ScholarWorks@경상국립대학교

초록

Existing machine learning approaches for lithium-ion battery state-of-charge estimation predominantly focus on increasing model complexity or extensive hyperparameter tuning, often overlooking the role of task-specific optimization dynamics. This study presents a task-aware optimization framework for SOC estimation by developing an advanced gradient descent algorithm and incorporating it into a gradient boosting regression model. This approach shifts the emphasis from model complication to optimization design. The proposed optimization algorithm integrates adaptive learning rate adjustment, gradient clipping, bias correction, and feature engineering to enhance training stability and convergence behavior under different operating conditions. The numerical behavior of the proposed optimizer was first verified using a single-neuron NumPy implementation, which was evaluated across three driving profiles and six temperature conditions. The optimizer was subsequently integrated into a gradient boosting regression model to form the proposed advanced gradient descent optimization-gradient-boosting regressor framework. Experimental results show that the proposed approach achieves high state-of-charge estimation accuracy, with average root mean square error and mean absolute error values of 0.103% and 0.015%, respectively. The task-aware optimization framework produced consistently improved estimation performance compared with conventional training strategies, demonstrating that optimization-centric design is an effective and underexplored method for enhancing state-of-charge estimation accuracy.

키워드