Using Genetic Programming to Learn Behavioral Models of Lithium Batteries

Abstract

Li-ion batteries play a key role in the sustainable development scenario, since they can allow a better management of renewable energy resources. The performances of Li-ion batteries are influenced by several factors. For this reason, accurate and reliable models of these batteries are needed, not only in the design phase, but also in real operating conditions. In this paper, we present a novel approach based on Genetic Programming (GP) for the voltage prediction of a Lithium Titanate Oxide battery. The proposed approach uses a multi-objective optimization strategy. The evolved models take in input the State-of-Charge (SoC) and provide as output the Charge/discharge rate (C-rate), which is used to evaluate the impact of the charge or discharge speed on the voltage. The experimental results showed that our approach is able to generate optimal candidate analytical models, where the choice of the preferred one is made by evaluating suitable metrics and imposing a sound trade-off between simplicity and accuracy. These results also proved that our GP-based behavioral modeling is more reliable and flexible than those based on a standard machine learning approach, like a neural network.