State of health forecasting of Lithium-ion batteries applicable in real-world operational conditions

Currently, several methods for battery state of health (SOH) prediction exist which are applicable to battery electric vehicles (BEV). However, only few research has been conducted on SOH forecasting based on features that encode causes for battery ageing applicable in real world applications. This paper proposes a machine learning method for SOH forecasting applicable for BEV fleet managers and battery designers in real world applications. As model inputs, we use the battery's operation time within certain operation ranges defined by combinations of the battery signals current, state of charge (SOC) and temperature. Different variants of this temporal aggregation of the battery operation time and of the operation ranges of the battery signals are examined. Our findings state that combining different cycle window widths w(w) to one training data set improves the generalization of the model. Also, we find that the fineness of the operational ranges of the signals does not limit the model's performance if w(w) is larger than 100 cycles or different w(w) are combined.

Automated summary

The study presents a machine learning method for predicting the state of health of batteries in electric vehicles, which can be applied in real-world applications. It was found that combining different cycle window widths into one training dataset improves the generalization of the model, and the granularity of the operational ranges of the signals does not limit the model's performance.