State of Charge Estimation Using a Neural Network for a Battery-Powered DC-DC Converter System

This paper presents a methodology for estimating the state of charge (SOC) of a battery using a feedforward neural network. The estimation is based on real-time measurements of the battery voltage under load, the output voltage of a buck DC-DC converter, and the converter duty cycle. A dataset for training the neural network is generated through simulations of the battery-converter-load system in PLECS, considering various SOC levels, duty cycles, and load resistances. The trained neural network is implemented in C code and integrated back into the PLECS simulation to evaluate its real-time performance. The proposed approach offers a data-driven SOC estimation method that leverages system-level dynamics, potentially improving the accuracy and adaptability of battery monitoring in power electronics applications.