A WIND TURBINE POWER FORECASTING METHOD BASED ON MTGP TRANSFER LEARNING

Wind turbine power forecasting plays an increasingly important role in the safety, reliability, and stability of both the power grid and the electricity market. Owing to the fluctuating and intermittent nature of the wind resource, it is very challenging to accurately forecast the wind power for either a newly constructed or existing farm. Most existing related research has not considered the data uncertainty and insufficiency in wind power forecasting so that the yielded model cannot produce the desirable result in practice. This paper presents a BDWPT-MTGP hybrid approach to accurately forecast the wind power in the scenario of data insufficiency. The Bayesian Discrete Wavelet Packet Transform (BDWPT) denoising approach is employed to improve wind power forecasting accuracy with data imperfection. A multi-task Gaussian process (MTGP) transfer learning model is developed for power forecasting of wind turbines based on limited operating data, thus addressing the data insufficiency issue. AMTGP-model is constructed by leveraging available wind farm data alongside limited power data. The proposed BDWPT-MTGP hybrid method is validated by using the operating data acquired from real-world wind turbines. Numerical results demonstrate that the proposed methodology provides a promising tool for accurately forecasting wind power with data uncertainty and insufficiency.