A method for predicting SOH of lithium-ion batteries

Abstract

The present invention provides a SOH prediction method for a lithium-ion battery. The method selects constant current charging phase data in an original data set to extract multi-physical field features, and constructs a multi-physical field feature data set, specifically including thermal features obtained by the average temperature of the battery in different periodic constant current charging stages, time features obtained by the length of the battery voltage time domain trajectory in different periodic constant current charging stages, and electrical features obtained by the battery incremental energy in different periodic constant current charging stages. The problem of relatively single extracted features is solved, and the aging process of the lithium-ion battery can be better mapped. In addition, the present invention constructs a prediction model based on an i-transformer algorithm, which can take into account the cross-sectional information of all features in a single time step and the overall information of a single feature in all time steps, thereby effectively capturing the long-term dependency between features and improving prediction accuracy.

Description

Technical Field

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

[0002] In recent years, lithium-ion batteries have been widely used in various fields as high-performance rechargeable power sources, including mobile devices, electric vehicles, and energy storage applications. Compared with other battery types, lithium-ion batteries have excellent characteristics, especially high energy density and long cycle life. However, as the number of charge and discharge cycles increases, the service life of lithium-ion batteries will decay, affecting the reliability and safety of the system. Therefore, it is crucial to accurately predict the state of health (SOH) of lithium-ion batteries.

[0003] At present, the prediction methods of lithium-ion battery SOH are mainly divided into two different methods: model-based methods and data-driven methods. Model-based methods include constructing equivalent circuit models and electrochemical models to simulate the degradation behavior of batteries. Constructing an equivalent circuit model is to regard the lithium-ion battery as an equivalent circuit, including devices such as resistors and capacitors, and to predict SOH by identifying device parameters and integrating algorithms; the electrochemical model is based on the chemical reactions and transmission processes occurring inside the lithium-ion battery to establish partial differential equations to describe the battery characteristics. However, as the battery ages, the error in parameter identification continues to increase, and the prediction performance is significantly affected by the model structure. It is impossible to accurately describe the degradation mechanism inside the battery, and the calculation is complex and difficult to apply in practice. The data-driven method trains the model through a large amount of data, so that the model can study the mapping rules between external battery parameters and SOH. This method does not require in-depth research on the internal mechanism of the battery, and the implementation difficulty is relatively low. It is the current research and development direction.

[0004] However, due to the uneven quality of battery operation data, the traditional data-driven method extracts relatively single features and cannot map the aging process of lithium-ion batteries well. In addition, traditional data-driven models usually use cross-sectional information of a single feature at a single time step to model, which cannot capture the long-term dependency between features well, affecting the accuracy of prediction. Summary of the invention

[0005] The purpose of the present invention is to provide a method for predicting the SOH of a lithium-ion battery, so as to solve the problem that the features extracted in the prior art are relatively single, which affects the accuracy of the prediction.

[0006] A method for predicting SOH of a lithium-ion battery, comprising:

[0007] Step S1, obtaining battery data of a charge and discharge cycle of a training lithium-ion battery to construct an original data set, wherein the original data set includes voltage, current, temperature and time of the training lithium-ion battery in the charge and discharge cycle;

[0008] Step S2, calculating the SOH of the training lithium-ion battery, selecting constant current charging phase data in the original data set to extract multi-physical field features, and constructing a multi-physical field feature data set, wherein the multi-physical field feature data set includes thermal features obtained by the average temperature of the battery in different periodic constant current charging stages, time features obtained by the time domain trajectory length of the battery voltage in different periodic constant current charging stages, and electrical features obtained by the incremental energy of the battery in different periodic constant current charging stages;

[0009] Step S3, constructing a prediction model based on the i-transformer algorithm, taking the multi-physical field feature data set as the input value and the SOH of the training lithium-ion battery as the output value, training the prediction model based on the i-transformer algorithm, defining an optimizer, and adjusting the loss function, thereby obtaining a trained prediction model;

[0010] Step S4: output the SOH of the lithium-ion battery to be predicted through the trained prediction model.

[0011] According to the SOH prediction method of lithium-ion batteries provided by the present invention, constant current charging phase data is selected in the original data set to extract multi-physical field features, and a multi-physical field feature data set is constructed, specifically including thermal features obtained by the average temperature of the battery in different periodic constant current charging stages, time features obtained by the length of the battery voltage time domain trajectory in different periodic constant current charging stages, and electrical features obtained by the battery incremental energy in different periodic constant current charging stages. This solves the problem of relatively single extracted features and can better map the aging process of lithium-ion batteries. In addition, the present invention constructs a prediction model based on the i-transformer algorithm, which can take into account the cross-sectional information of all features in a single time step and the overall information of a single feature in all time steps, thereby effectively capturing the long-term dependencies between features and improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic flow chart of the SOH prediction method for a lithium-ion battery provided by the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] See also Figure 1An embodiment of the present invention provides a method for predicting SOH of a lithium-ion battery, comprising steps S1 to S4.

[0015] Step S1, obtaining battery data of a charge and discharge cycle of a training lithium-ion battery to construct an original data set, wherein the original data set includes voltage, current, temperature and time of the training lithium-ion battery in the charge and discharge cycle.

[0016] In this embodiment, a battery testing system is used to perform a battery cycle aging experiment on cylindrical lithium-ion batteries of the same specifications at room temperature. The battery testing system includes a remote server, a Neware BTS 4000, an intermediate machine and an auxiliary channel. BTS 4000 is used to test training lithium-ion batteries, and the intermediate machine is used to connect the remote server and BTS 4000. The battery temperature is measured by the thermocouple sensor of the auxiliary channel, and the battery is measured by the Hall sensor inside the BTS4000. The rated capacity of the training lithium-ion battery is 2.5Ah. Constant current charging is performed at rates of 0.1, 0.2, 0.3, 0.5C, and 1C respectively until the battery terminal voltage reaches the upper cut-off voltage. Subsequently, the voltage is kept constant and constant voltage charging is performed until the current drops to the cut-off current. After a period of rest, 0.5C constant current discharge is performed until the battery is exhausted.

[0017] Step S2, calculate the SOH of the training lithium-ion battery, select constant current charging phase data in the original data set to extract multi-physical field features, and construct a multi-physical field feature data set, wherein the multi-physical field feature data set includes thermal features obtained by the average temperature of the battery in different periodic constant current charging stages, time features obtained by the length of the battery voltage time domain trajectory in different periodic constant current charging stages, and electrical features obtained by the battery incremental energy in different periodic constant current charging stages.

[0018] Among them, the battery's SOH is an important indicator to measure the current health of the battery. It is used to describe the battery's performance and remaining service life. It is an important indicator to evaluate the degree of battery aging or decline. It reflects the health of the battery and is usually expressed as a percentage. When a new battery leaves the factory, it is 100%. When defined by capacity decay, SOH is the ratio of the current remaining capacity of the battery to the rated capacity. Specifically,

[0019] The SOH of the training lithium-ion battery is calculated using the following formula:

[0020] ;

[0021] in, Indicates the SOH of the training lithium-ion battery, C actual is the actual current capacity of the training lithium-ion battery, C ratedis the rated capacity of the lithium-ion battery for training;

[0022] During the charge and discharge cycle of the training lithium-ion battery, the battery temperature is measured by the thermocouple sensor in the battery test system, and the average battery temperature at different constant current charging stages is extracted. As thermal characteristics, ,in T s is the total number of time steps, M 1. M 2. are the first time step, the second time step, and the T s Temperature per time step;

[0023] Draw the voltage time domain trajectory curve of the training lithium-ion battery in the constant current charging stage, and extract the battery voltage time domain trajectory length in the constant current charging stage of different cycles As a time feature, ,in, t 1 and t 2 respectively represent the start time and end time of the battery voltage in the time domain trajectory, Indicates the rate at which the battery voltage changes. x Indicates the number of charge and discharge cycles, dx for x The integral of

[0024] By calculating the incremental energy of the training lithium-ion battery in the constant current charging stage, the incremental energy change curve of each charge and discharge cycle is obtained, and the battery incremental energy in the constant current charging stage of different cycles is extracted. As electrical characteristics, ,in, E start is the initial energy level of the battery during the constant current charging phase, E end The final energy level of the battery during the constant current charging phase;

[0025] In specific implementation, the data outliers of thermal, time and electrical characteristics can be processed and normalized first, and then the multi-physics field feature data set can be constructed. , ,in, i The value of is 1, 2 or 3. i =1, X 1 indicates thermal characteristics, when i =2, X 2 represents the time feature. i =3, X 3 represents electrical characteristics, , , Respectively represent the first time step, the second time step, and the T s The eigenvalues ​​of the time steps.

[0026] Step S3, constructing a prediction model based on the i-transformer algorithm, taking the multi-physical field feature data set as the input value, the SOH of the training lithium-ion battery as the output value, training the prediction model based on the i-transformer algorithm, defining the optimizer, and adjusting the loss function, thereby obtaining a trained prediction model.

[0027] Among them, the prediction model constructed based on the i-transformer algorithm includes a full range of embedding modules based on variable tokens, an i-transformer module integrating multi-head attention mechanism and feedforward network, and an output module based on multi-layer perceptron.

[0028] The full range of variable token-based embedded modules satisfies the following formula:

[0029] ;

[0030] in, is a variable token containing the variable information features of each time step, Embedding Represents the embedding operation. The i-transformer uniquely embeds the variables at each time step into separate variable tokens In , priority is given to identifying correlations between variables rather than just processing aggregate data at a single time step.

[0031] The i-transformer module integrating the multi-head attention mechanism and the feedforward network includes L blocks, each block consists of a multi-head self-attention module, a layer normalization module and a feedforward network. The i-transformer module integrating the multi-head attention mechanism and the feedforward network satisfies the following formula:

[0032] ;

[0033] in, For the The sequence of layers to be enhanced R , , r 1. r 2. r 3 are i Take 1, 2, 3 , is the output feature enhancement sequence, iTransformer Block Indicates feature enhancement processing;

[0034] The multi-head self-attention module satisfies the following formula:

[0035] ;

[0036] in, b h Indicates h The attention output of the head, softmax is the normalization function, q h For the h The query matrix of each head, k h For the h The key matrix of each head, T represents transpose, d k is the projection dimension, v h For the h Multi-head attention is an extension of the self-attention mechanism, which uses multiple sets of weight matrices and Multiply them together to generate different query matrices, key matrices, and value matrices for each group of sequences.

[0037] Concatenate the attention outputs of all heads to form a multi-head attention vector, expressed as:

[0038] ;

[0039] in, is the multi-head attention vector, Indicates splicing, H Indicates the number of attention heads;

[0040] The convergence and stability of the deep learning neural network are enhanced by the layer normalization module, which solves the problem of over-smoothing of time series data caused by normalizing multivariate data at the same time step. The layer normalization module satisfies the following formula:

[0041] ;

[0042] in, is the normalized feature enhancement sequence, Mean It means taking the mean value, Var Indicates taking the variance;

[0043] The feedforward network (FFN) is used as a unified unit for token representation encoding. i-transformer applies FFN to the sequence representation of each variable token, encodes the observed time series through stacked inverted blocks, and represents future sequences through dense nonlinear connection decoding. The feedforward network satisfies the following formula:

[0044] ;

[0045] in, is the output of the feedforward network, is the activation function, Dropout is the regularization operation, represents the probability, W 1 and W 2 is the weight matrix, d 1 and d 2 is the bias vector. Through the regularization operation, the probability The input elements are set to zero and the outputs of some neurons are randomly discarded during the training process, so that each layer of the neural network has a different structure in each training iteration, preventing the neural network from being overly dependent on certain specific neurons, thereby improving the robustness and generalization ability of the network.

[0046] The output module based on the multi-layer perceptron satisfies the following formula:

[0047] ;

[0048] in, output is the final output of the prediction model. MLP It is a multi-layer perceptron.

[0049] In this embodiment, when training the prediction model based on the i-transformer algorithm, the optimizer uses Adam, the learning rate is set to 0.0001, and the training process satisfies the following formula:

[0050] ;

[0051] ;

[0052] in, m t and m t-1 are the first-order moment estimates of the current and previous iterations, respectively, v t and v t-1 are the second-order moment estimates of the current and previous iterations, respectively, is the current gradient, and is the decay rate. The learning rate can be set to achieve the optimal value based on the prediction effect.

[0053] In this embodiment, the loss function of the prediction model based on the i-transformer algorithm is expressed as:

[0054] ;

[0055] in, MAE represents the loss function, It is j The true value of the samples, It is j The predicted value of samples, J is the sample size.

[0056] The prediction model based on the i-transformer algorithm encapsulates the entire time series from each feature set into variable tokens to provide a global view, allowing the variable tokens to capture the complex patterns and dependencies of multi-physical field features over time; improves the model's ability to model the temporal continuity of features, improves the model's ability to capture long-term dependencies between features, the efficiency of feature utilization, and the coherence of the overall prediction, thereby improving the prediction effect of the SOH prediction model.

[0057] Step S4: output the SOH of the lithium-ion battery to be predicted through the trained prediction model.

[0058] In summary, according to the SOH prediction method of lithium-ion batteries provided by the present invention, constant current charging phase data is selected in the original data set to extract multi-physical field features, and a multi-physical field feature data set is constructed, specifically including thermal features obtained by the average temperature of the battery in different periodic constant current charging stages, time features obtained by the length of the battery voltage time domain trajectory in different periodic constant current charging stages, and electrical features obtained by the battery incremental energy in different periodic constant current charging stages. This solves the problem of relatively single extracted features and can better map the aging process of lithium-ion batteries. In addition, the present invention constructs a prediction model based on the i-transformer algorithm, which can take into account the cross-sectional information of all features in a single time step and the overall information of a single feature in all time steps, thereby effectively capturing the long-term dependencies between features and improving prediction accuracy.

[0059] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0060] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for predicting SOH of a lithium-ion battery, characterized in that: include: Step S1, obtaining battery data of a charge and discharge cycle of a training lithium-ion battery to construct an original data set, wherein the original data set includes voltage, current, temperature and time of the training lithium-ion battery in the charge and discharge cycle; Step S2, calculating the SOH of the training lithium-ion battery, selecting constant current charging phase data in the original data set to extract multi-physical field features, and constructing a multi-physical field feature data set, wherein the multi-physical field feature data set includes thermal features obtained by the average temperature of the battery in different periodic constant current charging stages, time features obtained by the time domain trajectory length of the battery voltage in different periodic constant current charging stages, and electrical features obtained by the incremental energy of the battery in different periodic constant current charging stages; Step S3, constructing a prediction model based on the i-transformer algorithm, taking the multi-physical field feature data set as the input value and the SOH of the training lithium-ion battery as the output value, training the prediction model based on the i-transformer algorithm, defining an optimizer, and adjusting the loss function, thereby obtaining a trained prediction model; Step S4, outputting the SOH of the lithium-ion battery to be predicted through the trained prediction model; Step S2 specifically includes: The SOH of the training lithium-ion battery is calculated using the following formula: ； in, Indicates the SOH of the training lithium-ion battery, C actual is the actual current capacity of the training lithium-ion battery, C rated is the rated capacity of the lithium-ion battery for training; During the charge and discharge cycle of the training lithium-ion battery, the battery temperature is measured by the thermocouple sensor in the battery test system, and the average battery temperature at different constant current charging stages is extracted. As thermal characteristics, ,in T s is the total number of time steps, M 1. M 2. are the first time step, the second time step, and the T s Temperature per time step; Draw the voltage time domain trajectory curve of the training lithium-ion battery in the constant current charging stage, and extract the battery voltage time domain trajectory length in the constant current charging stage of different cycles As a time feature, ,in, t 1 and t 2 respectively represent the start time and end time of the battery voltage in the time domain trajectory, Indicates the rate at which the battery voltage changes. x Indicates the number of charge and discharge cycles, dx for x The integral of By calculating the incremental energy of the training lithium-ion battery in the constant current charging stage, the incremental energy change curve of each charge and discharge cycle is obtained, and the battery incremental energy in the constant current charging stage of different cycles is extracted. As electrical characteristics, ,in, E start is the initial energy level of the battery during the constant current charging phase, E end The final energy level of the battery during the constant current charging phase; Building a Multiphysics Characterization Dataset , ,in, i The value of is 1, 2 or 3. i =1, X 1 indicates thermal characteristics, when i =2, X 2 represents the time feature, when i =3, X 3 represents electrical characteristics, , , Respectively represent the first time step, the second time step, and the T s The eigenvalues ​​of the time steps.

2. The SOH prediction method of a lithium-ion battery according to claim 1, characterized in that: In step S3, the prediction model based on the i-transformer algorithm is constructed, including a full range of embedding modules based on variable tokens, an i-transformer module integrating a multi-head attention mechanism and a feedforward network, and an output module based on a multi-layer perceptron; The full range of variable token-based embedded modules satisfies the following formula: ； in, is a variable token containing the variable information features of each time step, Embedding Indicates an embedding operation; The i-transformer module integrating the multi-head attention mechanism and the feedforward network includes L blocks, each block consists of a multi-head self-attention module, a layer normalization module and a feedforward network. The i-transformer module integrating the multi-head attention mechanism and the feedforward network satisfies the following formula: ； in, For the The sequence of layers to be enhanced R , , r 1. r 2. r 3 are i Take 1, 2, 3 , is the output feature enhancement sequence, iTransformer Block Indicates feature enhancement processing; The multi-head self-attention module satisfies the following formula: ； in, b h Indicates h The attention output of the head, softmax is the normalization function, q h For the h The query matrix of each head, k h For the h The key matrix of each head, T represents transpose, d k is the projection dimension, v h For the h The value matrix of the heads; Concatenate the attention outputs of all heads to form a multi-head attention vector, expressed as: ； in, is the multi-head attention vector, Indicates splicing, H Indicates the number of attention heads; The layer normalization module satisfies the following formula: ； in, is the normalized feature enhancement sequence, Mean It means taking the mean value, Var Indicates taking the variance; The feedforward network satisfies the following equation: ； in, is the output of the feedforward network, is the activation function, Dropout is the regularization operation, represents the probability, W 1 and W 2 is the weight matrix, d 1 and d 2 is the bias vector; The output module based on the multi-layer perceptron satisfies the following formula: ； in, output is the final output of the prediction model. MLP It is a multi-layer perceptron.

3. The SOH prediction method of a lithium-ion battery according to claim 2, characterized in that: In step S3, when training the prediction model based on the i-transformer algorithm, the optimizer uses Adam, the learning rate is set to 0.0001, and the training process satisfies the following formula: ； ； in, m t and m t-1 are the first-order moment estimates of the current and previous iterations, respectively, v t and v t-1 are the second-order moment estimates of the current and previous iterations, respectively, is the current gradient, and is the decay rate.

4. The SOH prediction method of a lithium-ion battery according to claim 3, characterized in that: The loss function of the prediction model based on the i-transformer algorithm is expressed as: ； in, MAE represents the loss function, It is j The true value of the samples, It is j The predicted value of samples, J is the sample size.

Images