WIND FARM CLUSTERING METHODS FOR POWER FORECASTING

This paper initiates the development of a hybrid model for wind farm output power forecasting based on spatiotemporal parameters of a studied site. The time-dependency of local wind patterns is addressed by developing three wind farm clustering, K-means, Agglomerative, and Density-Based Spatial Clustering of Applications with Noise. Clustered wind turbines obtain a more accurate representation for wind power forecasting. The results for this work will be later used to extract key features based on singular value decomposition (SVD) and build the forecast model. the emphasis in this paper is on the clustering method and not the forecasting algorithms. Hourly-wind data of an onshore wind farm in the US for one year are used for developing this model. The results will be further used in improving wind clustering algorithms, feature identification, and time-dependency analyses of short- to medium wind forecasting.