EV BATTERY HEALTH PREDICTION USING PYTHON

Abstract

Lithium-ion batteries are the most expensive and most critical component of an electric vehicle, and their condition decides the driving range, the safety of the vehicle, and its resale value. With every charge and discharge cycle, chemical and physical changes inside the cells reduce the capacity that the battery can store and increase its internal resistance. This gradual loss is described by the state of health, usually expressed as the present capacity as a percentage of the rated capacity. This paper presents a Python-based system that predicts the state of health and the remaining useful life of electric vehicle batteries using machine learning. The system uses publicly available battery ageing datasets in which lithium-ion cells were repeatedly charged and discharged under controlled conditions while voltage, current, temperature, and time were recorded for every cycle. These are supplemented by readings from a small test setup built by the project team, in which an 18650 cell is charged and discharged through an electronic load while an INA219 current and voltage sensor and an NTC thermistor are read by a microcontroller. The measurements are transferred to a computer, where each cycle is processed to calculate delivered capacity, energy, charge time, and average temperature. Feature engineering is central to the approach. For every cycle, features are extracted that are known to change as the battery ages, including the time taken to reach the upper voltage limit during constant-current charging, the duration of the constantvoltage phase, the voltage drop at the start of discharge, the mean discharge voltage, the peak temperature rise, and features from incremental capacity analysis. Several regression models are trained and compared, including Linear Regression, Support Vector Regression, Random Forest, Gradient Boosting, XGBoost, and a Long ShortTerm Memory network. Performance is measured using RMSE, MAE, MAPE, and the coefficient of determination.