Spatio-temporal degradation model with graph neural network and structured state space model for remaining useful life prediction

The advent of the big data era has driven the continuous advancement of deep learning techniques, which have recently achieved significant success in the field of remaining useful life (RUL) prediction. However, existing deep learning models often struggle to effectively capture both the long-term temporal dependencies and spatial dependencies in multi-sensor condition monitoring data. To address the above issues, this study proposes a spatio-temporal degradation model named spatio-temporal graph structured state space sequence model (ST-GS4D) to effectively extract degradation features for RUL prediction. The ST-GS4D is an innovative prediction framework that integrates structured state space sequence model with a diagonal state matrix (S4D) and graph convolutional network (GCN). The S4D is utilized to capture long-term temporal dependencies, while the GCN is employed to extract spatial dependencies, within multi-sensor time series data. To validate the capability of ST-GS4D in effectively capturing spatio-temporal dependencies from multi-sensor time series data for degradation feature extraction, two benchmark experiments are conducted on the C-MAPSS dataset and the dust filter dataset respectively. The experimental results indicate that the ST-GS4D model can effectively capture the degradation information from the multi-sensor time series data for accurate RUL prediction and outperform the existing methods.