Enhanced SOC Estimation Framework for Lithium-ion Batteries Based on Fusion of Sparrow Search Algorithm and Deep Neural Networks

Precise state of charge estimation is becoming increasingly critical as battery management systems face stricter functional demands. To address the difficulties in data-driven model hyperparameter optimization and the issue of insufficient generalization ability, this paper proposes an innovative model that integrates the sparrow search optimization algorithm and deep neural networks, fully utilizing the significant advantages of deep neural networks in feature automatic extraction, nonlinear mapping modeling, and large-scale data processing. Meanwhile, combining the unique global search and local development balance mechanism of the sparrow search optimization algorithm, it achieves adaptive optimization of hyperparameter combinations, effectively avoiding the limitations of traditional manual parameter tuning, such as low efficiency and poor generalization. The model was trained on dynamic driving cycle data at three different temperatures. Testing confirms the effectiveness of the model in state of charge estimation, exhibiting consistent performance across different temperatures, with an MAE within 1.6%, an RMSE within 2%, and an R2 exceeding 99%. This paper demonstrates that the proposed model exhibits excellent stability and accuracy.