Aplicação de técnicas de estatística multivariada para construção de modelo de seleção de módulos fotovoltaicos

Abstract

The objective of the following monograph is to propose the clustering of photovoltaic modules found in the Brazilian market, simplifying the comparability between these kinds of equipment considering the consumer point of view and allowing manufacturers to benchmark their products. To achieve this goal, clustering methods were applied, such as single-linkage, complete-linkage, average-linkage and Ward, with the intent of estimating the number of groups composed of 79 photovoltaic modules and considering 7 kinds of characteristics, being six of them indicatives of technical aspects of the modules and the seventh characteristic (the price) an indication of economic and financial perspective. Once estimated the number of groups following the aforementioned methods, the non-hierarchical clustering method of k-means is applied, in order to validate the clustering results comparatively from the best results obtained from the tested hierarchical clustering methods. In this monograph, it has been found that the clustering composed of 5 groups presents adequate results, especially when applied the k-means method. It has also been concluded that it is possible to cluster at least 4 of the 7 analyzed variables, with little loss of information quality (the variables peak power, average energy generation, area and efficiency can be replaced for a single variable – the efficiency was chosen between theses variables because of its emblematic nature). This reduction on the number of analyzed variables also goes in agreement with the objective of allowing the simplification of comparisons between photovoltaic modules.