Optimizing Battery’s Remaining Useful Life Prediction using PSO-LSTM Model

Abstract

Batteries now serve as the main solution amid the shift to renewable energy. Between 2011 and 2022, the National Aeronautics and Space Administration (NASA) recorded an approximate notable 40% increase in global CO2 concentration in parts per million (PPM). Electric Vehicles (EVs) have contributed to sustainable reductions in CO2 and PPM. Nevertheless, battery technology has drawbacks related to its end-of-life phase, commonly referred to as Remaining Useful Life (RUL), which can potentially lead to fires or dangerous toxic emissions from the cells. This study will employ a data-driven approach utilizing artificial intelligence to predict battery RUL, thereby effectively preventing potential battery failures with the developed model. The stochastic Particle Swarm Optimization - Long Short Term Memory (PSO-LSTM) model has been used to optimize window size and hidden layer hyperparameters. This was done to achieve a model with minimal error. PSO-LSTM forecasts data sourced from NASA’s Prognosis Centre of Excellence (PCoE), encompassing both “Cycle” and distinct “Capacity” measurements. PSO-LSTM model optimization produced RMSE of 0.074 and MAE values of 0.021 on 80% training data, and RMSE of 0.070 with MAE of 0.021 on 20% testing data. This stochastic PSO-LSTM model can be integrated into intelligent Battery Management Systems (BMS) for specific future automotive industry applications.