Feature Valuation Toward Improved State Estimation for Automotive Lithium-Ion Battery

作者：Yanzhi Wang, Jianxiao Wang, Jie Song · 年份：2024 · DOI：10.1109/ias55788.2024.11023788 · 被引用次数：1 · 研究领域：Fault Detection and Control Systems、Advanced Battery Technologies Research、Advanced Algorithms and Applications

In the rapidly evolving fields of energy storage and big data, data-driven models for estimating battery states have become increasingly prevalent. However, the accuracy of these models is greatly affected by the quality of feature data used in training, especially in real-world lithium-ion battery scenarios where data diversity and quality significantly vary from laboratory settings. Our paper introduces a feature valuation framework tailored to data-driven predictions, focusing on evaluating the averaged marginal improvement of features on the model's performance. Experimental findings demonstrate a prediction error of 4.07 % with only Battery and Power System, HVAC System, and Temperature Control and Monitoring features, yielding a 0.8 % and 2.2 % accuracy enhancement over the full dataset and BPS alone, respectively.