Research on chemical process fault diagnosis method based on multi-level spatiotemporal attention and double-layer bidirectional GRU

With the sustained advancement of the chemical industry, fault diagnosis assumes a vital role in ensuring the safety of chemical processes. With the growing complexity of chemical processes and the continuous increase in the number of pieces of equipment, process data exhibit characteristics of high dimensionality and strong temporal correlation, thereby rendering chemical process fault diagnosis confronted with severe challenges. However, traditional fault diagnosis methods are difficult to accurately capture key features and long-term dependencies when dealing with high-dimensional complex time-series data, and there are problems of insufficient diagnosis accuracy and poor robustness. Therefore, this paper proposes a new method for fault diagnosis for chemical processes, which combines multi-level feature extraction with spatiotemporal attention and a double-layer bidirectional gated recurrent unit. This method proposes a spatiotemporal attention mechanism that decouples the spatial and temporal dimensions of feature importance modeling to learn the importance of feature dimensions and time steps separately. This mechanism avoids coupling interference between spatiotemporal features and enhances the model’s ability to capture key spatiotemporal features. Bidirectional gated recurrent unit sequences are introduced to capture past and future context information and to abstract multi-level and multi-dimensional temporal features by increasing network depth. The effectiveness of the proposed method is validated via simulation experiments conducted on the Tennessee Eastman dataset and the continuous stirred tank reactor.