HyPELS: enhancing li-ion battery remaining useful life prediction with hybrid perspective ensemble learning strategy

Remaining useful life (RUL) prediction for Li-ion batteries typically relies on a single perspective, which leads to limited applicability and reduced prediction accuracy. To address the limitations of traditional methods, a hybrid perspective ensemble learning strategy (HyPELS)that integrates HIs and capacity degradation data is proposed. First, for the perspective of battery health indicators(HIs), the block-shuffled OS-ELM (BSOS-ELM)is proposed, which mitigates the issue of early data characteristics being overlooked when applying OS-ELM. Second, for the perspective of capacity degradation data, after decomposition using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), the obtained high-frequency signals are reconstructed using two reconstruction rules, which accounts for both local and global signal characteristics. Subsequently, during the training of the autoregressive integrated moving average model (ARIMA), we feed early capacity degradation data in reverse order, maximizing the utility of data while deepening the models understanding of the overall capacity degradation process. Finally, the predicted capacity obtained from both perspectives is constructed into a meta-dataset, utilizing BSOS-ELM as the meta-model for ensemble learning. HyPELS not only captures local variations but also learns the overall trend, fully leveraging the strengths of both perspectives to capture the multi-dimensional characteristics of battery aging comprehensively. Experimental results demonstrate that across multiple datasets, HyPELS outperforms other algorithms in various assessment criteria, fully illustrating the effectiveness of HyPELS and validating its advantages in enhancing prediction accuracy and robustness.