State of charge monitoring of Li-ion batteries for electric vehicles using GP filtering

Abstract

Electric vehicles are dependent on onboard battery management systems that protect the battery from functioning outside its safe operating limits by monitoring its state of charge (SOC). Advanced online monitoring techniques are required so that the performance of the energy management is not lowered severely. However, the behavior of batteries is difficult to be predicted online because of its nonlinearity, intrinsic variability and fluctuating environmental conditions. Gaussian Process (GP)-Bayesian filters are based on probabilistic non-parametric Gaussian models of hidden states using available measurements. As a result, model response variability can be explicitly incorporated into the prediction and measurement steps, which is usually not the case for more traditional filtering strategies that resort to parametric models for state estimation. In this work, GP models were incorporated into nonparametric filtering techniques to monitor the battery SOC online. Results show that Bayes’ filtering techniques increase the predictability of the SOC under uncertainty about the effect of environmental conditions on the SOC.