Estimation of Cooling Circuits' Temperature in Battery Electric Vehicles Using Karhunen Loeve Expansion and LSTM

Estimation of vehicle components' temperatures is essential to compute their thermal fatigue and life expectancy, and is typically based on lumped-parameter models, such as thermal networks. However, the design of such models requires expert knowledge and detailed information about the cooling circuit, thus data-driven approaches have been employed. In this article, a regression architecture for estimating the coolant temperature is proposed. The architecture utilizes a two branch LSTM regression model with optimized parameters and is separately modelling active and passive vehicle states. The proposed architecture was evaluated using data from a real battery electric vehicle from 1,000 hours of driving and showed an improvement in performance up-to 36.9% in terms of mean absolute error.