Hybrid framework for predicting and forecasting State of Health of Lithium-ion batteries in Electric Vehicles

Abstract

This paper has proposed a hybrid framework to accurately predict and forecast the State of Health (SOH) of Lithium-ion batteries for Electric Vehicles (EV) using noisy data. Due to significant environmental and sustainability benefits, the EVs are getting popular worldwide. The EVs are getting fully powered from Lithium-ion batteries instead of fossil fuel. Therefore, the Li-ion batteries in EV should be under progressively manage and control to ensure improved efficiency and safety to prevent failure. The State of Health (SOH) is one of the main indicator which is very decisive for reliable battery management system. This paper has presented a hybrid framework to reduce negative impact of noisy data for accurate prediction and forecasting of the SOH using a public but noisy dataset. The framework has used statistical and machine learning techniques, like Auto Regressive Integrated Moving Average (ARIMA), linear and Ridge regression, with Savitzky–Golay (S-G) filter to design hybrid models. The unique characteristic of these proposed models is their resistance against bad data to handle data fluctuation that may cause overfitting. Based on the experiment, the paper has presented comparative study on a number of performance metric, which show, in spite of its simplicity, the proposed prediction model shows better accuracy than existing similar techniques. Furthermore, five day-ahead forecasting is a dazzling characteristic of this framework. • Battery State of Health (SoH) prediction and forecasting framework for Electric Vehicles. • Hybrid models using Savitzky–Golay (S-G) to filter noise data to improve accuracy. • Hybrid feature selection process to identify significant battery parameters for SoH. • A comparative study to validate performance of the proposed framework with baseline.