Prediction of Generation in a Photovoltaic System through the application of Data Mining techniques

Abstract

This document presents a generation prediction model through data mining techniques for a photovoltaic plant located at Paragachi Community, belonging to Pimampiro (Imbabura), with a total of 14400 solar panels and 3.6 MW nominal power. This system does not have a battery bank for storage, for this reason, it does not provide energy at night, but during the day, it supplies the energy to 2000 households that represent Pimampiro’s urban population.

It begins with a univariate and multivariate analysis of the measurement variables, whose objective is to determine the behavior, incidence and the relationship of each variable in the generation of the photovoltaic system. With the variables of higher incidence as input, a learning machine is trained; it uses the technique of decision trees through random forest to predict the generation.

In renewable energies, the photovoltaic system is one of the most implemented and developed nowadays. However, predicting the amount of power it can generate is complicated by the stochastic behavior of the variables, limiting the entry of this technology into a competitive market, which can integrate into the National Interconnected System in an optimal and efficient way.