A deep integration model of temporal and spatial information for condition monitoring of steam turbine generator sets

Steam turbine generator sets are critical equipment in thermal power generation, and their operational status is typically monitored using a distributed control system (DCS). The DCS primarily relies on threshold values to identify faults. However, under complex and multivariate conditions, it lacks effective mechanisms for early warning and fault detection. A new method based on a bidirectional gated recurrent unit (BiGRU) and vision Transformer (ViT)for condition monitoring and fault diagnosis of the steam turbine generator set is proposed in this paper. BiGRU extracts local high-level temporal features from a time-series perspective, and its bidirectional structure captures bidirectional contextual information of the sequence. ViT effectively extracts complex global spatial features and captures global information through its self-attention mechanism. This approach facilitates stable and accurate condition monitoring and fault prediction, significantly enhancing the reliability of early warning systems. Kolmogorov-Smirnov test is employed to validate the faults predicted by the model, comparing the KS value and probability density distribution to reduce false alarms and misreports. The Kolmogorov-Smirnov test effectively verifies the faults, indicating that the new method proposed in this paper can be successfully applied to condition monitoring of steam turbine generator sets in thermal power plants .