Cascaded Modeling of Battery SOH and Dynamic Degradation Trajectory Using Segmented High-SOC Charging Data

Real-time estimation of the state of health (SOH) and prediction of the degradation trajectory are critical for ensuring the safety of lithium-ion batteries in consumer electronics and electric vehicles. However, existing approaches often decouple these two tasks and rely on long-term continuous data, which conflicts with the fragmented charging behaviors prevalent in real-world applications. To address these limitations, this paper proposes a cascaded modeling framework for concurrent SOH estimation and degradation trajectory prediction using only segmented data. First, features are extracted from accessible high-state-of-charge (SOC) charging segments. A hybrid feature screening strategy is then employed, combining correlation analysis with machine learning evaluation. Subsequently, a long short-term memory (LSTM) network is implemented for SOH estimation, followed by a convolutional neural network (CNN) for dynamic degradation trajectory prediction. A key innovation is the exploitation of intrinsic connections between the hidden-layer embeddings of the SOH estimator and underlying aging mechanisms. The entire output vector of the optimal dense layer is extracted as a health indicator to enhance trajectory prediction. Validation demonstrates SOH estimation with a root mean square error (RMSE) of 0.254% and an average trajectory prediction error of 31.82 cycles when using the predicted SOH. These results confirm that the cascaded modeling framework provides a robust, integrated solution for battery management in consumer applications.