Retired battery capacity estimation method and device and medium

Abstract

The invention relates to a decommissioned battery capacity estimation method and device and a medium. The method comprises the following steps: carrying out a cyclic aging experiment on a plurality of common batteries, calibrating the capacity of the batteries, measuring an electrochemical impedance spectrum, extracting the characteristics of the electrochemical impedance spectrum, and constructing a source domain data set; training a basic capacity estimation model based on the source domain data set; measuring electrochemical impedance spectrums of a large batch of retired batteries and extracting features; randomly selecting a small number of decommissioned batteries from a large batch of decommissioned batteries as transfer learning samples, calibrating the capacity, and constructing a target domain data set; migration from the basic capacity estimation model to target domain data is realized by adopting a feature matching-based migration learning method; and inputting the electrochemical impedance spectroscopy characteristics of a large batch of decommissioned batteries into the migrated model, and estimating the capacity of the decommissioned batteries. Compared with the prior art, the technical problems of long test time and high cost in the prior art are solved, repeated model training for different types of retired batteries is not needed, and the method has the advantages of low implementation cost, high applicability and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage batteries, and in particular to a method, device and medium for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning. Background Art

[0002] Due to their superior electrochemical properties, lithium-ion batteries have become the preferred power unit for electric vehicles. However, in recent years, the number of power batteries retired from electric vehicles has increased exponentially, placing a huge burden on the environment. Retired batteries can still serve as valuable energy carriers with a capacity of more than 70% of their initial capacity and can be used in applications with lower performance requirements, such as electric bicycles and energy storage systems. Therefore, cascade utilization is an effective way to safely and environmentally dispose of retired batteries. However, the capacity of retired batteries, even from the same vehicle, can show considerable differences. Retired batteries experience complex and changing operating conditions when used on-board, making it challenging to accurately predict the capacity degradation trajectory. In addition, historical data of retired batteries is usually difficult to obtain, and it is necessary to rely on on-site charge and discharge tests to calibrate the battery capacity. This approach results in long testing time, low efficiency, and high cost, which directly affects the economic feasibility of cascade utilization. Therefore, a pressing challenge is to develop efficient methods to provide fast and accurate capacity estimation for retired batteries. Summary of the invention

[0003] The purpose of the present invention is to provide a method, device and medium for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning, so as to ensure that the trained model can achieve effective generalization and accurate capacity estimation under different operating conditions, materials and application scenarios.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] A method for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning comprises the following steps:

[0006] S1. Perform cycle aging experiments on multiple common batteries. During the aging process, calibrate the battery capacity and measure the electrochemical impedance spectrum. Extract the electrochemical impedance spectrum features and use the corresponding capacity as a label to construct a source domain dataset.

[0007] S2, training a basic capacity estimation model based on the source domain dataset;

[0008] S3, measure the electrochemical impedance spectra of a large number of retired batteries and extract features;

[0009] S4. Randomly select a small number of retired batteries from a large number of retired batteries as transfer learning samples and calibrate their capacities. Based on the electrochemical impedance spectroscopy characteristics of a small number of retired battery samples and their corresponding capacities as labels, construct a target domain dataset.

[0010] S5. Use a feature matching-based transfer learning method to achieve the migration of the basic capacity estimation model to the target domain data;

[0011] S6. Input the electrochemical impedance spectroscopy characteristics of a large number of retired batteries into the migrated model to estimate the capacity of the retired batteries.

[0012] The battery capacity calibration method is:

[0013] Use the current corresponding to the battery to charge and discharge between the upper and lower cut-off voltages of the battery, where charging is constant current and constant voltage charging, and discharging is constant current discharging. Set sufficient rest time between charging and discharging, and define the battery capacity as the average value of the charging capacity and the discharging capacity.

[0014] The electrochemical impedance spectroscopy is measured by:

[0015] The electrochemical impedance spectroscopy of the battery is measured using a constant current titration method and an electrochemical workstation, wherein the excitation current is determined according to the battery characteristics. If the measured electrochemical impedance spectroscopy curve has noise, the excitation current is increased and remeasured until a complete and smooth electrochemical impedance spectrum is obtained.

[0016] The method for extracting the electrochemical impedance spectroscopy characteristics is:

[0017] The real and imaginary parts of the electrochemical impedance spectrum are extracted as characteristic sequences, that is, the electrochemical impedance spectrum Z is expressed in complex form as:

[0018] Z=Z'+jZ"

[0019] Where Z' is the real part, representing the resistance of the battery at a given frequency; Z" is the imaginary part, representing the reactance of the battery at a given frequency.

[0020] The source domain dataset is defined as:

[0021]

[0022]

[0023] Among them, the superscript S represents the source domain; X i represents the feature sequence extracted from the i-th electrochemical impedance spectroscopy sample in the data set; Z' i,k and Z" i,k are the real and imaginary parts of the i-th electrochemical impedance spectrum at the k-th frequency; y i Indicates the capacity label corresponding to the i-th impedance spectrum sample, m s Represents the total number of EIS samples in the source domain dataset.

[0024] The target domain dataset is defined as:

[0025]

[0026] Among them, the superscript T represents the target domain, m t Represents the total number of impedance spectra in the target domain dataset.

[0027] The feature matching-based transfer learning method refers to adding a linear transformation layer before the basic capacity estimation model, and adapting to the changes of new data only by adjusting the weights of the newly added linear transformation layer, specifically:

[0028] The capacity labels of the samples in the target domain dataset are normalized to the source domain. The normalized capacity labels are:

[0029]

[0030] in, and Represent the minimum and maximum values ​​of the capacity in the source domain, respectively. and Respectively represent the minimum and maximum values ​​of the capacity in the target domain, represents the capacity label corresponding to the i-th impedance spectrum sample in the target domain dataset, represents the normalized capacity label of the i-th impedance spectrum sample in the target domain dataset;

[0031] Search the capacity label with the smallest difference from the normalized capacity label from the source domain as the optimal matching sample;

[0032] The mapping between the target domain and the source domain is achieved by fitting the linear relationship between the feature sequence of the optimal matching sample and the feature sequence of the target domain sample:

[0033]

[0034] Among them, z k represents the real or imaginary part of the impedance spectrum at the kth frequency; a k and b k It represents the fitting weight, the superscript S represents the source domain, and the superscript T represents the target domain.

[0035] The S6 comprises the following steps:

[0036] S61, inputting the electrochemical impedance spectroscopy characteristics of each retired battery into the capacity estimation model after migration, and outputting the capacity estimation result;

[0037] S62, performing inverse normalization on the output capacity estimation result, and re-mapping the output back to the target domain, wherein the inverse normalization method is:

[0038]

[0039] in, and They represent the estimated capacity of the target domain after inverse normalization and the estimated capacity of the target domain before inverse normalization, respectively.

[0040] A retired battery capacity estimation device based on electrochemical impedance spectroscopy and transfer learning comprises a memory, a processor, and a program stored in the memory, wherein the processor implements the method described above when executing the program.

[0041] A storage medium stores a program, which implements the method described above when executed.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] (1) The present invention uses electrochemical impedance spectroscopy as a feature, which is simple and fast to obtain and is not affected by variable historical charging and discharging conditions. It can achieve efficient data collection of large quantities of retired batteries and frequent estimation of capacity.

[0044] (2) The feature matching-based transfer learning method proposed in the present invention can effectively transfer the basic model established based on laboratory aging data to retired battery data for use. It is not limited by working conditions, materials and application scenarios and has strong versatility. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of the method of the present invention;

[0046] Figure 2 The battery capacity attenuation curve and impedance spectrum change diagram of the source domain data set in the embodiment of the present invention;

[0047] Figure 3 is a capacity distribution diagram of retired batteries in an embodiment of the present invention;

[0048] Figure 4 This is a graph showing changes in the impedance spectrum of a retired battery in an embodiment of the present invention;

[0049] Figure 5 Graph showing capacity estimation results of retired batteries in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0051] This embodiment provides a method for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning. Figure 1 As shown, the following steps are included:

[0052] S1. Perform cycle aging experiments on multiple common batteries. During the aging process, calibrate the battery capacity and measure the electrochemical impedance spectrum. Extract the electrochemical impedance spectrum features and use the corresponding capacity as a label to construct a source domain dataset.

[0053] Among them, the battery capacity calibration method is:

[0054] Use the current corresponding to the battery to charge and discharge between the upper and lower cut-off voltages of the battery, where charging is constant current and constant voltage charging, and discharging is constant current discharging. Set sufficient rest time between charging and discharging, and define the battery capacity as the average value of the charging capacity and the discharging capacity.

[0055] The measurement method of electrochemical impedance spectroscopy is:

[0056] Since the impedance of a vehicle battery is usually in the milliohm range, this embodiment uses a constant current titration method. Using the constant current titration method, an electrochemical workstation is used to measure the electrochemical impedance spectrum of the battery, wherein the excitation current is determined according to the battery characteristics. If the measured electrochemical impedance spectrum curve has noise, the excitation current is increased and re-measured until a complete and smooth electrochemical impedance spectrum is obtained.

[0057] The method for extracting the electrochemical impedance spectroscopy features is:

[0058] The real and imaginary parts of the electrochemical impedance spectrum are extracted as characteristic sequences, that is, the electrochemical impedance spectrum Z is expressed in complex form as:

[0059] Z=Z'+jZ"

[0060] Where Z' is the real part, representing the resistance of the battery at a given frequency; Z" is the imaginary part, representing the reactance of the battery at a given frequency.

[0061] Then, the source domain dataset is defined as:

[0062]

[0063] Among them, the superscript S represents the source domain; X i represents the feature sequence extracted from the i-th electrochemical impedance spectroscopy sample in the data set; Z' i,k and Z" i,k are the real and imaginary parts of the i-th electrochemical impedance spectrum at the k-th frequency; y i Indicates the capacity label corresponding to the i-th impedance spectrum sample, m s Represents the total number of EIS samples in the source domain dataset.

[0064] In this embodiment, the battery used for training the model is a ternary lithium-ion battery with a charge and discharge cut-off voltage of 4.2V and 2.5V respectively, and a nominal capacity of 2750mAh; the retired battery is a lithium iron phosphate battery with a charge and discharge cut-off voltage of 3.65V and 2.5V respectively, and a nominal capacity of 8600mAh, but the actual application is not limited to this.

[0065] Four ternary lithium batteries were tested for charge and discharge cycles under different working conditions until the battery life decayed to 80% of the initial capacity, i.e. the end of life. Battery capacity calibration and impedance testing were performed at intervals of 25 to 100 cycles. The capacity calibration used a nominal current of 0.5C, charging was constant current and constant voltage charging, discharging was constant current discharge, and the capacity was the average value of the charge and discharge quantity. Impedance was measured by constant current titration, and the root mean square of the excitation current was 500mA. The impedance measurement frequency range was 0.1Hz to 10kHz, and 10 points were recorded in every 10 times the frequency range. Figure 2 The battery capacity decay curve and the impedance spectrum of battery #01 are shown. The impedance gradually increases and shifts as the battery ages.

[0066] The real and imaginary parts are extracted from the electrochemical impedance spectrum as feature sequences, and the corresponding capacities are used as labels to construct a source domain dataset for training machine learning models.

[0067] S2. Train a basic capacity estimation model based on the source domain dataset.

[0068] In this embodiment, a capacity estimation model is constructed using a long short-term memory neural network, but the actual application is not limited to this. Model training is performed on a computer with a 3.20 GHz processor and 16 GB RAM. The Adam optimizer is used to optimize the model parameters, and the initial learning rate is set to 10 -3 .

[0069] S3. Measure the electrochemical impedance spectra of a large number of retired batteries and extract features.

[0070] S4. Randomly select a small number of retired batteries from a large number of retired batteries as transfer learning samples and calibrate their capacities. Based on the electrochemical impedance spectroscopy characteristics of a small number of retired battery samples and using their corresponding capacities as labels, construct a target domain dataset.

[0071] In this embodiment, the target domain dataset is defined as:

[0072]

[0073]

[0074] The superscript T represents the target domain, m t Represents the total number of impedance spectra in the target domain dataset.

[0075] In this embodiment, the target domain data set consists of 81 retired lithium iron phosphate batteries, and the battery capacity distribution is as follows: Figure 3 As shown. The impedance measurement of retired batteries also uses constant current titration, and the root mean square of the excitation current is 5000mA. The impedance measurement frequency range is 0.1Hz to 10kHz, and 10 points are recorded in every 10 times frequency range. Compared with the time of charge and discharge test capacity, the impedance test time of each battery is about 1.67min, which greatly improves the test efficiency. Figure 4 The impedance spectrum of the batch of batteries is shown. Ten batteries are randomly selected from the batch of batteries as transfer learning samples.

[0076] S5. Use the feature matching-based transfer learning method to achieve the migration of the basic capacity estimation model to the target domain data.

[0077] The feature matching-based transfer learning method refers to adding a linear transformation layer before the basic capacity estimation model, and adapting to the changes of new data only by adjusting the weights of the newly added linear transformation layer. Specifically:

[0078] The capacity labels of the samples in the target domain dataset are normalized to the source domain. The normalized capacity labels are:

[0079]

[0080] in, and Represent the minimum and maximum values ​​of the capacity in the source domain, respectively. and Respectively represent the minimum and maximum values ​​of the capacity in the target domain, represents the capacity label corresponding to the i-th impedance spectrum sample in the target domain dataset, represents the normalized capacity label of the i-th impedance spectrum sample in the target domain dataset;

[0081] Search the capacity label with the smallest difference from the normalized capacity label from the source domain as the optimal matching sample;

[0082] The mapping between the target domain and the source domain is achieved by fitting the linear relationship between the feature sequence of the optimal matching sample and the feature sequence of the target domain sample:

[0083]

[0084] Among them, z k represents the real or imaginary part of the impedance spectrum at the kth frequency; a k and b k It represents the fitting weight, the superscript S represents the source domain, and the superscript T represents the target domain.

[0085] This embodiment uses a feature matching-based transfer learning method based on 10 randomly selected batteries to migrate the established basic model to the target domain dataset for use. The key to this method is to find the best matching object by normalizing the target domain label to the range of the source domain label, and then apply a linear transformation to map the features between the target domain sample and the best matching object in the source domain. Only by adjusting the weights of the newly added layer can the changes in the relationship between the input features and the output labels be adapted.

[0086] S6. Input the electrochemical impedance spectroscopy characteristics of a large number of retired batteries into the migrated model to estimate the capacity of the retired batteries.

[0087] S6 includes the following steps:

[0088] S61, inputting the electrochemical impedance spectroscopy characteristics of each retired battery into the capacity estimation model after migration, and outputting the capacity estimation result;

[0089] S62, inverse normalize the output capacity estimation result, and remap the output back to the target domain, wherein the inverse normalization method is:

[0090]

[0091] in, and They represent the estimated capacity of the target domain after inverse normalization and the estimated capacity of the target domain before inverse normalization, respectively.

[0092] Since the model outputs capacity labels that match the capacity range of the source domain, it is necessary to remap the output back to the target domain by denormalizing the output values.

[0093] In this embodiment, the accuracy of the retired battery capacity estimation is evaluated by using the mean absolute percentage error (MAPE) and the root mean square percentage error (RMSPE) as evaluation indicators. Figure 5 The results of retired battery capacity estimation are shown, with MAPE and RMSPE of 2.75% and 5.04% respectively. This result shows that the proposed method can reduce the cost of repeated testing, modeling and training of specific retired batteries while maintaining the accuracy of capacity estimation, which helps to improve the efficiency of large-scale retired battery grading.

[0094] This embodiment also provides a retired battery capacity estimation device based on electrochemical impedance spectroscopy and transfer learning, including a memory, a processor, and a program stored in the memory, and the processor implements the method described above when executing the program.

[0095] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0096] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning, characterized in that: The following steps are involved: S1. Perform cycle aging experiments on multiple common batteries. During the aging process, calibrate the battery capacity and measure the electrochemical impedance spectrum. Extract the electrochemical impedance spectrum features and use the corresponding capacity as a label to construct a source domain dataset. S2, training a basic capacity estimation model based on the source domain dataset; S3, measure the electrochemical impedance spectra of a large number of retired batteries and extract features; S4. Randomly select a small number of retired batteries from a large number of retired batteries as transfer learning samples and calibrate their capacities. Based on the electrochemical impedance spectroscopy characteristics of a small number of retired battery samples and their corresponding capacities as labels, construct a target domain dataset. S5. Use a feature matching-based transfer learning method to achieve the migration of the basic capacity estimation model to the target domain data; S6. Input the electrochemical impedance spectroscopy characteristics of a large number of retired batteries into the migrated model to estimate the capacity of the retired batteries.

2. According to claim 1, a retired battery capacity estimation method based on electrochemical impedance spectroscopy and transfer learning is characterized in that: The battery capacity calibration method is: Use the current corresponding to the battery to charge and discharge between the upper and lower cut-off voltages of the battery, where charging is constant current and constant voltage charging, and discharging is constant current discharging. Set sufficient rest time between charging and discharging, and define the battery capacity as the average value of the charging capacity and the discharging capacity.

3. The method for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning according to claim 1, characterized in that: The electrochemical impedance spectroscopy is measured by: The electrochemical impedance spectroscopy of the battery is measured using a constant current titration method and an electrochemical workstation, wherein the excitation current is determined according to the battery characteristics. If the measured electrochemical impedance spectroscopy curve has noise, the excitation current is increased and remeasured until a complete and smooth electrochemical impedance spectrum is obtained.

4. The method for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning according to claim 1, characterized in that: The method for extracting the electrochemical impedance spectroscopy characteristics is: The real and imaginary parts of the electrochemical impedance spectrum are extracted as characteristic sequences, that is, the electrochemical impedance spectrum Z is expressed in complex form as: Z=Z'+jZ" Where Z' is the real part, representing the resistance of the battery at a given frequency; Z" is the imaginary part, representing the reactance of the battery at a given frequency.

5. A method for estimating capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning according to claim 4, characterized in that: The source domain dataset is defined as: Among them, the superscript S represents the source domain; X i represents the feature sequence extracted from the i-th electrochemical impedance spectroscopy sample in the data set; Z' i,k and Z" i,k are the real and imaginary parts of the i-th electrochemical impedance spectrum at the k-th frequency; y i Indicates the capacity label corresponding to the i-th impedance spectrum sample, m s Represents the total number of EIS samples in the source domain dataset.

6. A method for estimating capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning according to claim 5, characterized in that: The target domain dataset is defined as: Among them, the superscript T represents the target domain, m t Represents the total number of impedance spectra in the target domain dataset.

7. The method for estimating capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning according to claim 1, characterized in that: The feature matching-based transfer learning method refers to adding a linear transformation layer before the basic capacity estimation model, and adapting to the changes of new data only by adjusting the weights of the newly added linear transformation layer, specifically: The capacity labels of the samples in the target domain dataset are normalized to the source domain. The normalized capacity labels are: in, and Represent the minimum and maximum values ​​of the capacity in the source domain, respectively. and Respectively represent the minimum and maximum values ​​of the capacity in the target domain, represents the capacity label corresponding to the i-th impedance spectrum sample in the target domain dataset, represents the normalized capacity label of the i-th impedance spectrum sample in the target domain dataset; Search the capacity label with the smallest difference from the normalized capacity label from the source domain as the optimal matching sample; The mapping between the target domain and the source domain is achieved by fitting the linear relationship between the feature sequence of the optimal matching sample and the feature sequence of the target domain sample: Among them, z k represents the real or imaginary part of the impedance spectrum at the kth frequency; a k and b k It represents the fitting weight, the superscript S represents the source domain, and the superscript T represents the target domain.

8. The method for estimating the capacity of retired batteries based on electrochemical impedance spectroscopy and transfer learning according to claim 7, characterized in that: The S6 comprises the following steps: S61, inputting the electrochemical impedance spectroscopy characteristics of each retired battery into the capacity estimation model after migration, and outputting the capacity estimation result; S62, performing inverse normalization on the output capacity estimation result, and re-mapping the output back to the target domain, wherein the inverse normalization method is: in, and They represent the estimated capacity of the target domain after inverse normalization and the estimated capacity of the target domain before inverse normalization, respectively.

9. A retired battery capacity estimation device based on electrochemical impedance spectroscopy and transfer learning, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 8 is implemented.

Images