Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Fen Bilimleri Enstitüsü, Türkiye
Tezin Onay Tarihi: 2019
Tezin Dili: İngilizce
Öğrenci: ZÜLFİYE EBRU ÖZTÜRK
Danışman: Özlem İlk Dağ
Özet:In recent years, many developments have been implemented by the players of the sector to provide sustainable energy flow in Turkey. One of them is the obligation of recording electricity interruption statistics. The Turkish energy regulatory compels new rules to local electricity distribution companies about recording their interruption statistics, including the reasons for electricity interruption, after 2003. However, all of the local distribution companies do not use the same standard to record these statistics. This situation causes complexities for decision makers and researchers for modeling electricity interruptions. In this study, we aimed to find appropriate longitudinal models for the dataset of electricity interruptions. However, the observed data in this study is discrete count type and most of them are zero. Markov Chain Monte Carlo Generalized Linear Mixed Models (abbreviated MCMCglmm), especially the type of zero-inflated and hurdle could be appropriate for these type of data. Therefore, Poisson, zero-inflated Poisson, and hurdle-Poisson distributed models were implemented to a real electricity interruption count dataset belonging to Çankırı in this study. The models have been implemented by using MCMCglmm package in R. To compare the models, Deviance Information Criteria (DIC) and posterior predictive checks were used. Geweke-Halfwidth and Heiderberger-Welch diagnostic tests were used to detect convergence and stationary status of the models. Despite the excessive vi zero in the dataset, it was observed that Poisson MCMCglmm estimates were better than the models of zero-inflated Poisson and hurdle Poisson MCMCglmm. Furthermore, Poisson MCMCglmm gave better estimation results in shorter computational time as well.