Deep-Learning-Based SoC Estimation for $\mathbf{LiFePO}_{4}$ Batteries Using CNN-GRU-TCN Hybrid Architecture

Accurate estimation of the state of charge (SOC) is critical for the efficient and relative operation of a battery management system (BMS) in energy storage applications. This paper presents a novel hybrid deep learning model combining convolutional neural networks (CNNs), gated recurrent units (GRUs), and temporal convolutional networks (TCNs), which combines the RNN model features and temporal dependencies of non-linear features like voltage, current, and temperature to create a relationship with SOC. The complex relationships between temporal dependencies and the monitoring signals are derived from the DL methods of lithium iron phosphate <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\mathbf{LiFePO}_{4})$</tex> batteries. The proposed model leverages the feature extraction capabilities of CNNs, the temporal dynamics modeling of GRUs, and the long effective memory capabilities of sequence prediction strengths of TCN to enhance the accuracy and robustness of SOC estimation. We conducted experiments with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{LiFePO}_{4}$</tex> data, taken from Influx DB, processed, and used for training and validation of models with an 80:20 ratio. Additionally, we compared the performance of our model with that of LSTM, CNN-LSTM, GRU, CNN-GRU, and CNN-GRU-LSTM. The experimental results confirm that our proposed CNN-GRU-TCN hybrid model outperforms the other models in SOC estimation in LiFePO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</inf> batteries.