A BiLSTM-FM-Res and Power Curve-Based State-of-Health Estimation Method for Lithium-Ion Batteries Using Partial Charging Segments

The state of charge (SOC) intervals during electric vehicle (EV) charging and discharging lack completeness, and variations in lithium-ion battery performance under different operating conditions increase the complexity of estimating battery state of health (SOH). Traditional battery health factor extraction methods, which rely on changes in charging voltage, often neglect information from the constant voltage (CV) charging phase, limiting their accuracy under random partial charging scenarios. To address this, this study integrates battery voltage and current signals during charging to derive charging power signals. Then, the area of the charging power curve is extracted through integral operations as a novel health indicator. Furthermore, this study proposes a bidirectional long short-term memory - feature mapping - residual connection (BiLSTM-FM-Res) framework based on a stacked deep learning architecture. This model first employs bidirectional long short-term memory (BiLSTM) to extract temporal features from sequential data. Feature mapping is subsequently applied to enable deep mining of critical patterns while suppressing noise interference in SOH estimation. Finally, residual connections and information fusion techniques are implemented to achieve multi-source feature integration, ensuring maximal utilization of operational data. Experimental validation across three real-world battery datasets demonstrates the proposed SOH estimation framework. The Pearson coefficient between the power curve area and battery SOH degradation trends exceeds 0.998 across varying charging conditions, confirming the effectiveness of this health indicator. Evaluation metrics show root mean square error (RMSE) values below 1%, mean absolute error (MAE) below 0.8%, and coefficient of determination (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">R</i>²) exceeding 0.979 between estimated and true SOH values. These results validate the robustness, flexibility, and practicality of the proposed battery SOH estimation framework for real-world applications.