Attention Mechanism in Deep Learning for Wind Power Forecasting

Wind energy is the fast moving renewable energy sources in the world. However, wind energy deals with many challenges namely seed capital, motionless property of wind plants and most difficult to track down economical areas. To deal with these challenges, this study proposes a fusion of attention mechanism with deep-learning models for wind power forecasting. Attention mechanism is essentially a way to non-uniformly weight the contribution of input feature vectors so, as to optimize the process of learning targets. Several models have been considered to evaluate the performance. Convolutional Neural Network (CNN) helps in extracting short term attributes to obtain high-dimensional attributes, whereas issue of inaccurate prediction imputable to merging of original data can be solved by Gated Recurrent Unit (GRU). GRU helps in extracting long-term trend of high dimensional attribute. Using original data of wind power, we verify that attention mechanism with deep learning models provide better performance for wind power forecasting.