Proceedings of the

The 33rd European Safety and Reliability Conference (ESREL 2023)
3 – 8 September 2023, Southampton, UK

A Data-Driven Failure Prediction Method for Offshore Wind Turbines Using Long Short-Term Memory Model

Heping Lia, Li Chenb and Kejia Yangc

School of Economics and Management, China University of Petroleum (Beijing), China.


As operating in harsh marine environment, the offshore wind turbines often lead to high failure rate which affect the efficiency and reliability of wind power generation significantly. To improve the power generation capacity and decrease the breakdown time, this paper proposes an early failure prediction model for offshore wind turbines using Long Short-Term Memory (LSTM) model. With the SCADA data, a main feature is distinguished by the coefficient analysis to each failure mode. LSTM is used to capture the representation between the main feature and other relative features in the normal operation. When a failure occurs, the consistency of the representation should change dramatically. Therefore, a rule is set to distinguish the pattern between the normal operation and the failures using residual value indicator. With the SCADA data set of an offshore wind farm provided by EDP, it is proved that the algorithm can warn the hydraulic group, bearing and transformer failure about 31 hours, 5 hours and 15 hours in advance respectively.

Keywords: Failure prediction, Offshore wind turbines, Long Short-Term Memory (LSTM).

Download PDF