A Data-Driven Framework for Battery Capacity Estimation in Real-World Electric Vehicles Using Virtual Impedance and Incremental Capacity Analysis

Accurate battery capacity estimation is critical for ensuring the safe and reliable operation of electric vehicles (EVs) and addressing user range anxiety. However, predicting battery health is challenging due to the non-linearity, non-measurability, and complex multi-stress operating conditions that characterize battery performance. Incremental capacity curves and electrochemical impedance spectroscopy (EIS) are effective tools for reflecting battery aging, but their practical application has limitations. This paper presents a novel method for battery capacity estimation using charging segment data derived from real-world operating conditions monitored by the vehicle's Battery Management System (BMS). The proposed approach begins with a detailed statistical analysis of voltage data to determine optimal charging capacity intervals and involves selecting appropriate voltage ranges to compute equivalent full-charge capacities. Experimental tests are performed to measure battery charging capacities across various temperatures, and temperature corrections are applied to ensure accurate capacity labeling. Next, several virtual impedances at low frequencies are calculated and the peak and valley values of the incremental capacity (IC) curve are identified. These derived features are then utilized to train a Transformer model for battery capacity estimation. To enhance the model's adaptability, transfer learning techniques are applied, allowing the model to be effectively used across different vehicle types. Experimental results demonstrate that this approach substantially improves the accuracy of battery capacity estimation. By providing a more precise understanding of battery health, this approach contributes to enhanced EV performance and user experience, addressing key challenges related to battery management and range anxiety.