Abstract
Accurate estimation of state of charge (SOC) is essential for the applications of lithium-ion battery. Although many machine learning based SOC estimation algorithms have been proposed, the two common and long-lasting problems remain unsolved, i.e. tedious neural network training process and unsatisfying robustness to measurement noises. To solve these two problems while providing highly accurate SOC estimation, this paper proposes a novel recurrent equilibrium network (REN) based algorithm. Direct parameterization technique is employed to substantially simplify the neural network training. The proposed algorithm produces outstanding SOC estimation results under varying temperatures. Its efficacy is verified by experimental results.
Funding Information
  • future battery industries CRC's mine electrification project

This publication has 22 references indexed in Scilit: