State-of-charge estimation of lithium-ion battery based on clockwork recurrent neural network

State of charge (SOC) is the most important parameter in battery management system (BMS). Firstly, in this paper, a new structure of standard recurrent neural network (RNN), named clockwork recurrent neural network (CWRNN) is introduced, which divides hidden layer into separate modules, assigns each module a different specify clock speed to solve long-term dependencies. Secondly, because of each module in CWRNN has different clock speeds, it makes computation only at its prescribed clock period, rather than compute and update all the inner parameters at every time step, so that CWRNN can reduce the training and computation cost obviously. Finally, employed network is verified at dynamic drive cycle at different temperature. The result shows that proposed network has satisfactory estimation results, such as the root mean square error (RMSE) is less than 1.29%. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This paper presents a novel neural network structure called CWRNN, which effectively addresses long-term dependencies, reduces training and computation costs, and is validated under different temperature conditions.