Capacity degradation influenced state of charge and life cycle prediction through tri-layered WNN model with extensive energy storage characterization

Abstract

The conventional coulomb counting method for state of charge (SoC) estimation in battery management systems (BMS) is hindered by its inability to account for self-discharge and capacity degradation, leading to inaccurate SoC readings and suboptimal predictions of battery life cycles. To address this research gap, the study proposes a machine learning-driven novel tri-layered cascaded model to precisely predict the state of charge and accurately forecast the life cycle of various types of battery energy storage systems, considering the effects of self-discharge and capacity degradation. The hyperparameter tuned proposed optimal tri-layered cascaded machine learning model was trained and tested using real-time data acquired from a hardware test bench model, integrating the IT6006C-300-75 programmable bi-directional power supply and IT9000 interfacing software. In addition to that, to forecast the life cycle through the process of layered prediction, the power density and the energy density has been predicted and projected with a detailed mathematical derivation as the functions of time. The proposed tri-layered machine learning model not only refines the SoC measurement process by considering the impact of the self-discharge but also integrates capacity degradation into lifecycle prediction, thus offering a robust solution for advanced BMS in complex micro-grid environments.