Improved Multi-head Bi-directional Long and Short-term Memory Temporal Convolutional Network for Lithium-ion Batteries State of Charge Estimation in Energy Storage Systems

Lithium-ion batteries with their high voltage, large capacity, high discharge rate, no memory effect, and green environmental protection advantages are widely used in communication, power stations, backup power, and other energy storage fields. Accurate estimation of the state of charge (SOC) of lithium-ion batteries is a key prerequisite to ensure the safe, reliable, and efficient operation of battery systems. To address this core challenge, this paper innovatively proposes a composite model combining the Kepler optimization algorithm, temporal convolutional network (TCN), bi-directional long and short-term memory network (BiLSTM), and multi-head attention mechanism (MHA). Kepler optimization algorithm was used to search the optimal hyperparameters in the TCN-BiLSTM structure so that the model could adjust the structural parameters and extract the input features accurately under different working conditions and temperatures. The multi-head self-attention mechanism is introduced to assign different weights to the feature outputs extracted by time convolution according to the different importance of information to improve the adaptability of the model. Finally, the proposed model is tested and compared with other models under different temperatures and working conditions.