A MACHINE LEARNING FRAMEWORK FOR REAL-TIME EV BATTERY RUL PREDICTION AND CHARGING OPTIMIZATION

Abstract

Electric Vehicle (EV) adoption is rapidly accelerating, with global EV sales surpassing 14 million units in 2023, representing a 35% increase over the previous year. Studies indicate that battery performance and energy prediction account for nearly 45% of operational inefficiencies in EVs, while accurate charging strategies can reduce energy waste by up to 30%. Despite this potential, existing systems often rely on manual estimations or simplistic rule-based approaches for battery monitoring, which suffer from inconsistent results due to human error, lack of adaptability to real-time data, and inability to model non-linear battery degradation behaviours. To overcome these limitations, this study proposes a machine learning-based predictive framework using Bagging with Decision Tree Regressor (DTR) to estimate the Remaining Useful Life (RUL) of EV batteries. The methodology begins with collecting high-dimensional EV battery datasets, followed by robust preprocessing including feature scaling, outlier removal, and unique value analysis. Baseline models such as Support Vector Regressor (SVR) and Deep Neural Network (DNN) are first applied for comparative analysis. The proposed ensemble model leverages the Bagging technique, which trains multiple DTRs on different bootstrap samples and averages their predictions to reduce variance and improve accuracy. The system is further integrated into a Django-based web application that handles user interaction, model training, and real-time prediction using uploaded battery data. The result is a scalable, interpretable, and highly accurate prediction tool that can significantly optimize EV charging schedules and battery life cycle management