The early warning method for offshore wind turbine gearbox oil temperature based on FSTAE-ATT

Abstract

Offshore wind turbine gearboxes often experience malfunctions due to harsh environmental conditions, resulting in significant downtime and financial losses. This study presents an innovative early warning system for monitoring gearbox oil temperature using a novel FSTAE-ATT model. The system leverages SCADA data and employs Feature Mode Decomposition (FMD) to enhance feature extraction from gearbox oil temperature measurements. The FSTAE-ATT model integrates Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal dependencies, augmented by a self-attention mechanism to highlight critical features. The model's reconstruction error serves as an early warning indicator for gearbox oil temperature anomalies. The effectiveness of the FSTAE-ATT model was validated using real-world data from an offshore wind farm in Yangjiang, Guangdong, China. Comparative analysis with other models, including STAE, STAE-ATT, AE, TAE, and SAE, demonstrated that the FSTAE-ATT model outperforms them with lower RMSE (e.g., 0.003452 for unit #40) and MAE (e.g., 0.002828 for unit #40) metrics. Additionally, significantly earlier warning times (e.g., up to 22 h and 36 min for unit #40), provide substantial lead time for preventative maintenance. This work contributes to advancing offshore wind turbine condition monitoring and fault detection, enhancing the sustainability and profitability of offshore wind energy systems.