Abstract :
The degree of complexity and the large volume of information normally associated with electrical distribution networks pose the most significant challenges for the engineers and/or operators dealing with such networks. Yet, despite the complexity of these systems, similar patterns of behavior can be identified among the same type of equipment, thus allowing a significant reduction in the number of individuals to be analyzed. Such reduction is made possible by the use of classification techniques aimed at classifying or grouping similar individuals in a single group (category or family). This paper presents and discusses the application of classification techniques to solve specific problems in distribution systems. Their common feature is the need to represent large groups of individuals, in a reduced way identifying their typical patterns or elements. The first one is to set patterns for low voltage circuits based on a sample obtained from field data collection. Each LV circuit was represented by a set of attributes, considering that some of them, as circuit configuration, length and conductors were known only for the selected sample. The other circuits, about 50000 of them, were then classified, according to the previously established patterns. Two classification techniques were used: artificial neural networks and hierarchical classification, as part of an R&D project for RGE1, to assess technical energy losses The second problem was to identify, among a large group of distribution transformers, those groups or categories sharing similar features regarding both the physical aspects and the load characteristics. A statistical classification technique — cluster analysis — was used with data obtained from the set of 35000 transformers as well as from the load curve measurements of a sample of 140 transformers. The results of this classification process were used in a study made for AES Su1l. This research aimed at esta- lishing the optimal transformer load and loss of life. The third problem was to characterize the load in a utility´s concession area. The data obtained from load curve measurements, taken from samples representing each type of consumer, stratified by level of consumption or demand, per level of voltage, render the typical curves and their market percentage. The process is divided in steps utilizing the techniques of hierarchical classification and cluster analysis. The methodology was used in the tariff review process of COPEL1 and CELESC1.
