An Interpretable Electric Vehicles Battery State of Charge Estimation Using MHDTCN-GRU

Lithium-ion batteries are the driving force behind electric vehicles and portable electronic devices. Accurate estimation of the state of charge in lithium-ion batteries is crucial for optimizing battery performance and improving energy efficiency. This paper proposes a novel hybrid model that combines a multi-head dilated temporal convolutional network architecture with a gated recurrent unit to anticipate the state of charge levels. The novel multi-head architecture of the dilated temporal convolutional network facilitates simultaneous learning of patterns across different scales, allowing the model to adapt to new patterns quickly. The diverse dilation rates in the dilated temporal convolutional network enhance the model's capability to capture long-term sequences, while the gated recurrent unit focuses on short-term dependencies, offering a versatile state of charge estimation method suitable for various environmental conditions. Additionally, the incorporation of the explainable artificial intelligence technique - Shapley Additive exPlanations aids in achieving global interpretability for state of charge prediction, offering a precise quantification of the influence of individual attributes. Comprehensive experiments were conducted across various temperatures and driving cycles to demonstrate the effectiveness of the proposed model. The computation results indicate the proposed method's adaptability to varying conditions, achieving high estimation accuracy and robustness with a mean absolute percentage error and root mean square percentage error of 0.54% and 0.84%, respectively, along with a parameter count of 3,74,433. Moreover, the proposed architecture enhances state of charge estimation performance compared to existing models across multiple datasets while maintaining a more efficient parameter count.