Early Warning and Identification of Wind Turbine Faults Based on Temporal Network Flow Entropy

Abstract

Fault warning and identification are crucial for preventing accidents, enhancing performance and ensuring the reliability of wind turbine (WT). While SCADA systems are widely used for WT fault monitoring, they face inherent challenges including a lack of historical fault data and poor disturbance resistance. To address these issues, we introduces a novel data-driven method for early fault detection in WT, termed temporal network flow entropy (TNFE). In this approach, WTs are mapped into a multi-node complex network based on the correspondence between their internal components and monitored variables by the SCADA system. TNFE then quantifies changes in the inter-variable correlation information through network entropy, enabling early fault warnings by tracking the system’s information flow. Dominant variable nodes are identified from the network under different conditions, and dynamic network markers (DNMs) are constructed to pinpoint WT defects. Comparative studies show that TNFE outperforms other data-driven approaches in diagnostic accuracy and robustness against random noise. Case studies confirm that the TNFE method not only provides timely warnings of early WT defects but also effectively identifies defect types.

Keywords

• Wind turbine
• SCADA data
• network flow entropy
• dynamical network marker (DNM)
• fault diagnosis