Boiler combustion state anomaly early warning method based on graph neural network
By constructing a boiler combustion state anomaly detection method based on graph neural networks, a spatiotemporal heterogeneous graph structure of boiler combustion is built. The graph is then sparsified and reconstructed using a graph sparse training model. This solves the problems of limited computing resources and difficulty in capturing multivariate coupled changes in existing technologies, and achieves efficient and real-time anomaly detection and accurate diagnosis, thereby improving the safety and timeliness of boiler operation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YUHENG POWER STATION OF SHAANXI HUADIAN YUHENG COAL POWER CO LTD
- Filing Date
- 2026-03-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for detecting abnormal boiler combustion conditions have limited computational resources and strict time delays in large-scale, strongly coupled boiler monitoring networks, making real-time operation difficult. Furthermore, traditional methods struggle to effectively capture dynamic coupling changes among multiple variables, lack the ability to accurately locate the root cause of anomalies, and suffer from insufficient system robustness. They often experience delayed, false, or missed reports, making it difficult to provide maintenance personnel with high-value fault tracing and decision support.
A graph neural network-based approach is adopted to construct a spatiotemporal heterogeneous graph structure for boiler combustion. Dynamic sparsification is performed using a Graph Sparse Training model. Pruning and growth are combined with structural sensitivity and temporal mutual information indices to generate a dynamic sparse graph structure. Adjacency relationships and temporal trajectories are reconstructed, node-level anomaly scores are calculated, and physical consistency calibration is performed to locate potential root cause nodes and root cause edges, thus forming anomaly detection results.
It enables efficient real-time detection of abnormal boiler combustion status under limited computing power, provides accurate and auditable anomaly diagnosis results, improves the real-time performance and throughput of detection, enhances the ability to process multivariate coupled information, and ensures the safety of boiler operation and the timeliness of early warning.
Smart Images
Figure CN122413201A_ABST