A novel MIC-BOA-TiDE fusion framework with kernel density estimation for point and probabilistic remaining useful life prediction of lithium-ion batteries

Accurate prediction of battery remaining useful life (RUL) is crucial for enhancing equipment safety and reliability as well as for sustainable environmental development. This paper proposes a lithium-ion battery RUL prediction model that combines the maximum information coefficient (MIC), the Bayesian optimization algorithm (BOA), kernel density estimation (KDE), and the time-series dense encoder (TiDE). First, using NASA's publicly available lithium-ion battery cycle-life data and the CAL-CE dataset, multiple health indicators including constant-current charge time, discharge time, and IC-curve peak values are extracted and selected via MIC. Next, the TiDE model is employed for accurate RUL prediction, with its key hyperparameters optimized by BOA to boost predictive performance. Finally, KDE is adopted to produce probabilistic RUL forecasts and construct confidence intervals that quantify prediction uncertainty, thereby refining the overall assessment. Comparative experiments demonstrate that the MIC-BOA-TiDE framework reduces the capacity-forecast MAE by 68.9% versus a back-propagation network and by 46.0% versus a GRU baseline, while its RUL error converges to virtually zero, underscoring its superior accuracy and stability. Additionally, the KDE-based interval prediction results show that at the 95% confidence level, the coverage probability on the CS2_35 dataset reaches 95.27% with an average interval width of 0.0731, confirming the model's effectiveness in quantifying predictive uncertainty.