is as follows:

[0106] S201, constructing a TSLANet network including an adaptive spectrum block and an interactive convolution block, and training and verifying the TSLANet network, the implementation method of which is as follows:

[0107] S2011, build a TSLANet network containing adaptive spectrum blocks and interactive convolution blocks;

[0108] S2012. Use battery operation data to train the TSLANet network, and use the health status during the charging and discharging process as the true value of the TSLANet network training;

[0109] S2013, verify the TSLANet network using the cross-validation method;

[0110] S202, input the preprocessed time series data into the trained and verified TSLANet network, and obtain the feature vector by fusing the extracted features. The implementation method is as follows:

[0111] S2021, input the preprocessed time series data into the trained TSLANet network;

[0112] S2022, using an adaptive spectrum block to perform Fourier analysis on the preprocessed time series data to obtain an enhanced time series;

[0113] S2023, extracting features in the enhanced time series using an interactive convolution block to obtain output features;

[0114] S2024. Perform fusion processing on the output features to obtain a feature vector.

[0115] In this embodiment, Figure 3 As shown, this embodiment extracts features from battery test data, and obtains a training set and a validation set. The TSLANet network is trained by the training set, and the TSLANet network is validated by the validation set, and the trained TSLANet network is then used to predict SOH.

[0116] In this embodiment, the specific steps of constructing the TSLANet network are as follows: Figure 4 As shown, the TSLANet network integrates two new components, the adaptive spectrum block ASB and the interactive convolution block ICB. The two components form a single layer that can be expanded to multiple layers. The adaptive spectrum block ASB uses fast Fourier transform to convert time series data to the frequency domain, and uses adaptive thresholding to attenuate high-frequency noise and highlight related spectral features. The inverse Fourier transform is then used to reconstruct the time domain features, now with reduced noise and enhanced representation. The interactive convolution block ICB is a streamlined convolution block that uses different kernel sizes to interactively improve features and improve adaptability to temporal dynamics in time series. Together, these components form a cohesive structure that balances local and global time series feature extraction for time series analysis.

[0117] The construction of the TSLANet network embedding layer divides the input time series into several segments, maps them to another dimension and embeds the position information to form a new time series with location-related characteristics. Given an input time series S, where each input time series S∈R C×L has C channels and sequence length L. First, the input time series S is divided into a set of M segments {P1, P2, ..., P M}, where each fragment P i is a small segment of the input time series S, the dimension of the segment is determined by a predefined size P, so that each segment P i ∈R C×P . Map each fragment to another dimension P′, namely P i →P i ′∈R C ×P′ , the position embedding is added to each segment to preserve the temporal order of interruptions during segmentation. The position embedding of the i-th segment is denoted as E i , which is a vector aligned with the segment in dimension. The augmented segment is obtained by adding two inputs, namely S PEi =P i ′+E i , SPE={S PE1 ,P PE2 ,...,P PEM}. It is worth noting that position embeddings are learnable parameters, allowing the model to effectively capture temporal relationships in time series data.

[0118] In this embodiment, the formula for fast Fourier transform calculation is as follows:

[0119] F=F[S PE ]∈G C×L′

[0120] Where F[·] represents a one-dimensional FFT operation, and L′ represents the length of the transformed frequency domain sequence, which may be different from the sequence length L depending on the implementation of FFT and the nature of the time series data. Each channel of the time series is transformed independently to obtain a comprehensive frequency domain representation F, which encapsulates the spectral characteristics of the original time series on all channels and contains the original frequency domain data. High-frequency components usually represent rapid fluctuations that deviate from the underlying trend, making the frequency domain series look more random and difficult to interpret. Therefore, the adaptive spectrum block includes an adaptive local filter that allows the TSLANet network to dynamically adjust the filtering level according to the characteristics of the data set and remove these high-frequency noise components, which is conducive to processing non-stationary data whose spectrum may change over time. The included adaptive local filter adaptively sets the appropriate frequency threshold for each specific time series data.

[0121] In this embodiment, the frequency domain representation F of the time series is obtained by FFT operation. First, the power spectrum of the frequency domain representation F is calculated, which helps to identify the dominant frequency component. The power spectrum P is the square of the frequency component amplitude: P = |F| 2 , which can provide a measure of the intensity of different frequencies in the time series data. The key to effective noise reduction is to adaptively filter out high-frequency components from the power spectrum P. The present invention achieves this through a trainable threshold θ, which is adjusted according to the spectral characteristics of the data. This threshold θ is set as a learnable parameter and optimized by back propagation during the training process, especially That is, the partial derivative of the threshold θ to the loss function L enables θ to distinguish between the basic signal frequency and noise. The expression of the adaptive threshold is as follows:

[0122]

[0123] in, represents element-wise multiplication, (P>θ) is a binary mask, using which frequencies with power above a threshold θ are retained, while other frequencies are filtered out. The adaptability of the threshold θ ensures that ASB can effectively remove high frequencies while retaining key information. By adaptively selecting the frequency threshold, ASB customizes its filtering process to each specific time series dataset, thereby improving the overall effectiveness of the model in processing a variety of data scenarios.

[0124] After adaptive filtering of the frequency domain data, the adaptive spectrum block in the TSLANet network uses two sets of learnable filters; a global filter learned from the frequency domain representation F, and a local filter learned from the filtered adaptive filtering data. Let W G and W L They are learnable global filters and local filters respectively. The expression of obtaining the corresponding features using filters is as follows:

[0125]

[0126] Among them, F G represents the features filtered by the global filter, F L Represents the features filtered by the local filter.

[0127] The filtered features are then integrated to capture a comprehensive and detailed spectrum as follows:

[0128] F integrated =F G +F L

[0129] Among them, F integratedRepresents comprehensive frequency domain data.

[0130] The comprehensive frequency domain data is converted back to the time domain using IFFT to obtain a time domain signal S′. The formula of the IFFT is as follows:

[0131] S′=F -1 [F integrated ]∈R C×P′

[0132] Using IFFT ensures that the enhanced features are consistent with the original data structure of the input time series.

[0133] In this embodiment, the proposed interactive convolution block, such as Figure 4 As shown in the figure, using a two-layer convolutional structure, the design of ICB includes parallel convolutions with different kernel sizes to capture local features and longer-distance dependencies. Specifically, the first convolutional layer is designed to capture local features in the data with a smaller kernel, and the second convolutional layer is designed to identify broader and longer-distance dependencies with a larger kernel. ICB components enable the output of each layer to regulate the feature extraction of another layer. Element multiplication promotes the interaction between features extracted at different scales, thereby better modeling complex relationships.

[0134] The time series S′ obtained by converting the frequency domain data back to the time domain is used as the input of ICB. The internal process of ICB can be expressed by the following formula:

[0135]

[0136] Among them, Conv1(·) and Conv2(·) represent two one-dimensional convolutional layers, and φ represents the Gaussian nonlinear unit activation function, which enables the ICB component to better learn and represent complex time series data features.

[0137] The activated features are added and the output features are calculated through the convolution layer Conv3(·), and the formula for the convolution layer calculation is as follows:

[0138] O ICB =Conv3(A1+A2)

[0139] Among them, O ICB Represents the output features.

[0140] In this embodiment, the battery operation data in the battery aging data set of the Center for Advanced Lifecycle Engineering at the University of Maryland is used to train the TSLANet network, and the health status during the charging and discharging process is used as the true value for model training. The cross-validation method is then used to verify the generalization ability and estimation accuracy of the model, so as to achieve subsequent accurate estimation of SOH.

[0141] In this embodiment, the preprocessed time series data is input into the TSLANet network, and the preprocessed time series data is processed by the above-mentioned specific TSLANet network. The feature extraction is performed through the adaptive spectrum block ASB and the interactive convolution block ICB. The adaptive spectrum block ASB is used to capture the long-term and short-term interactions in the time series data, and the noise of the time series data is reduced by the adaptive threshold to obtain an enhanced time series; the interactive convolution block ICB is used to capture the complex time pattern in the enhanced time series through the convolution operation, and the output features are extracted. The extracted features are fused to form a feature vector containing rich time series information. The quality of the feature vector directly affects the accuracy of SOH estimation.

[0142] S203, obtaining a prediction result using a neural network according to the feature vector, estimating the battery health status of the prediction result, and optimizing the neural network using the estimation result;

[0143] S204: construct a battery health status estimation model based on the trained and verified TSLANet network and the optimized neural network.

[0144] In this embodiment, the feature vector is input into the neural network, and the neural network parameters are continuously optimized by comparing the estimated value with the true value to construct an optimized neural network. The neural network is adapted to the neural network for SOH estimation and can effectively utilize the feature vector input into the neural network to estimate the battery health state. Finally, a battery health state estimation model is constructed based on the trained and verified TSLANet network and the optimized neural network.

[0145] S3. Input the current operating status of the lithium-ion battery system into a battery health status estimation model to estimate the current health status of the lithium-ion battery.

[0146] In this embodiment, the operation status of the current lithium-ion battery system is input into the battery health state estimation model. First, in the battery health state estimation model, the operation status of the current lithium-ion battery system is cleaned and input into the TSLANet network. The adaptive spectrum block ASB in the TSLANet network captures the medium- and short-term interactions of the operation status of the current lithium-ion battery system. The noise of the operation status of the current lithium-ion battery system is reduced by an adaptive threshold to obtain an enhanced time series. Then, the interactive convolution block ICB is used to capture the complex time pattern in the enhanced time series through a convolution operation, and the output features are extracted. The extracted output features are fused to form a feature vector containing the operation status of the current lithium-ion battery system. Finally, the feature vector is input into the optimized neural network. After continuous iteration of the optimized neural network, the current lithium-ion battery health state is estimated, and the accurate estimation of the battery health state is achieved, thereby providing strong support for the safe and efficient operation of the battery system.

[0147] Example 2

[0148] Based on the same inventive concept, this embodiment also provides a computer device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor. When the processor executes the computer program, any or all steps of a method for estimating the health status of a lithium-ion battery based on a TSLANet network described in Example 1 are implemented.

[0149] Example 3

[0150] Based on the same inventive concept, this embodiment also provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, any or all steps of a lithium-ion battery health status estimation method based on a TSLANet network described in Example 1 are implemented.

[0151] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0152] The present application is described with reference to the flowcharts and / or block diagrams of the methods, systems, and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

Claims

1. A method for estimating the health status of a lithium-ion battery based on a TSLANet network, characterized in that: The following steps are involved: S1. Acquire multidimensional time series data of lithium-ion batteries during actual operation, and preprocess the multidimensional time series data; S2. Using the pre-processed multi-dimensional time series data, based on the TSLANet network and combined with the neural network, a battery health status estimation model is constructed; S3. Input the current operating status of the lithium-ion battery system into a battery health status estimation model to estimate the current health status of the lithium-ion battery.

2. The method for estimating the health status of a lithium-ion battery based on a TSLANet network according to claim 1, characterized in that: The S1 comprises the following steps: S101, obtaining a lithium-ion battery aging data set, merging the original data in the lithium-ion battery aging data set by charging and discharging, and obtaining multidimensional time series data of the lithium-ion battery during actual operation; S102, cleaning the multidimensional time series data in the actual operation process, wherein the cleaning process includes removing abnormal values ​​and missing values, and calculating the health status in the charging and discharging process; S103, obtaining incremental capacity and drawing an IC curve according to the cleaned multi-dimensional time series data; S104, processing the IC curve using a Kalman filter algorithm, and analyzing the battery aging characteristics contained in the processed IC curve; S105, extracting the charging and discharging characteristics of the lithium-ion battery, and using the extracted charging and discharging characteristics of the lithium-ion battery and the battery aging characteristics as health factors; S106. Use the Pearson correlation coefficient to evaluate the correlation between the health factors and the SOH values, filter out the health factors with the lowest correlation, and define the filtered health factors as the preprocessed time series data.

3. The method for estimating the health status of a lithium-ion battery based on a TSLANet network according to claim 2, characterized in that: The S103 is specifically: According to the cleaned multi-dimensional time series data, the relationship between the capacity increment and voltage change of the lithium-ion battery during the constant current charging or discharging stage is analyzed, the incremental capacity is obtained and the IC curve is drawn.

4. The method for estimating the health status of a lithium-ion battery based on a TSLANet network according to claim 2, characterized in that: The S104 is specifically as follows: Determine the state equation and observation equation of the battery system, and use the state estimation and input of the battery system at the previous moment to predict the state of the battery system at the current moment; The state of the battery system at the current moment is used to update the Kalman gain, the state of the battery system and the covariance matrix to obtain a processed IC curve, and the battery aging characteristics contained in the processed IC curve are analyzed.

5. The method for estimating the health status of a lithium-ion battery based on a TSLANet network according to claim 1, characterized in that: The S2 comprises the following steps: S201, constructing a TSLANet network including an adaptive spectrum block and an interactive convolution block, and training and verifying the TSLANet network; S202, inputting the preprocessed time series data into the trained and verified TSLANet network, and obtaining a feature vector by fusing the extracted features; S203, obtaining a prediction result using a neural network according to the feature vector, estimating the battery health status of the prediction result, and optimizing the neural network using the estimation result; S204: construct a battery health status estimation model based on the trained and verified TSLANet network and the optimized neural network.

6. The method for estimating the health status of a lithium-ion battery based on a TSLANet network according to claim 2, characterized in that: The S201 includes the following steps: S2011, build a TSLANet network containing adaptive spectrum blocks and interactive convolution blocks; S2012. Use battery operation data to train the TSLANet network, and use the health status during the charging and discharging process as the true value of the TSLANet network training; S2013. Use cross-validation method to verify the TSLANet network.

7. The method for estimating the health status of a lithium-ion battery based on a TSLANet network according to claim 1, characterized in that: The S202 comprises the following steps: S2021, input the preprocessed time series data into the trained TSLANet network; S2022, using an adaptive spectrum block to perform Fourier analysis on the preprocessed time series data to obtain an enhanced time series; S2023, extracting features in the enhanced time series using an interactive convolution block to obtain output features; S2024. Perform fusion processing on the output features to obtain a feature vector.

8. The method for estimating the health status of a lithium-ion battery based on a TSLANet network according to claim 7, characterized in that: The expression of the output feature is as follows: <h2 style=";text-align:left;direction:ltr">O<h2 style=";text-align:left;direction:ltr"> ICB <h2 style=";text-align:left;direction:ltr"> =Conv3(A1+A2) Among them, Conv1(·) represents the first convolutional layer, Conv2(·) represents the second convolutional layer, Conv3(·) represents the third convolutional layer, φ represents the Gaussian nonlinear unit activation function, represents element multiplication, S′ represents the time domain signal, A1 represents the adjusted first convolutional layer feature, A2 represents the adjusted second convolutional layer feature, O ICB Represents the output features.

9. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, any or all steps of a method for estimating the health status of a lithium-ion battery based on a TSLANet network as described in any one of claims 1 to 8 are implemented.

10. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and when the computer program is executed by the processor, it implements any or all steps of a method for estimating the health status of a lithium-ion battery based on a TSLANet network as described in any one of claims 1 to 8.

Images