Presentation Information
[23p-P05-39]Multi-scale Analysis of Voltage Curves for Accurate and Adaptable Prediction of Full Lifecycle for Lithium-ion Batteries
〇(D)Hongmin JIANG1, YAN Lijing2, MA Tingli1 (1.Kyushu Inst. Tech., 2.China Jiliang Univ.)
Keywords:
Lithium-ion batteries,Data-driven method,Health indicator
Health status prediction of lithium-ion batteries (LIBs) is critical for the stable operation of electrical equipment. The data-driven approach can fit the degradation laws based on the historical cyclic data and identify potential problems in time. Existing prediction methods primarily rely on monitoring various data parameters, such as capacity, voltage, temperature, current, impedance, etc. However, challenges persist regarding monitoring accuracy and cost-effectiveness with current methods, and integrating multiple parameters remains a pressing issue. Herein, this study focuses on only analyzing voltages, employing multiscale feature extraction, and modeling techniques, aiming to improve the accuracy and adaptability in predicting the full lifecycle of LIBs. A novel cyclic voltage data pre-processing technique was introduced, involving the selection of feature parameters with high correlation to battery life across both time and frequency domains. The proposed multi-scale Graph Convolutional Network model effectively captures the feature variations in the graph-structured data at different positions within the sampling window. The single-step prediction experiments demonstrate the capability to predict subsequent capacity degradation solely based on the voltages of any consecutive 15 cycles, with the root mean square error in testing below 2% and exhibiting adaptable capability across different battery datasets. This study highlights the significance of a multi-scale approach to voltage analysis in cyclic data, leveraging advanced modeling techniques, not only to enhance the accuracy and adaptability of full lifecycle predictions of LIBs but also to offer a robust framework for overcoming prevailing challenges in monitoring accuracy and cost-effectiveness.