Lithium-ion battery (LIB) terminal voltage models are investigated using two modelling approaches. The first model is a third-order Thevenin equivalent circuit model (ECM), which consists of an open-circuit voltage in series with a nonlinear resistance and three parallel RC pairs. The parameters of the ECM are obtained by fitting the model to hybrid pulse power characterization (HPPC) test data. The parametrization of the ECM is performed through quadratic-based programming. The second is a novel modelling approach based on long short-term memory (LSTM) recurrent neural networks to estimate the battery terminal voltage. The LSTM is trained on multiple vehicle drive cycles at six different temperatures, including −20°C, without the necessity of battery characterization tests. The performance of both models is evaluated with four automotive drive cycles at each temperature. The results show that both models achieve acceptable performance at all temperatures. However, the LSTM performs better in 92% of the cases, especially at lower temperatures, where it has as much as two-thirds lower error than the ECM approach.
Comparative Study between Equivalent Circuit and Recurrent Neural Network Battery Voltage Models
Sae Technical Papers
SAE WCX Digital Summit ; 2021
2021-04-06
Aufsatz (Konferenz)
Englisch
A Comparative Study of Recurrent Neural Network Architectures for Battery Voltage Prediction
SAE Technical Papers | 2021
|A Comparative Study of Recurrent Neural Network Architectures for Battery Voltage Prediction
British Library Conference Proceedings | 2021
|A Comparative Study of Recurrent Neural Network Architectures for Battery Voltage Prediction
British Library Conference Proceedings | 2021
|A comparative study of equivalent circuit models for Li-ion batteries
Online Contents | 2012
|