This paper proposed a training-based method for wind turbine signal forecasting. This proposed model employs a convolutional network, a long short-term memory network as well as a multi-task learning ideas within a signal frame. This method utilized the convolutional network for exploitation of spatial properties from wind field. As well, the mentioned long short-term memory is used for training dynamic features of the wind field. The ideas stated together have been utilized for modeling the impacts of spatio-dynamic construction of wind field on wind turbine responses of interest. So, we implemented this multi-task training method for forecasting the generated WT energy and demand at the same time through a single forecast method, which is the deep neural-network. Performance of our suggested model is confirmed by a real wind field information that is produced by Large Eddy Simulation. This data also include wind turbine reaction information that is simulated using aero-elastic wind turbine construction analyzing software. The obtained results depict that the suggested method can forecast two outputs with a five-percent error by a so short term prediction, which is shorter than 1 m.

Item Type:	Refereed Article
Keywords:	wind signal, forecasting, long short term memory network, multi task learning, deep neural networks
Research Division:	Engineering
Research Group:	Electrical engineering
Research Field:	Electrical energy generation (incl. renewables, excl. photovoltaics)
Objective Division:	Energy
Objective Group:	Renewable energy
Objective Field:	Wind energy
UTAS Author:	Lee, B (Mr Brendan Lee)
ID Code:	135048
Year Published:	2019
Web of Science® Times Cited:	67
Deposited By:	Engineering
Deposited On:	2019-09-24
Last Modified:	2020-01-14
Downloads:	0