Avaliação de desempenho de uma usina solar fotovoltaica com base no agrupamento k-means

Abstract

In 2021, Brazil reached, according to the Brazilian Association of Photovoltaic Solar Energy (ABSOLAR), the mark of 10 GW of operational power in large-scale solar photovoltaic plants, as well as in small and medium-sized systems, corresponding to a value above 70\% of the power of the Itaipu hydroelectric plant. With that, the country is the only Latin American in the Top 15 of the world ranking of installed solar energy capacity prepared by the International Renewable Energy Agency (IRENA). For this reason, it is important to analyze the use of energy resources, seeking to understand the behavior and performance of photovoltaic systems, identifying possible irregularities in the work. Therefore, in this work, a data mining method will be implemented, applied in data clustering, known as K-means. This method aims to group 36 transformer stations, belonging to 4 photovoltaic plants, in different sets, in such a way that the components of the same group have similar characteristics. The K-means clustering algorithm was developed in Python programming language, using the inverter power and solar irradiance variables as data input. The input data were extracted from the Supervisory Control and Data Acquisition (SCADA) and discretized at 60-minute intervals. This methodology allows analyzing the performance of systems through groups, rather than individual analyses, easing the effort to monitor the operation of photovoltaic plants with a large number of inverters. The results obtained demonstrate that inverters with similar specifications, arranged in the same photovoltaic system, can present divergent behaviors. In groups where the performance of inverters is inferior to the others, it is necessary to carry out a study to understand the possible reasons that resulted in such deficiencies.