Akademik Veri Yönetim Sistemi
Tezin Türü: Doktora
Tezin Yürütüldüğü Kurum: Gazi Üniversitesi, Fen Bilimleri Enstitüsü, Türkiye
Tezin Onay Tarihi: 2020
Tezin Dili: Türkçe
Öğrenci: MEDİNE ÇOLAK
Danışman: RAMAZAN BAYINDIR
Özet:Since solar radiation intensity and photovoltaic power generation data are indispensable inputs for photovoltaic energy systems, high accuracy and consistent solar radiation and photovoltaic power prediction are essential requirements in practice. In this thesis study, solar parameters have been modeled by experimental methods and metaheuristic optimization algorithms with high ability to find global optimum have been hybridized to artificial neural networks. In the process of creating hybrid estimation models, the grey wolf, the ant lion and the whale optimization algorithms were integrated into the multilayer perceptron algorithm. Air temperature, relative humidity and diffuse horizontal solar radiation parameters were used in the structure with 3 inputs and 2 inputs to estimate the daily total horizontal solar radiation. In order to estimate the daily photovoltaic power generation, air temperature, relative humidity, total horizontal solar radiation and diffuse horizontal solar radiation parameters were used in the structure with 4 inputs, 3 inputs and 2 inputs. In addition, the performances of the developed hybrid estimation models were tested in terms of the hyperbolic tangent, sinus and sigmoid activation functions used in the multilayer perceptron algorithm. The comparison of mean absolute error, mean absolute percentage error and root mean squared error performances of the developed estimation models shows that the grey wolf optimization algorithm-based multilayer perceptron model provides the best results for estimating the daily total solar radiation and the daily photovoltaic power generation. In addition, it is considered that the developed hybrid estimation models can successfully be applied on the datasets containing different time intervals to be used in the future.