Unveiling MISFN: A Multiattribute Information Segmentation and Fusion Network for Advanced Wind Turbine Anomaly Monitoring

Abstract

The increasing demand for reliable wind turbine performance has highlighted the critical need for advanced anomaly monitoring systems. Existing strategies are often found to struggle with the efficient extraction and fusion of multiattribute data, which are essential for ensuring the secure operation of wind turbines. In response, a novel approach, multi-attribute information segmentation and fusion network (MISFN), is proposed to enhance anomaly monitoring through spatiotemporal feature extraction from supervisory control and data acquisition (SCADA) systems. In this network, information segmentation convolution (ISC) is proposed to direct the flow of multiattribute SCADA data, enabling effective information diversion while also extracting spatiotemporal features and captures long-range dependencies. These spatiotemporal features are then refined by the Transformer, which further enhances the model’s ability to detect anomalies in the data. Explained variance score (EVS) is developed as an evaluation metric to assess the deviation between predicted and actual system states, enabling early and accurate anomaly detection. MISFN was validated using SCADA data from two real-world wind farms, where anomalous states were successfully monitored with high reliability and precision. Through comparative experiments, the superiority of this model over traditional approaches was demonstrated, confirming its effectiveness for real-time wind turbine anomaly monitoring. This work is considered a significant advance in the integration of multi-attribute data fusion and spatiotemporal feature extraction, paving the way for more intelligent and proactive wind turbine monitoring solutions.