ScholarWorks@경상국립대학교

초록

Wind energy, as a key aspect of renewable energy, is drawing growing attention for its potential to lower greenhouse gas emissions and provide a clean energy alternative. Nonetheless, the inherent variability of wind energy necessitates ensuring the reliable operation of wind turbine systems and reducing downtime caused by failures. This study introduces a novel fault detection approach for wind turbine systems based on a combination of Bidirectional Long Short-Term Memory (Bi-LSTM) networks and an Attention Mechanism, addressing the limitations of traditional diagnostic methods. By utilizing operational parameters gathered from SCADA as input data, our proposed model captures temporal dependencies in both directions while focusing on the most relevant features for fault classification. Extensive experimental results using real-world SCADA data from an offshore wind farm demonstrate that our proposed model achieves superior fault detection performance with an accuracy of 99.40%, outperforming conventional machine learning methods including Random Forest (87.50%), AdaBoost (89.00%), and Multi-Layer Perceptron (91.34%). Moreover, the model maintains computational efficiency with an average detection time of only 0.0008 s, making it suitable for real-time monitoring applications. The results confirm the effectiveness of our approach in enhancing wind turbine reliability and reducing maintenance costs through early and accurate fault detection.

키워드