Ultra-Short-Term Platform Motion Prediction Method of Large Floating Wind Turbines Based on LSTM Network

Abstract

The motion prediction of large floating wind turbine platforms is the key technology to realize the control of active ballast systems and intelligent operation and maintenance monitoring. However, the complex environment of floating wind turbines makes ultra-short-term predictions that only rely on physical models and numerical simulation methods very challenging. Therefore, this paper proposes an innovative ultra-short-term prediction method for floating wind turbine platform motion based on the long-short-term memory (LSTM) neural network. Measured data have been used to verify the feasibility and uncertainty of this method in terms of surge motion. The results show that the ultra-short-term prediction method proposed in this paper can obtain a better accuracy. For example, the maximum mean square error of surge motion prediction in the 60 s under working condition is only about 1%. The ultra-short-term motion prediction of large floating wind turbines proposed in this paper provides solid technical support for future intelligent operation and maintenance of floating wind farms.

Keywords

• large floating wind turbines
• ultra-short-term prediction
• long-short-term memory (lstm) network
• uncertainty