Machine Learning Based Soc and Soh Prediction for Renewable Battery Systems

Abstract

A considerable problem posed on the operational front for Battery Energy Storage Systems (BESS) is the intermittency and non-linearity of renewable sources such as solar and wind. The traditional battery management systems are based on deterministic rule-based methods, which contribute to the inaccuracy of estimation of state of charge (SOC) and state of health (SOH) under dynamic conditions. The proposed predictive framework is based on machine learning to deal with integrating renewable energy into battery systems, using multivariate time series to help in modelling, feature normalization, artificial neural networks for nonlinear mapping, coupled with adaptive control to make real-time estimation. The SOC and SOH accuracy as per MATLAB-based verification with dynamic load and generation profiles turned out to be 97% and 96%, respectively, with 35% of error reduction. The proposed system promotes energy conservation, opens an easy channel for maintenance and enhances the life of a battery, making it a results-driven and scalable solution for smart grids and electric vehicles.