A Deep Learning Hybrid Model for Predicting Remaining Capacity of Lithium Ion Batteries

Accurate prediction of the remaining capacity of lithium-Ion batteries plays an extremely important role in practical applications. In response to the issues of nonlinear aging trajectories during battery cycling and insufficient accuracy in remaining capacity prediction, we have introduced an innovative deep learning model. We integrate the Conditional Probability Optimization Algorithm with a deep learning model. The model incorporates Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Units (BIGRU), and Attention mechanisms. In the data preprocessing stage, we employ the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) to enhance data quality by decomposing and denoising the raw signals. Additionally, Crested Porcupine Optimizer(CPO) is introduced to perform global optimization of the model's hyperparameters, ensuring optimal performance and robustness. The experimental results demonstrate that our model significantly improves the prediction accuracy.