Machine Learning Based State of Charge Estimation of the Li-Ion Batteries

Abstract

This thesis focuses on the development of an accurate and reliable machine learning based State of Charge (SoC) estimation method for Lithium-Ion (Li-ion) batteries. Pre cise approximation of SoC is critical in the effective and secure functioning of batteries, particularly in electric vehicles and renewable energy storage systems. The proposed method utilizes machine learning algorithms to predict the SOC of the battery by an alyzing the battery voltage, current, and temperature data. The methodology involves the collection of data from the battery under various operating conditions, followed by data Pre-processing. Three models trained and tested in the study are Long Short Term Memory, Convolution Neural Network and Deep Neural Networks. Pre-processed data is then fed into a machine learning model (LSTM,CNN and DNN) for training and validation using 80/20 and 80/20 split, to estimate the SoC. The execution of the presented approach is assessed using experimental data and the performance parameter used for comparison and evaluation is Mean Squared Error (MSE). The performance of our implemented models is compared with the models of the reviewed research articles. Comparisons show that LSTM based model has higher performance (MSE=0.014) com pared to the other reviewed implementations. Finally, the trained model is tested on a real-time dataset generated using the developed Battery Management System. Results show acceptable results indicating its potential for practical applications in the field of energy storage.