Estimating State of Charge and State of Health of Electrified Vehicle Battery by Data Driven Approach: Machine Learning

Abstract

The recent development, increased interest and achievements in artificial intelligence and Machine Learning (ML) have facilitated the development of novel methods for estimating State of charge (SoC) and State of heath (SoH) of electrified car batteries. SoC and SoH are critical to the performance, passenger comfort, and safety of electric vehicles (EVs), as well as minimizing costs associated with overdesign or oversizing of the battery pack. Two methods of ML techniques: Feedforward Back Propagation Neural Network (FBPNN) and Cascaded Feedforward Neural Network (CFNN) for estimation of SoC and SoH for electrified car batteries have been proposed using real time sample data retrieved from NASA Ames Prognostics and Panasonic 18650PF Li-Ion Battery Data Repository. The input Data set contains discharging current, ambient temperature and battery voltage. The ML algorithms have been trained using three inputs and battery states (SoH and SoC) of electrified car batteries as targets. The MATLAB based nntool toolbox has been utilized for estimation purpose. The results demonstrated that proposed CFNN has better performance in estimation and have smaller overshoots and undershoots from the actual value than the FBPNN.