Proposal for a KDD-based procedure to obtain a set of intelligent systems training applied to the identification of failures in hydroelectric power plants

This paper presents a procedure based on KDD (Knowledge Discovery Data), which allows the analysis of a data set to obtain structured information from the behavior of the system under specific conditions, such as system failure conditions at a hydroelectric power plant. By applying this procedure, the information obtained, it is structured in such a mode so that it can be used on the training of intelligent systems focused on fault diagnosis. The former procedure is necessary in the intelligent systems development stage because obtaining an effective training set requires extreme time and effort. The procedure was applied in the historical records of the Amaime hydroelectric power plant, located in Palmira, Valle del Cauca, Colombia, aiming to obtain patterns of behavior of the protection system which can be translated to different failures. This was possible by integrating a data mining technique such as hierarchical clustering and the statistical technique called the interpolation function. The main achievement of this work is to present a structured procedure that reduces the time to obtain a training set. In this specific case, the training set for mechanical failure of a hydroelectric power station was obtained, which can be used in the development of an intelligent system for failures diagnosis.