A lithium-ion battery state of health evaluation modeling method based on transfer learning

Abstract

This invention provides a modeling method for assessing the health status of lithium-ion batteries based on transfer learning, comprising: determining the performance parameters of a lithium-ion battery energy storage system that exhibit changing trends throughout its lifespan; using the discharge capacity of the lithium-ion battery energy storage system as the health status parameter, dividing it into four levels and assigning corresponding labels; acquiring performance parameter data and label data from the development and testing phases of the lithium-ion battery energy storage system, and constructing a source domain dataset F. S Using source domain dataset F S The mapping relationship model between six performance parameters and state labels established using the stacked autoencoder algorithm is trained to obtain the model parameter set; the target domain dataset F is constructed. T This invention optimizes the model parameter set and verifies the accuracy of the mapping relationship model to obtain a lithium-ion battery health status assessment model. It solves the technical challenge of fusing performance information from multiple performance parameters during the use of lithium-ion battery energy storage systems, enabling the classification of their health status.

Description

Technical Field

[0001] This invention relates to a modeling method for assessing the health status of lithium-ion batteries based on transfer learning, belonging to the field of lithium-ion battery fault prediction and health management (PHM) technology. Background Technology

[0002] Lithium-ion battery energy storage systems are typical secondary energy sources, capable of multiple charge-discharge cycles, and widely used in various industrial and social sectors. During the research and development of lithium-ion batteries, to fully improve product performance, continuous research and development tests, cycle charge-discharge tests, and accelerated life tests are conducted. Product development departments accumulate a wealth of experimental and usage data, covering the battery's life cycle. However, current battery health status classification is mostly based on rough assessments using cycle counts and service life, with multi-source data not being fully mined and utilized, making it difficult to determine the health level of battery energy storage systems and thus support usage decisions. Given this multi-source data context, how to mine multi-stage and multi-type performance data of lithium-ion battery energy storage systems, establish a mapping relationship between life cycle data and status labels, and transfer the model from high-rate charge-discharge and high-stress conditions in research and development tests to actual usage conditions, enabling the assessment and classification of the health status of lithium-ion battery energy storage systems, and assisting equipment users in making usage and maintenance decisions, has become an urgent problem to be solved by lithium-ion battery energy storage system research and development and equipment operation departments. Summary of the Invention

[0003] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide a modeling method for assessing the health status of lithium-ion batteries based on transfer learning. Following the idea of ​​"model parameter transfer + fine-tuning", it fully integrates the life cycle data accumulated by lithium-ion battery energy storage systems in various types of tests, and solves the problem of accurately assessing the health status during use.

[0004] The technical solution of this invention is: a lithium-ion battery health status assessment modeling method based on transfer learning, comprising:

[0005] Determine the performance parameters of a lithium-ion battery energy storage system that exhibit changing trends throughout its lifespan;

[0006] After each discharge cycle, the discharge capacity of the lithium-ion battery energy storage system is used as the health status parameter, and the system is divided into four levels, with corresponding labels set for each level.

[0007] Acquire performance parameter data and tag data from the research and development and testing phases of lithium-ion battery energy storage systems, sort them chronologically, and construct a source domain dataset F. S , including input set X S and output set Y S ;

[0008] Using source domain dataset F S The mapping model between six performance parameters and state labels established using the stacked autoencoder algorithm is trained to obtain the source domain-oriented F S Model parameter set

[0009] Construct the target domain dataset F T The dataset is divided into a target domain normal state dataset and a target domain abnormal state dataset. The target domain normal state dataset is used to optimize the model parameter set. Optimize to obtain the target domain F T Model parameter set Next, the target domain abnormal state dataset is used to test the model parameter set. Conduct tests to verify the accuracy of the mapping model;

[0010] The validated mapping model is a lithium-ion battery health status assessment model, which can perform health assessments on lithium-ion battery energy storage systems.

[0011] Preferably, the performance parameters include direct and indirect measurements, specifically:

[0012] Directly measured values ​​include: battery internal temperature, charge / discharge internal resistance, and maximum voltage difference of a single cell;

[0013] Indirect measurements include: median discharge voltage, constant current charging time, and constant voltage charging voltage.

[0014] The maximum value of direct measurement within a discharge cycle is recorded in real time as a data sample, while the discharge capacity and indirect measurement are recorded directly after a discharge cycle is completed.

[0015] Preferably, the discharge capacity is divided into four levels, specifically defined as follows:

[0016] A discharge capacity below 85% is considered "abnormal" and should be labeled "3".

[0017] A discharge capacity between 85% and 90% is considered "accelerated decay," and the label is set to "2."

[0018] A discharge capacity between 90% and 95% is considered "decay" and is labeled "1".

[0019] A discharge capacity greater than or equal to 95% is defined as good, and the label is set to "0".

[0020] Preferably, the source domain dataset F S The construction process is as follows:

[0021] Construct source domain dataset FS input set X S Time: Obtain six performance parameters of a lithium-ion battery energy storage system at different times exhibiting degradation characteristics: battery internal temperature x1, charge / discharge internal resistance x2, maximum voltage difference of a single cell x3, median discharge voltage x4, constant current charging time x5, and constant voltage charging voltage value x6; represent the input set X in matrix form. S :

[0022]

[0023] Where k is the discharge period, and t represents the time when the parameter is recorded in the first discharge period;

[0024] Construct source domain dataset F S Output set Y S During the research and development of lithium-ion battery energy storage systems, system status tags are set based on the discharge capacity for each operation; the output set Y is represented in matrix form. S :

[0025]

[0026] Among them, Z1, Z2, Z3, and Z4 are set to the labels corresponding to the four discharge capacity ranges: when the battery discharge capacity is labeled "0", Z1 is set to 1, and Z2, Z3, and Z4 are set to 0; similarly, when the battery discharge capacity labels are "1", "2", and "3" respectively, Z2, Z3, and Z4 are set to 1 respectively, and the rest are set to 0.

[0027] Preferably, the source domain dataset F is used. S The model for the mapping relationship between six performance parameters and state labels, established using the stacked autoencoder algorithm, is trained, including the following steps:

[0028] S5-1 on the source dataset F S Input set X S The data is normalized so that the sample data values ​​are within the range of [0,1].

[0029] S5-2 Utilizing the normalized input set X S The mapping model between performance parameters and state labels is trained through two processes: forward unsupervised learning and backward supervised fine-tuning, to obtain the source domain-oriented F. S Model parameter set The model parameter set includes the weights and biases of the stacked autoencoder.

[0030] Preferably, in forward unsupervised learning, a stacked autoencoder network is used to obtain the input set X. S The reconstructed value z, with input set X SThe mean squared error between the reconstructed value z and the target value z is used as the objective function J. The model is trained layer by layer with the goal of minimizing J, and the optimal values ​​of the weights and biases are obtained. This optimal value is denoted as θ, which represents the model parameter set. S The objective function J is specifically:

[0031]

[0032] Where k is the discharge period, z i x is the reconstructed value of the performance parameters obtained through the self-stacked encoder. i The input parameters are the performance parameters in the input set.

[0033] Preferably, a backpropagation algorithm is used for supervised backpropagation fine-tuning, specifically as follows:

[0034] The state classification results output by the stacked autoencoder during the forward unsupervised learning process. and source domain output set Y S The corresponding state label y establishes the minimum cost function J. B :

[0035]

[0036] By minimizing the cost function J B To optimize and update the model parameter set θ S The weights and biases in the formula are defined by the following rules:

[0037]

[0038] Where L is the number of stacked autoencoder layers. Represents the model parameter set θ S In the nth layer stacked autoencoder, the weights between the i-th input layer neuron and the j-th hidden layer neuron; Represents the model parameter set θ S In the nth layer of the stacked autoencoder, the bias of the i-th input layer neuron is α, where α is the learning rate.

[0039] After completion and supervised fine-tuning, the source domain F is obtained. S Model parameter set

[0040] Preferably, based on the construction of source domain F S The method of constructing the target domain dataset F T When dividing the target domain into normal state datasets and abnormal state datasets, the data labels are divided based on battery capacity:

[0041] The performance parameters corresponding to the discharge capacity labels "0" and "1" are used as the input set for the normal state of the target domain.

[0042] The performance parameters corresponding to the discharge capacity labels “2” and “3” are used as the input set for abnormal states in the target domain.

[0043] Secondly, a terminal device is provided, comprising:

[0044] Memory, used to store at least one instruction executed by a processor;

[0045] The processor is used to execute instructions stored in the memory to implement the above-mentioned lithium-ion battery health status assessment modeling method based on transfer learning.

[0046] Thirdly, a computer-readable storage medium is provided, which stores computer instructions that, when executed on a computer, cause the computer to perform the aforementioned lithium-ion battery health status assessment modeling method based on transfer learning.

[0047] Compared with the prior art, the present invention has the following advantages:

[0048] (1) This invention innovatively proposes to integrate life cycle data under high rate and high stress conditions in the research and development and testing of lithium-ion battery energy storage system, and classifies the battery into healthy, degraded, accelerated degraded and abnormal states according to discharge capacity. It also uses stacked autoencoders to characterize the mapping relationship between performance parameters and healthy state, and realizes an accurate description of performance parameters from battery internal temperature, charge and discharge internal resistance, maximum voltage difference of single cell, median discharge voltage, constant current charging time, constant voltage charging voltage value to healthy state.

[0049] (2) This invention innovatively proposes to transfer the model parameters obtained during the development and testing of lithium-ion battery energy storage systems to the actual operation process. In combination with the characteristics of low rate and conventional stress in actual use, the model parameters are transferred and fine-tuned to improve the generalization ability of the model, thus realizing the transfer of knowledge accumulated in the product development process to the model in operation. Attached Figure Description

[0050] Figure 1 This is a flowchart of the method of the present invention;

[0051] Figure 2 This is a schematic diagram of the stacked autoencoder model of the present invention;

[0052] Figure 3 This is a schematic diagram illustrating the transfer learning process of the present invention;

[0053] Figure 4 This is a schematic diagram of the confusion matrix calculated by the present invention. Detailed Implementation

[0054] The technical problem solved by this invention is to overcome the shortcomings of the prior art and propose a modeling technology method for health status assessment of lithium-ion battery energy storage systems based on transfer learning. Following the idea of ​​"model parameter transfer + fine-tuning", it fully integrates the life cycle data accumulated by lithium-ion battery energy storage systems in various types of tests, and solves the problem of accurately assessing the health status during use.

[0055] A method for hierarchical modeling of the health status of lithium-ion battery energy storage systems based on transfer learning, such as... Figure 1 As shown, the main steps include:

[0056] S1. Determine the performance parameters of the lithium-ion battery energy storage system that exhibit changing trends throughout its lifespan, using discharge capacity as the criterion for state classification. The performance parameters include six parameters in total, categorized into direct and indirect measurements. Direct measurements include: battery internal temperature, charge / discharge internal resistance, and maximum differential voltage between individual cells. Indirect measurements include: median discharge voltage, constant current charging time, and constant voltage charging voltage. Direct measurements are recorded in real-time, and the maximum value within a discharge cycle is used as the data sample. Indirect measurements are recorded directly after a discharge cycle is completed.

[0057] S2. After each discharge cycle, the discharge capacity of the lithium-ion battery energy storage system is used as the health status parameter.

[0058] S3. The discharge capacity is divided into four levels: "good", "degradation", "accelerated degradation" and "abnormal". The capacity is defined as "abnormal" when it is below 85%, "accelerated degradation" when it is between 85% and 90%, "degradation" when it is between 90% and 95%, and "good" when it is greater than or equal to 95%.

[0059] S4, in the source domain F S To obtain data from the development and testing phases of lithium-ion battery energy storage systems, since high-rate and high-environmental-stress charge-discharge tests are usually conducted during the development phase, full life-cycle data can be obtained. After sorting by time sequence, health and degradation state data labels are set according to the discharge capacity defined by S3 to obtain a labeled dataset.

[0060] Source domain F S The dataset construction process is as follows:

[0061] Source domain F S Input set construction: Obtain six performance parameters of the lithium-ion battery energy storage system at different times exhibiting degradation characteristics: x1 battery internal temperature, x2 charge / discharge internal resistance, x3 maximum voltage difference of a single cell, x4 median discharge voltage, x5 constant current charging time, and x6 constant voltage charging voltage. The source domain F is represented in matrix form. S input set X SThe format is as follows.

[0062]

[0063] Where k is the discharge cycle, and t represents the time when the parameter is recorded in the first discharge cycle.

[0064] Source domain F S Output set construction: During the research and testing of lithium-ion battery energy storage systems, system status labels are set based on the discharge capacity for each operation: "0" represents a battery discharge capacity greater than 95%, "1" represents a discharge capacity between 90% and 95%, "2" represents a discharge capacity between 85% and 90%, and "3" represents a discharge capacity below 85%. The source domain F is represented in matrix form. S Output set Y S The format is as follows:

[0065]

[0066] Where k is the discharge cycle. When the battery discharge capacity is "0" (labeled), Z1 is set to 1, and Z2, Z3, and Z4 are set to 0. Similarly, when the battery discharge capacity is "1", "2", or "3" (labeled), Z2, Z3, and Z4 are set to 1, and the rest are set to 0.

[0067] S5. Establish a mapping model between six performance parameters and status labels using the Stack Auto-encoders (SAE) algorithm, including four steps:

[0068] (1) For the source domain F S The six performance parameters mentioned in step S1, namely battery internal temperature, charging and discharging internal resistance, maximum voltage difference of single cell, median discharge voltage, constant current charging time, and constant voltage charging voltage, are normalized to ensure that the sample data ranges between [0,1].

[0069] (2) Using the source domain input set X S The model for the mapping relationship between performance parameters and state labels is trained, including two processes: forward unsupervised learning and backward supervised fine-tuning.

[0070] In the forward unsupervised learning process, the source domain input set X is used. S Perform model training.

[0071] A single autoencoder consists of an input layer and hidden layers, such as Figure 2 As shown, a stacked autoencoder is composed of multiple autoencoders stacked together. The input of the first autoencoder in a stacked autoencoder is 6 performance parameters, and then they are stacked layer by layer. The hidden layer of the previous autoencoder becomes the input layer of the next autoencoder, and so on.

[0072] Each autoencoder uses weight parameters W and bias b to represent the transitive relationship between the input layer and the hidden layer. A stacked autoencoder consists of two parts: encoding and decoding. The encoding process is from the input layer to the hidden layer, and the decoding process is from the hidden layer to the output layer.

[0073] Let the number of stacked layers of the stacked autoencoder be L. Express the weight matrix parameters W and bias vector b of the stacked autoencoder in matrix form:

[0074] W = [w (1) ,w (2) ,L,w (n) ,L,w (L) ]

[0075] b = [b (1) ,b (2) ,L,b (n) ,L,b (L) ]

[0076] by Let W represent the elements in W, specifically the weights between the i-th input layer neuron and the j-th hidden layer neuron in the n-th stacked autoencoder. Let b represent the element, that is, the bias vector of the i-th input layer neuron in the n-th stacked autoencoder.

[0077] Stacked autoencoder networks with input set X S The mean square error J between the reconstructed value z and the target value z is used as the objective function to find the optimal values ​​of the weight matrix parameter W and the bias vector b.

[0078]

[0079] Where k is the discharge period, z i x is the reconstructed value of the performance parameters obtained through the self-stacked encoder. i The input parameters are the performance parameters in the input set.

[0080] The model is trained layer by layer with the goal of minimizing the loss function J, obtaining the weight matrix parameters W and the bias vector b. Let θ be the model parameter set obtained in this step. S .

[0081] (3) Using the source domain output set Y S Based on the backpropagation algorithm (BP) for θ S Perform reverse supervised fine-tuning.

[0082]

[0083] In the formula, The input performance parameters of the stacked autoencoder are the state classification results output by the input performance parameters during the forward unsupervised learning process, and y is the source domain output set Y. S The corresponding state label, the BP backpropagation fine-tuning algorithm minimizes the cost function J B To optimize model parameters.

[0084] Update θ S The weight matrix parameter W in s and bias vector b s The rules are as follows:

[0085]

[0086] Represents the model parameter set θ S In the nth layer stacked autoencoder, the weights between the i-th input layer neuron and the j-th hidden layer neuron. Represents the model parameter set θ S In the nth layer of the stacked autoencoder, the bias of the i-th input layer neuron is α, where α is the learning rate.

[0087] After fine-tuning, the source domain F is obtained. S Model parameter set

[0088] S6. Transfer the model parameter set Migrate to target domain F T and in the target domain F T Internally, model parameter learning is performed, such as Figure 3 As shown. Because lithium-ion battery energy storage systems are typically shallowly charged and discharged during use, and the discharge rate is relatively low, in order to ensure service life, the accumulated data is mostly health data, with relatively little fault data.

[0089] In the target domain F T The model parameter optimization learning includes four steps:

[0090] (1) In the target domain F T In accordance with the rules specified in S3, data tags are divided based on battery capacity, and the performance parameters corresponding to tags "0" and "1" are used as the normal state input set, denoted as X. T-normal This label is denoted as the normal output set, and is labeled as Y. T-normal The performance parameters labeled "2" and "3" are used as the abnormal state input set, denoted as X. T-failure This label is denoted as the abnormal state output set, and is called Y. T-failure ;

[0091] (2) Based on the SAE algorithm, using the target domain F T The dataset contains normal state input and output data sets that correspond to the model parameter set θ.s * Optimize to obtain the target domain F T The weight matrix parameter W T and bias vector b T The model parameter set θ T * ;

[0092] (3) Using the target domain F T The abnormal state input and output sets are tested, and the confusion matrix for state classification is calculated, such as... Figure 4 As shown, the accuracy of the transfer learning health status assessment model based on different operating modes of SAE lithium-ion battery energy storage system is verified.

[0093] (4) After the verification is completed, the model with a certain accuracy can be used to conduct a health assessment of the lithium-ion battery energy storage system. At the same time, as the target domain data is continuously accumulated, (1) to (4) can be repeated to continuously optimize the model parameters.

[0094] In a second aspect, the present invention provides a terminal device, comprising:

[0095] Memory, used to store at least one instruction executed by a processor;

[0096] The processor is used to execute instructions stored in the memory to implement the above-mentioned lithium-ion battery health status assessment modeling method based on transfer learning.

[0097] Thirdly, the present invention provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the above-described modeling method for assessing the health status of lithium-ion batteries based on transfer learning.

[0098] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A lithium-ion battery health status assessment modeling method based on transfer learning, characterized in that... include: Determine the performance parameters of a lithium-ion battery energy storage system that exhibit changing trends throughout its lifespan; After each discharge cycle, the discharge capacity of the lithium-ion battery energy storage system is used as the health status parameter, and the system is divided into four levels, with corresponding labels set for each level. Acquire performance parameter data and tag data from the research and development and testing phases of lithium-ion battery energy storage systems, sort them chronologically, and construct a source domain dataset F. S , including input set X S and output set Y S ; Using source domain dataset F S The mapping model between six performance parameters and state labels, established using the stacked autoencoder algorithm, is trained to obtain a model for the source domain dataset F. S Model parameter set Construct the target domain dataset F T The dataset is divided into a target domain normal state dataset and a target domain abnormal state dataset. The target domain normal state dataset is used to optimize the model parameter set. Optimize to obtain the target domain dataset F T Model parameter set Next, the target domain abnormal state dataset is used to test the model parameter set. Conduct tests to verify the accuracy of the mapping model; The verified mapping model is a lithium-ion battery health status assessment model, which can perform health assessments on lithium-ion battery energy storage systems. Performance parameters include direct and indirect measurements, specifically: Directly measured values ​​include: battery internal temperature, charge / discharge internal resistance, and maximum voltage difference of a single cell; Indirect measurements include: median discharge voltage, constant current charging time, and constant voltage charging voltage. The maximum value of direct measurement within a discharge cycle is recorded in real time as a data sample, while the discharge capacity and indirect measurement are recorded directly after a discharge cycle is completed.

2. The lithium-ion battery health status assessment modeling method based on transfer learning according to claim 1, characterized in that: The discharge capacity is divided into four levels, specifically defined as follows: A discharge capacity below 85% is considered "abnormal" and should be labeled "3". When the discharge capacity is between 85% and 90%, it is considered "accelerated decay" and the label is set to "2". When the discharge capacity is between 90% and 95%, it is considered "decay" and the label is set to "1". A discharge capacity greater than or equal to 95% is defined as good, and the label is set to "0".

3. The lithium-ion battery health status assessment modeling method based on transfer learning according to claim 2, characterized in that: Source domain dataset F S The construction process is as follows: Construct source domain dataset F S input set X S Time: Obtain six performance parameters of a lithium-ion battery energy storage system at different times exhibiting degradation characteristics: battery internal temperature x1, charge / discharge internal resistance x2, maximum voltage difference of a single cell x3, median discharge voltage x4, constant current charging time x5, and constant voltage charging voltage value x6; represent the input set X in matrix form. S : Where k is the discharge period, and t represents the time when the parameter is recorded in the first discharge period; Construct source domain dataset F S Output set Y S During the research and development of lithium-ion battery energy storage systems, system status tags are set based on the discharge capacity for each operation; the output set Y is represented in matrix form. S : Among them, Z1, Z2, Z3, and Z4 are set to the labels corresponding to the four discharge capacity ranges: when the battery discharge capacity is labeled "0", Z1 is set to 1, and Z2, Z3, and Z4 are set to 0; similarly, when the battery discharge capacity labels are "1", "2", and "3", Z2, Z3, and Z4 are set to 1 respectively, and the rest are set to 0.

4. The lithium-ion battery health status assessment modeling method based on transfer learning according to claim 1, characterized in that: Using source domain dataset F S The model for the mapping relationship between six performance parameters and state labels, established using the stacked autoencoder algorithm, is trained, including the following steps: S5-1 on the source dataset F S Input set X S The data is normalized so that the sample data values ​​are within the range of [0,1]. S5-2 Utilizing the normalized input set X S The mapping model between performance parameters and state labels is trained through two processes: forward unsupervised learning and backward supervised fine-tuning, to obtain a model oriented towards the source domain dataset F. S Model parameter set The model parameter set includes the weights and biases of the stacked autoencoder.

5. The lithium-ion battery health status assessment modeling method based on transfer learning according to claim 4, characterized in that: In forward unsupervised learning, a stacked autoencoder network is used to obtain the input set X. S The reconstructed value z, with input set X S The mean squared error between the reconstructed value z and the target value z is used as the objective function J. The model is trained layer by layer with the goal of minimizing J, and the optimal values ​​of the weights and biases are obtained. This optimal value is denoted as θ, which represents the model parameter set. S The objective function J is specifically: Where k is the discharge period, z i x is the reconstructed value of the performance parameters obtained through the self-stacked encoder. i The input parameters are the performance parameters in the input set.

6. The lithium-ion battery health status assessment modeling method based on transfer learning according to claim 5, characterized in that: The backpropagation algorithm is used for supervised backpropagation fine-tuning, specifically as follows: The state classification results output by the stacked autoencoder during the forward unsupervised learning process. and source domain output set Y S The corresponding state label y establishes the minimum cost function J. B : By minimizing the cost function J B To optimize and update the model parameter set θ S The weights and biases in the formula are defined by the following rules: Where L is the number of stacked autoencoder layers. Represents the model parameter set θ S In the nth layer stacked autoencoder, the weights between the i-th input layer neuron and the j-th hidden layer neuron; Represents the model parameter set θ S In the nth layer of the stacked autoencoder, the bias of the i-th input layer neuron is α, where α is the learning rate. After completing the supervised fine-tuning, we obtain the dataset F oriented towards the source domain. S Model parameter set 7. The lithium-ion battery health status assessment modeling method based on transfer learning according to claim 3, characterized in that: Based on the construction of the source domain dataset F S Methods for constructing target domain dataset F T When dividing the dataset into target domain normal state dataset and target domain abnormal state dataset, the data labels are divided based on discharge capacity: The performance parameters corresponding to the discharge capacity labeled "0" and "1" are used as the input set for the normal state of the target domain. The performance parameters corresponding to the discharge capacity labels "2" and "3" are used as the input set for abnormal states in the target domain.

8. A terminal device, characterized in that, include: Memory, used to store at least one instruction executed by a processor; A processor for executing instructions stored in memory to implement the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-7.

Images