Multi-Scale Self-Attention Convolutional Neural Network for Energy Storage State-of-Charge Estimation

State-of-Charge (SOC) estimation in Li-ion batteries increases the efficiency of energy management systems, extending battery life. Accurate SOC prediction helps users better meet their energy needs while optimizing the energy consumption of devices. However, the use of batteries under different operating conditions makes SOC estimation a challenge. In this paper, a deep learning-based model that can accurately predict the SOC of li-ion battery cells under different temperatures has been proposed. The model is able to extract different feature maps by analyzing the input sequence data at different scales. In this way, correlations between different time periods can be revealed. Another feature of the model is that it can detect long-term patterns in each feature map by analyzing the obtained multi-scale features with a self-attention mechanism. In this way, SOC prediction will be improved not only based on previous data but also based on the relevant points in the sequence. In the experiments for positive temperatures, an average of $\mathbf{29.1 \%}$ success increase was achieved for LA92 drive cycle and $\mathbf{12.3 \%}$ for UDDS. In the experiments for all temperatures, an average of 61.9% success increase was observed for LA92 drive cycle and $\mathbf{46.2 \%}$ for UDDS.