Comparative Analysis of Clustering Techniques for a Hybrid Model Implementation

Abstract

This research is oriented to compare the performance of two clustering methods in order to know what is the best one for archiving high quality hybrid models. For testing purposes, a real dataset has been obtained of a bio-climate house located in Sotavento Experimental Wind Farm, in Xermade (Lugo) in Galicia (Spain). Between several systems installed in the house, the thermal solar generation system has been the chosen one for studying its behaviour and experimenting with the clustering techniques. Two approaches have been utilized for establishing the quality of each clustering algorithm. The first one is based on typical error measurements implied in a regression procedure such as Multi Layer Perceptron. Whereas, the second one, is oriented to the utilization of three unsupervised learning metrics (Silhouette, Calinski-Harabasz and Davies-Bouldin).