State of Charge Estimation for Lithium-Ion Batteries via Recursive Gated Recurrent Unit Neural Networks

Accurate estimation of the State of Charge (SOC) of lithium-ion batteries is crucial for ensuring the safe operation of battery systems. Currently, SOC estimation methods based on recurrent neural networks such as the Gated Recurrent Unit (GRU) have garnered significant attention, as they can achieve accurate SOC estimation without the need for predefined battery models. However, these estimation methods suffer from excessively high computational complexity, making them difficult to apply in practical engineering scenarios. To address the issue of high computational complexity caused by the need for extensive hidden state iterations in traditional GRU neural networks for SOC estimation, a recursive update approach with temporal inheritance of network hidden states is proposed. By improving the output structure of the GRU network, this method enables accurate SOC estimation for the current time step with only a single network computation on the sampled data. Compared to traditional GRU methods reported in the literature, this recursive GRU method reduces computational load by over 99% while maintaining the same level of SOC estimation accuracy, demonstrating promising application potential. Additionally, to tackle the issue of insufficient battery training data in certain application scenarios, the method incorporates transfer learning to rapidly complete network training. Validation using laboratory test datasets and publicly available datasets shows that the method can accurately estimate SOC for lithium-ion batteries under different temperature conditions, aging states, and battery models, with a maximum estimation error of no more than 3%.