A comparative analysis of neural and fuzzy cluster techniques applied to the characterization of electric load in substations

The hourly electric load behavior is necessary information to execute planning and operation activities of power systems. This information can be obtained through load characterization consisting of measurement and analysis of several load profiles aiming to find the typical profile that reflects consumer's behavior served by a certain utility or even the profiles of the consumers served by a specific substation. In this work, six techniques were used to perform clustering and classification of bus load profile in utilities substations: K-means, four variations of self organizing maps and fuzzy C-means. Many simulations with different parameters were used to improve clustering performance. Merit indexes of clustering performance (intracluster and intercluster) were used to compare these techniques. These indexes were not sufficient to guarantee satisfactory results but indicate reasonable performance. The combination of all techniques helps decision maker to understand results and increase the possibility to extract from data implicit information and potentially useful.