A Self-Organizing Map Approach for Time-of-Use Feature Based Wind Resource Clustering

The variable nature of wind and solar photovoltaic renewable energy sources, especially under high grid penetration levels, impacts adversely on the electrical grid operations. These impacts can include increased use of costly peaking generation as well as increased ramping of conventional generation. The operational challenges posed by variable renewable energy can be mitigated by optimal geographic siting of the renewable energy plants. Such siting optimization studies typically use large, high-resolution temporospatial datasets that are difficult to interpret and present great computational burdens. This paper investigates a data-reduction methodology using a self-organizing map approach for the clustering of temporospatial wind speed profiles with the view to reduce the dimensionality of the dataset. The clustering algorithm is implemented using the statistical means and standard deviations of the associated wind speed profiles for pre-defined time-of-use periods. The results obtained with the self-organizing map approach are compared with results obtained using the k-means and k-medoids algorithms. Overall, the results for the three clustering methodologies show good agreement. The clustered resource maps obtained with time-of-use statistical features are of value for siting and interpreting wind generation capacity such that the aggregated power generation profile is optimized in terms of time-of-use grid support.