Resumo:
In recent years, renewable and sustainable energy sources have attracted the attention of
various investors and stakeholders, such as energy sector players and consumers. Electric
power systems have experienced the rapid insertion of distributed renewable generating
sources and, as a result, face planning and operational challenges as new connections
are made to the grid. It is very difficult to observe and anticipate the required levels of
photovoltaic generation, which are tasks considered inherent to a quick insertion into the
electrical grid. This distributed/renewable generation must be integrated in a coordinated
way, so that there is no negative impact on the electrical performance of the grid, increas ing the complexity of energy management. In this work, a multivariate strategy, based on
design of experiments (DOE), is addressed for the prediction of photovoltaic generation
using a new approach for parameterization and combination of a set of artificial neural
networks (ANN). Two main questions will be explored: how to select the ANNs and how
to combine them in the forecast by sets (ensemble). As a complement to this methodology,
the reduction of dimensionality of climate data through Principal Component Analysis
(PCA) is also presented. The design of experiments (DOE) approach is applied to the
PV generation time series factors and to the ANN factors. Then, a cluster analysis is
performed to select the networks that obtained the best results. From this point, a mixture
analysis (MDE) is used to determine the ideal weights for the formation of the ensemble.
The methodology is detailed throughout the work and, based on the combination of fore casts, the photovoltaic generation was estimated for a set of specific panels, located in the
south of the State of Minas Gerais. Therefore, a more comprehensive study, which con sidered a dataset of seventeen generation plants, with seasonal characteristics, was also
examined. The versatility of the proposed method allowed changing the number of factors
to be used in the experimental arrangement, in the forecasting model and in the desired
forecasting horizon and, consequently, improving the determination of the forecast for the
studied scenarios.
