FAST CHARGING STRATEGIES AND OPTIMISATION AND SOH ESTIMATION USING DIFFERENT ML BASED ALGORITHMS FOR A LI ION BATTERY

Abstract

The increasing demand for electric vehicles (EVs) and portable electronics has highlighted the critical need for fast charging solutions and battery optimization strategies. This abstract provides an overview of the key aspects surrounding fast charging and optimization techniques in batteries. Fast charging, typically defined as the ability to charge a battery to a significant capacity within a short period, presents a range of challenges. These include thermal management, electrochemical stress, and overall battery lifespan. To address these challenges, researchers and engineers have developed various strategies and optimizations. This abstract explores the following areas: Battery Chemistry and Materials, Thermal Management, Charging Protocols, Battery Management Systems (BMS), Fast Charging Infrastructure, Artificial Intelligence and Machine Learning, Environmental and Economic Considerations. Accurate estimation of battery state of health is very important for computing remaining lifetime of a battery and also to get rid of any battery risk. Although estimating SOH (State of Health) is challenging because of the nonlinear electrochemical behaviour of lithium-ion batteries and due to uncertain operating conditions of EVs. There exist several methods to estimate the state of health of the battery. In general, the existing SOH approaches are associated with the following drawbacks: due to time-consuming process, they are challenging to implement, accuracy highly depends on the battery model parameters, and during experimental data set preparation, many assumptions are observed. These are some problems that makes these SOH estimating methods not suitable for EVs. This thesis demonstrates deep learning-based data-driven methods for estimating SOH by collecting the past data from lithium-ion batteries. This work is done to propose a different SOH estimation method using Feedforward Neural Network (FFNN), Convolution Neural Network (CNN), and Long Short Term Memory (LSTM) Neural Network. Among these three methods, LSTM is preferred as it has the least mean square error.