CEEMDAN–TCN–BiLSTM-based multiscenario prediction of lithium-ion battery state of health

Accurate state-of-health (SOH) prediction of lithium-ion batteries is challenging due to nonstationary degradation behavior and multistage aging characteristics. In this study, a hybrid CEEMDAN–temporal convolutional network (TCN)–bidirectional long short-term memory (BiLSTM) framework is adopted for battery SOH estimation. CEEMDAN is used to decompose capacity degradation data into multiscale components, alleviating nonstationarity, while a TCN and a BiLSTM network are employed to capture local temporal features and long-term degradation dependencies, respectively. The framework is evaluated using a publicly available battery aging dataset under multiple operating scenarios, including full charge–discharge cycles, partial charge–discharge conditions, noise-contaminated data, and different discharge current and cutoff voltage settings. Comparative results with benchmark data-driven models show improved prediction accuracy and stability across different scenarios, indicating that the adopted framework is effective for modeling complex battery degradation behavior.