A Novel Hybrid Approach for Remaining Useful Life and Short Term Capacity Prediction of Lithium- ion Batteries

Accurately determining the remaining usable life of lithium-ion batteries is essential for effective battery management. This paper presents a novel hybrid method for predicting both the Remaining Useful Life (RUL) and short-term capacity of batteries, known as ZOA-MSVM, which integrates the Zebra Optimization Algorithm (ZOA) with Multi-kernel Support Vector Machine (MSVM). The objective of the paper is to provide precise forecasts of future capacities and RUL for lithium-ion batteries (LIBs) while effectively managing uncertainty. The ZOA-MSVM model is trained using data from LIBs, leveraging the Zebra Optimization Algorithm to identify optimal kernel parameters, penalty factors, and weight coefficients for the MSVM model. This trained model predicts battery capacity and RUL by considering key factors such as capacity, current, and voltage during discharge operations. When compared to existing techniques, including artificial neural networks (ANN), ant lion optimizer-optimized ANN, and adaptive network-based fuzzy inference systems, the ZOA-MSVM method demonstrates improved performance with reduced errors during both charging and discharging. Notably, the uncertainty intervals produced by the ZOA-MSVM technique often encompass the actual remaining life of the battery, and its charging error is lower than that of other current methods. Utilizing the NASA battery dataset, an observed data sequence is generated for regression prediction. Results indicate that the enhanced algorithm significantly improves prediction accuracy and generalization ability, while also reducing training time and computational complexity.