Remaining discharge energy estimation for lithium-ion batteries using pattern recognition and power prediction

Abstract

The remaining discharge energy (RDE) of a battery is an important value for estimating the remaining range of a vehicle. Prediction based methods for calculating RDE have been proven to be suitable for improving energy estimation accuracy. This paper aims to further improve the estimation accuracy by incorporating novel load prediction techniques with pattern recognition into the RDE calculation. For the pattern recognition, driving segment data was categorised into different usage patterns, then a rule-based logic was designed to recognise these, based on features from each pattern. For the power prediction, a clustering and Markov modelling approach was used to group and define power levels from the data as states and find the probabilities of each state-to-state transition occurring. This data was defined for each pattern, so that the logic could inform what data should be used to predict the future power profile. From the predicted power profile, the RDE was calculated from the product of the predicted load and the predicted voltage, which was obtained from a first-order battery model. The proposed algorithm was tested in simulation and real-time using battery cycler data, and compared against other prediction-based methods. The proposed method was shown to have desirable accuracy and robustness to modelling errors. The primary conclusion from this research was using pattern recognition can improve the accuracy of RDE estimation.