Remaining useful life prediction of lithium-ion batteries based on an incremental internal resistance aging model and a gated recurrent unit neural network

Accurate and interpretable remaining useful life (RUL) prediction of lithium-ion batteries (LIBs) is critical for operational safety and system optimization. However, existing data-driven approaches frequently suffer from inadequate physical interpretability and diminished generalization capability under diverse aging scenarios. To address these challenges, this study proposes a hybrid RUL prediction framework by integrating an incremental internal resistance aging model (IIRAM) with a gated recurrent unit (GRU) neural network. A state of charge (SOC) normalization strategy is established to project charge/discharge voltage profiles onto a unified scale. Subsequently, an analytical IIRAM comprising three sub-models—voltage, incremental internal resistance (IIR), and incremental capacity (IC)—is constructed. A differential modeling approach is employed to formulate an analytical voltage model, from which the mathematical models for IIR and IC are derived to quantitatively characterize degradation behavior of LIBs. Eight health indicators are extracted from the IIRAM and input into the GRU network, thereby constructing the IIRAM-GRU hybrid model for RUL forecasting. Experimental validation on the NASA and CALCE battery datasets demonstrate that the proposed method consistently outperforms conventional deep learning models, achieving enhanced predictive accuracy, improved robustness, and augmented physical interpretability, thereby providing a reliable and practical solution for real-world battery prognostics. • A SOC normalization strategy using percentage of charge (POC) is introduced. • An IIRAM is constructed to characterize the aging of IIR, IC, and voltage. • A differential modeling method is proposed to develop an analytical voltage model. • Analytical models for IIR and IC are derived from the voltage model. • Eight HIs from IIRAM are input into a GRU for accurate capacity and RUL prediction.