CN-122017639-A - Battery health state self-adaptive assessment method and system based on transfer learning

CN122017639ACN 122017639 ACN122017639 ACN 122017639ACN-122017639-A

Abstract

The invention provides a battery health state self-adaptive assessment method and system based on transfer learning in the technical field of lithium battery management and state monitoring, wherein the method comprises the steps of S1, collecting charging voltage data of a target lithium battery to generate an increment capacity curve, S2, taking a voltage value corresponding to the maximum increment capacity as a center reference, self-adaptively selecting voltage segment data from the charging voltage data based on a preset voltage range window, S3, fixing sharing parameters for extracting general characteristics in a pre-trained battery health state assessment model, selecting preset number of segment sub-data from the voltage segment data, and finely adjusting the battery health state assessment model to adapt to the target lithium battery, and S4, enabling the voltage segment data to belong to the finely-adjusted battery health state assessment model to obtain a battery health state assessment result. The method has the advantage that the accuracy, the robustness and the generalization capability of the battery health state assessment are greatly improved.

Inventors

YANG LIU
GUO WEIQIANG
WU GUIFANG
CHI SHENGSONG

Assignees

福建星云电子股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (10)

1. A self-adaptive evaluation method of battery health status based on transfer learning is characterized by comprising the following steps: Step S1, collecting charging voltage data of a target lithium battery in a constant-current charging stage, calculating and generating an incremental capacity curve based on the charging voltage data, and performing smoothing treatment on the incremental capacity curve to inhibit noise; Step S2, searching the maximum increment capacity on the increment capacity curve after the smoothing treatment, taking a voltage value corresponding to the maximum increment capacity as a center reference, and adaptively selecting voltage fragment data from the charging voltage data based on a preset voltage range window; Step S3, fixing sharing parameters for extracting general features in a pre-trained battery health state evaluation model, and selecting a preset number of segment sub-data from the voltage segment data to finely adjust the battery health state evaluation model so as to adapt to a target lithium battery; And S4, enabling the voltage segment data to belong to the finely-adjusted battery health state evaluation model, and obtaining a battery health state evaluation result.
2. The method for adaptively estimating a battery state of health based on transfer learning as set forth in claim 1, wherein in said step S1, said incremental capacity curve is smoothed by a Gaussian filter algorithm.
3. The method for adaptively estimating a battery state of health based on transfer learning as set forth in claim 1, wherein in the step S2, the width of the voltage range window is preset according to the incremental capacity curve characteristics of lithium batteries of different models, so as to ensure that the voltage range window can cover an electrochemical characteristic region strongly related to the battery state of health.
4. The method for adaptively estimating the battery health status based on the transfer learning of claim 1, wherein in the step S3, the battery health status estimation model is a convolutional neural network model, and the network structure comprises a convolutional layer, a pooling layer and a full-connection layer which are sequentially connected; The convolution layer is used for scanning and processing the input voltage segment data through convolution check and superposing deviation to extract battery state characteristics; the full-connection layer is used for outputting a battery state evaluation result according to the battery state characteristics after dimension scaling; The sharing parameters include at least weight parameters of the convolutional layer.
5. The method for adaptively estimating the state of health of a battery based on transfer learning as set forth in claim 1, wherein in the step S3, the model for estimating the state of health of the battery is pre-trained by minimizing a mean square error between a predicted SOH, i.e., a result of estimating the state of health of the battery, by taking as input a data set constructed from historical voltage segment data labeled with a true SOH.
6. A battery state of health self-adaptive evaluation system based on transfer learning is characterized by comprising the following modules: The incremental capacity curve generation module is used for collecting charging voltage data of the target lithium battery in a constant-current charging stage, calculating and generating an incremental capacity curve based on the charging voltage data, and smoothing the incremental capacity curve to inhibit noise; The voltage segment self-adaptive selection module is used for searching the maximum increment capacity on the increment capacity curve after the smoothing processing, taking a voltage value corresponding to the maximum increment capacity as a center reference, and self-adaptively selecting voltage segment data from the charging voltage data based on a preset voltage range window; The migration learning module is used for fixing shared parameters for extracting general features in the pre-trained battery health state evaluation model, selecting a preset number of segment sub-data from the voltage segment data, and finely adjusting the battery health state evaluation model to adapt to a target lithium battery; and the health state evaluation module is used for enabling the voltage segment data to belong to the finely-adjusted battery health state evaluation model to obtain a battery health state evaluation result.
7. The adaptive battery state of health assessment system according to claim 6, wherein said incremental capacity curve generation module smoothes said incremental capacity curve by a Gaussian filter algorithm.
8. The adaptive battery state of health assessment system according to claim 6, wherein the width of the voltage range window is preset according to the incremental capacity curve characteristics of lithium batteries of different models in the voltage segment adaptive selection module, so as to ensure that the voltage range window can cover an electrochemical characteristic interval strongly related to the battery state of health.
9. The adaptive battery health status assessment system based on transfer learning of claim 6, wherein the battery health status assessment model is a convolutional neural network model, and the network structure comprises a convolutional layer, a pooling layer and a full-connection layer which are sequentially connected; The convolution layer is used for scanning and processing the input voltage segment data through convolution check and superposing deviation to extract battery state characteristics; the full-connection layer is used for outputting a battery state evaluation result according to the battery state characteristics after dimension scaling; The sharing parameters include at least weight parameters of the convolutional layer.
10. The adaptive battery state of health assessment system based on transfer learning of claim 6, wherein the transfer learning module is configured to pre-train a data set constructed from historical voltage segment data labeled with true SOH as input by minimizing a mean square error between predicted SOH, i.e., a battery state of health assessment result.

Description

Battery health state self-adaptive assessment method and system based on transfer learning Technical Field The invention relates to the technical field of lithium battery management and state monitoring, in particular to a self-adaptive battery state of health assessment method and system based on transfer learning. Background With the wide application of new energy automobiles, portable electronic devices and large-scale energy storage systems, lithium batteries are particularly important as key power sources for accurate assessment and safety monitoring of State of Health (SOH). SOH is a core index reflecting the degradation degree of battery performance, and directly influences the use safety, cruising ability and operation reliability of an energy system of equipment. Therefore, the development of the efficient and accurate battery state of health assessment method has important practical significance and wide application value. However, the conventional battery SOH evaluation method has the following technical limitations: (1) Model stiffness and special parameter dependence the existing methods are mostly based on preset, solidified electrochemical models or empirical decay formulas. Once established, the model is difficult to adjust, and the evaluation effect of the model is seriously dependent on special parameters such as internal resistance, capacity and the like acquired under specific working conditions. When applied to lithium batteries of different batches, models or materials, such fixed models tend to evaluate misalignment, lacking the necessary adaptability, due to performance differences between the batteries. In essence, they cannot take advantage of existing knowledge (such as the large amount of data of other types of batteries) to quickly adapt to new target batteries. (2) In the context of rapid iterations of lithium battery technology, conventional approaches often require a completely new evaluation model to be built from scratch, due to lack of sufficient historical data, in the face of new or popular models of batteries. The process involves the re-execution of complete feature engineering, model structure design and parameter training, resulting in long development cycle, large consumption of computing resources, high adaptation cost, and difficulty in meeting the application requirements of rapid batch deployment. (3) The lack of generalization across domains is that the feature extraction rules of an assessment model trained on a specific dataset are deeply coupled with the inherent characteristics of the battery, making it difficult for its "learned knowledge" to migrate to other different types of batteries. The high dependence on the training data field makes the generalization capability of the model seriously insufficient when facing the new field (new battery type) with scarce data, and restricts the universality of large-scale engineering application. Therefore, how to provide a self-adaptive evaluation method and system for battery health status based on transfer learning, so as to improve the accuracy, robustness and generalization capability of battery health status evaluation, is a technical problem to be solved urgently. Disclosure of Invention The invention aims to solve the technical problem of providing a self-adaptive evaluation method and a self-adaptive evaluation system for the battery health state based on transfer learning, which can improve the accuracy, the robustness and the generalization capability of the battery health state evaluation. In a first aspect, the present invention provides a method for adaptively estimating a battery state of health based on transfer learning, including the steps of: Step S1, collecting charging voltage data of a target lithium battery in a constant-current charging stage, calculating and generating an incremental capacity curve based on the charging voltage data, and performing smoothing treatment on the incremental capacity curve to inhibit noise; Step S2, searching the maximum increment capacity on the increment capacity curve after the smoothing treatment, taking a voltage value corresponding to the maximum increment capacity as a center reference, and adaptively selecting voltage fragment data from the charging voltage data based on a preset voltage range window; Step S3, fixing sharing parameters for extracting general features in a pre-trained battery health state evaluation model, and selecting a preset number of segment sub-data from the voltage segment data to finely adjust the battery health state evaluation model so as to adapt to a target lithium battery; And S4, enabling the voltage segment data to belong to the finely-adjusted battery health state evaluation model, and obtaining a battery health state evaluation result. Further, in the step S1, the incremental capacity curve is smoothed by a gaussian filter algorithm. Further, in step S2, the width of the voltage range window is preset according to the incr