Neural network-based prediction of the state of charge of Lithium-ion battery

Abstract: The problem of predicting the state of charge (SOC) of lithium-ion batteries (LIB) for energy storage systems of technological equipment and mobile objects is considered. A neural network-based approach is proposed to overcome the limitations of tra-ditional mathematical models that require prior knowledge of battery parameters. Three types of architectures are examined-multilayer perceptrons (MLP), recurrent neural networks (RNN), and gated recurrent units (GRU) – to analyze their ef-fectiveness in SOC prediction. A methodology is developed for training these neural network models on experimental data, in-cluding time series of voltage, current and temperature under dynamic discharge conditions. Differences in accuracy and computational costs are confirmed: MLP networks train faster but fail to capture temporal dependencies, while RNN and GRU networks demonstrate lower test errors but demand higher com-putational resources. The practical applicability of the results is validated through the optimization of SOC estimation algo-rithms under real-world conditions.

Keywords: gru, rnn, neural network architectures: mlp, neural net-works, state of charge, lithium-ion battery