Modeling of Traction Batteries for Rail Applications Using Artificial Neural Networks

Abstract

The development of novel operational strategies for battery electric trains requires a vehicle model including the traction battery. This paper proposes a method to generate accurate traction battery models on system level for application in a simulation model of battery electric multiple units. Artificial neural networks are used to identify the coherences within real system data from a traction battery used in an electric bus. Two approaches are examined to estimate the terminal voltage: a feedforward neural network and a long short-term memory network. Model generation is followed by a comparison with an existing physics-based battery model in order to prove the increase of accuracy.