Battery state-of-the-health prediction using AI techniques

Abstract

Since emission issues have sounded the alarm bell, energy security and environmental protection issues have become increasingly prominent. Meeting the requirement of zero pollution, clean energy replacement, and high energy efficiency, many improvements have been adopted in the transportation areas such as the development of electric vehicles (EVs). As one of the key essential parts of an electric vehicle, Li-ion battery has been widely used in EV for its high energy density, long life cycle, and high safety level. At present, Lithium-ion batteries are used to build battery packs in series and parallel in electric vehicles. However, aging happens in the ability of a battery that storing energy and providing power decrease over battery life cycles. To evaluate and predict if the consumed battery should be replaced, the state of health (SOH) is brought forward, which is an essential parameter to determine battery degradation state. It should be noted that experimental methods are impractical to monitor every electric vehicle, which can be time-consuming and costly. The data-driven method only builds a learning model containing input variables and output variables from the perspective of data to find out the characteristic of battery SOH change, which is simple and easy to implement. In this dissertation, a novel SOH estimation is proposed based on the charging voltage curve using a random forest (RF) method. The results show that according to the battery charging curve, an accurate SOH estimation can be achieved.