Fuzzy Time Series Methods Applied to Short -Term Photovoltaic Power Forecasting Forecasting

Trabalho de Conclusão de Curso apresentado ao Instituto Latino-Americano de Tecnologia, Infraestrutura e Território da Universidade Federal da Integração Latino-Americana, como requisito parcial à obtenção do título de Bacharel em Engenharia de Energia.Abstract— Solar photovoltaic energy has shown a significant growth in the last decade. In the face of this growth, there are challenges to consider for the high penetration rates of solar photovoltaic, since this type of energy generation is variable. Photovoltaic generation forecasting is one of the alternatives to face such challenges. In this work we propose the realization of indirect forecasts of photovoltaic generation using Fuzzy Time Series. Considering short-term horizon predictions, two Fuzzy Time Series methods are used and evaluated to obtain a Global Horizontal Irradiance value. The first order called Multivariate WEIGHTED-FTS method applies linear chronological weights and, the second is FIG-FTS, Fuzzy Information Granular method, a higher order multivariate method which works as a wrapper that transforms real multivariate time series values into fuzzy univariate time series. The Global Horizontal Irradiance values obtained from both methods, were submitted to the spatial smoothing process, obtaining spatial irradiance on which a first-order low pass filter was applied considering the physical parameters of the photovoltaic simulated power system. This proposed indirect forecasting of photovoltaic generation was statistically evaluated and the results showed good model performance. Error statistics, such as RMSE and MBE, show that the higher order FIG-FTS method performs better than the WEIGHTED-FTS method in forecasting the GHI. Using the direct proportionality of the power of an FTS with the Global Horizontal Irradiance, spatial smoothing is performed by applying the low-pass filter considering the system area, resulting in a power simulation, also evaluated with statistical metrics. These results are analyzed and it is concluded that the application of the method is suitable for photovoltaic generation forecasting