Banca de DEFESA: ANTÔNIO FABRÍCIO GUIMARÃES DE SOUSA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : ANTÔNIO FABRÍCIO GUIMARÃES DE SOUSA
DATA : 12/02/2019
HORA: 08:30
LOCAL: Auditório NTB - PROPPIT
TÍTULO:

Artificial Neural Networks and Prediction Intervals for solar irradiance forecasting

PALAVRAS-CHAVES:

Solar energy, Solar irradiance forecasting, Artificial Neural Network, Particle
Swarm Optimization, Prediction Intervals.

PÁGINAS: 67
GRANDE ÁREA: Outra
ÁREA: Ciências Ambientais
RESUMO:

Solar energy is a clean renewable source with an important role in the global energy supply.
An accurate knowledge on solar irradiance prediction is particularly required for proper
development and planning of photovoltaic (PV) energy systems. The present work has the
objective to implement algorithms for accurately predicting solar irradiance, both in
conventional Artificial Neural Network (ANN) point forecasting and in Prediction Intervals
(PIs) outcomes, providing a further comparison between them in order to obtain the best one.
These algorithms are based on the models: Multilayer Perceptron network (MLP), Elman
neural network (ELMAN), Nonlinear Auto-Regressive network (NAR), Nonlinear AutoRegressive network with exogenous inputs (NARX) and Lower Upper Bound Estimation
(LUBE), this last one is trained by Particle Swarm Optimization (PSO) for PIs estimation.
Meteorological data collected from a station in Amazon region in Brazil have been used to
train and validate the models. The results demonstrated that all ANN models yield good
accuracy in terms of prediction error, less than 10% for normalized root mean square error
(nRMSE) and normalized mean absolute error (NMAE), and higher than 0.90 for
determination coefficient (R²), in 7 of 9 prediction models results. And the estimated average
scenario obtained in the PI forecasting also showed reliable results, similar to the ANN
models and without significant difference from them, according to statistical tests. The
created PIs achieved Coverage Probability (CP) higher than 94% and a relative small interval
width of 32.5%, in relation to the real data and the forecasted results made by the others ANN
models, obtaining different scenarios prediction of the local solar irradiance simultaneously.
Demonstrated results indicate that PI estimation by the PSO-based LUBE method is very
efficient in for solar irradiance forecasting, and proved as a powerful tool to be used in PV
systems projects.

MEMBROS DA BANCA:
Presidente - 1963026 - ANDERSON ALVARENGA DE MOURA MENESES
Externo ao Programa - 1646831 - FABIO MANOEL FRANCA LOBATO
Externo ao Programa - 1917410 - JOSE ROBERTO BRANCO RAMOS FILHO
Externo ao Programa - 1835583 - RODOLFO MADURO ALMEIDA