A Particle Swarm Optimized Multi-Model Framework for Remaining Useful Life Prediction of Lithium-Ion Batteries Using Domain-Driven Feature Engineering

With respect to battery management and safe operation and maintenance scheduling of electric vehicles (EVs), it is very important to predict the remaining useful life (RUL) of lithium-ion batteries (LIBs). Accurate prediction of RUL can bring secure working conditions, avert internal and external failure, and, last, avoid any undesirable consequences. However, achieving accurate prediction of RUL is complicated for EV applications due to various reasons such as the complex operational characteristics, dynamic changes in the model parameters during the aging process, extraction of battery parameters, data preparation, and hyper-parameter tuning of the predictive model. This research proposes a novel approach that integrates Particle Swarm Optimization (PSO) with a multi-model technique for RUL prediction. The framework integrates many machine learning (ML) models and deep learning (DL) models. Combining domain knowledge, advanced optimization techniques, and learning models to make high-accuracy RUL predictions reduces maintenance costs and improves battery management systems. This study uses domain-driven feature engineering to extract battery-specific indicators, including voltage drops, charging time, and temperature fluctuations, to increase model accuracy. Among the evaluated models, LSTM demonstrates superior performance, achieving a mean absolute error (MAE) of 0.34, a root mean square error (RMSE) of 0.76, and an R2 of 0.93, providing the best results in RUL prediction. The proposed research uniquely integrates PSO-based optimization with domain-driven feature engineering across multiple machine learning and deep learning models, demonstrating a unified and novel approach that significantly improves the prediction accuracy of RUL in LIBs.