A support vector machine-based state-of-health estimation method for lithium-ion batteries under electric vehicle operation

Abstract

Capacity and resistance are state-of-health (SOH) indicators that are essential to monitor during the application of batteries on board electric vehicles. For state-of-health determination in laboratory environment, standard battery performance tests are established and well-functioning. Since standard performance tests are not available on-board a vehicle, we are developing a method where those standard tests are applied virtually to a support vector machine-based battery model. This data-driven model is solely based on variables available during ordinary electric vehicle (EV) operation such as battery current, voltage and temperature. This article contributes with a thorough experimental validation of this method, as well as the introduction of new features – capacity estimation and temperature dependence. Typical EV battery usage data is generated and exposed to the suggested method in order to estimate capacity and resistance. These estimations are compared to direct measurements of the SOH indicators with standard tests. The obtained estimations of capacities and instantaneous resistances demonstrate good accuracy over a temperature and state-of-charge range typical for EV operating conditions and allow thus for online detection of battery degradation. The proposed method is also found to be suitable for on-board application in respect of processing power and memory restrictions.