Corrected State-of-Health Representation for Lithium-Ion Battery Degradation Forecasting via Model-Parameter Identification and Multisource Health Indicator Fusion

Abstract

Accurate state-of-health (SOH) assessment and degradation trend forecasting are essential for lithium-ion battery safety, maintenance, and lifetime management. In practical applications, SOH estimation remains largely capacity-based and relies on current, voltage, temperature, and capacity-related sensing data acquired by battery management systems. Although capacity is a key aging indicator, it cannot fully characterize battery degradation because impedance growth, polarization variation, thermal inconsistency, and cell-to-cell voltage dispersion are strongly coupled under complex operating conditions. Therefore, SOH estimation should incorporate multidimensional sensor-derived degradation information to more accurately capture battery health evolution. To address this issue, this study proposes a corrected SOH representation and degradation trend forecasting framework based on multisource health indicator fusion and hierarchical prediction. First, capacity-based SOH is adopted as the benchmark; impedance- and polarization-related parameters are identified using a first-order Thevenin model and forgetting-factor recursive least squares (FFRLS), and temperature and voltage consistency indicators are extracted from selected state-of-charge (SOC) intervals. Next, gray relational analysis (GRA) selects SOH-sensitive core indicators, and G1-CRITIC combined weighting constructs a fused health indicator, which is nonlinearly mapped to generate the corrected SOH. Furthermore, a fungal growth optimization (FGO)-DATCN-GPR model integrates FGO, a dual-attention temporal convolutional network, and Gaussian process regression for hyperparameter optimization, global trend learning, and local residual compensation, respectively. Validation using datasets from 20 electric vehicles and a commercial energy-storage cabinet demonstrates favorable overall forecasting accuracy and cross-scenario competitiveness against expanded benchmark and ablation models, indicating the value of multichannel sensing information for battery health monitoring and degradation forecasting.