An unsupervised spatiotemporal graphical modeling approach for wind turbine condition monitoring. (November 2018)

• Record Type: Journal Article
• Authors: Yang, Wenguang; Liu, Chao; Jiang, Dongxiang
• Is Part Of: Renewable energy. Volume 127(2018)
• Page Start: 230
• Page End: 241
• Publication Date: 2018-11
• Subjects: Wind turbine -- Unsupervised condition monitoring -- Spatiotemporal graphical modeling; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2018.04.059
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: The vast installment of wind turbines and the development of condition monitoring system provides large amounts of operational data for condition monitoring and health management, while the lack of labeled data becomes one of the major challenges for the data analytics. To address this issue, this work presents an unsupervised anomaly detection approach for wind turbine condition monitoring, where a spatiotemporal graphical modeling method, spatiotemporal pattern network (STPN), is applied to extract the spatial and temporal features between the variables in the system, and an energy-based model, stacked Restricted Boltzmann Machine (RBM) is used to capture the system-wide patterns and then applied for condition monitoring. Case studies on three data sets are carried out including: (1) anomaly detection on a benchmark model for fault detection and isolation, (2) anomaly detection on an experimental data set with the normal condition and 11 fault conditions and (3) online condition monitoring using real data from a wind farm in northwest China. The results show that the proposed approach is capable of detecting the anomalies without the need for labeling data. Highlights: Spatiotemporal pattern network (STPN) for wind turbine condition monitoring. Energy-based Restricted Boltzmann Machine (RBM) model for unsupervised learning. STPN for offline system behavior analysis and online feature extraction. STPN + RBM framework outperforms FastDTW algorithm on experimental data set. Case study on SCADA data of wind turbines validates the presented approach.