Remaining useful life prediction of lithium-ion batteries based on TCN-DCN fusion model combined with IRRS filtering

Lithium batteries are widely used in various applications such as electronic products, power generation and energy storage. Accurately predicting the remaining useful life of lithium batteries is critical to improving the reliability of energy systems. However, current deep learning-based prediction methods tend to involve complex models and fail to effectively uncover potential relationships among data. To overcome this limitation, a fusion TCN-DCN data-driven model is proposed in this study. First, degradation features with high correlation were extracted from the charging and discharging processes of lithium batteries. Then, an improved robust smoothing regression method was used to deal with the noise in the data by smoothing and denoising. A small number of samples were then used to pretrain the TCN model to optimize its parameters, and the trained TCN were integrated into the deep network of the DCN model to construct the TCN-DCN model. Finally, existing methods were compared with proposed method, which demonstrated superior performance in terms of both prediction accuracy and computational speed.