A review on methods for state of health forecasting of lithium-ion batteries applicable in real-world operational conditions

Abstract

The ageing of Lithium-ion batteries can be described as change of state of health (∆SOH). It depends on the battery's operation during charging, discharging, and rest phases. Mapping the operation conditions during these phases for long time windows to a ∆SOH enables forecasting the battery's SOH. With SOH forecasting fleet managers of battery electric vehicle (BEV) fleets can plan vehicle replacement and optimize the fleet's operational strategy. Inspired by the applicability from a user's perspective of fleet managers and battery designers, this work motivates and defines key criteria for SOH forecasting models. The key criteria concern the encoding of information in the model inputs, model transferability to other batteries, and the applicability to 2nd life battery applications. Based on these key criteria we review SOH forecasting models. Currently, only few models satisfy the majority of the defined key criteria, while three others only fail at two key criteria. The majority (71 %) of the methods use machine learning models which can be seen as current research trend due to the complex dependence of battery operational data and battery ageing. We show limitations of the applicability and comparability of existing models due to different data sets, different metrics, different output values, and different forecast horizons. Furthermore, code and data are only rarely shared and publicly available.