A Novel Combined Model for Condition Monitoring of Steam Turbine-Generator Sets

State monitoring is an indispensable part of thermal power generation, providing early warnings of potential faults and preventing severe accidents, thereby ensuring production safety and economic efficiency. The Distributed Control System (DCS) plays a pivotal role in thermal power plants, acting as the central nervous system that connects various devices and systems, enabling comprehensive integration from data acquisition and monitoring to control, and ensuring the efficient and safe operation of power plants. The steam turbine generator is a critical component in thermal power generation, and monitoring its operational status is a crucial task. This study proposes a BiGRU-ViT model, which combines Bidirectional Gated Recurrent Unit (BiGRU) and Vision Transformer (ViT). BiGRU extracts local features from the perspective of time series, capturing bidirectional contextual information due to its bidirectional structure. ViT, on the other hand, extracts spatial features, leveraging the self-attention mechanism to capture global information. The integration of these two models allows for a broader perspective in predicting target parameters. Tests using actual DCS data demonstrate that the BiGRU-ViT model can provide early warnings of potential faults, verifying the model's effectiveness and suggesting its applicability in the state monitoring of steam turbine generators in thermal power plants.