İzmir Bakırçay Üniversitesi’nin yenilenebilir enerji üretiminin ve enerji tüketiminin tahmin edilmesi
Yükleniyor...
Dosyalar
Tarih
2022
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Bakırçay Üniversitesi Lisansüstü Eğitim Enstitüsü
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
İzmir Bakırçay Üniversitesi’nin ekolojik ve ekonomik olarak kendi kendine yetebilen bir üniversite olma hedefinin gerçekleşebilmesi için elektrik üretiminin ve tüketiminin doğru öngörülüp doğru tahmin edilmesi son derece önemlidir. İzmir Bakırçay Üniversitesi enerji ihtiyacını yenilenebilir enerji kaynakları ile karşılamayı hedeflemektedir. Yenilenebilir enerji kaynakları, fosil yakıt santrallerinin aksine üretimi kontrol edilebilen enerji kaynakları değildir. Bu durumdan dolayı üretimin tahmin edilmesi gerekir. Elektrik üretiminin ve tüketiminin doğru tahmin edilmesi yalnızca üniversite için değil ulusal çapta da büyük önem arz etmektedir. Üretimin ve tüketimin doğru tahmin edilebilmesi, şebekede bulunacak elektrik miktarının doğru şekilde ayarlanabilmesi ve kaybın azaltılabilmesine katkı sağlar. Bu tahminlemenin saatlik olarak yapılması şebekede bulunacak elektrik miktarının daha hassas ayarlanabilmesini sağlamanın yanı sıra akıllı bina, akıllı kampüs, akıllı şebeke gibi uygulamalar için büyük önem arz etmektedir. Halihazırda kurulu bulunan güneş enerji santraline ait 1 Temmuz 2019 tarihi ile 31 Ekim 2021 tarihleri arasındaki elektrik üretimi ve aynı tarihler arasındaki elektrik tüketimi farklı tahmin yöntemleri ile tahmin edilmiştir. Elektrik üretiminin tahmini için sıcaklık, küresel güneş radyasyonu, bulutluluk, yağış miktarı gibi meteorolojik parametreler kullanılmıştır. YSA’lar ile yapılan tahminleme için doğrusal olmayan otoregresif ağ ve dış girdili doğrusal olmayan otoregresif ağ modelleri kullanılmıştır. Kullanılan bu modellerde eğitim algoritması olarak bayes optimizasyon algoritması ve Levenberg-Marquardt algoritması kullanılmıştır. Yapılan denemelerin sonucunda dış girdili doğrusal olmayan ağ hem bayes optimizasyon algoritması ile hem Levenberg-Marquardt algoritması ile doğrusal olmayan ağa kıyasla daha başarılı tahminlerde bulunmuştur. Eğitim algoritması olarak kullanılan Bayes optimizasyon algoritması da hem dış girdili doğrusal olmayan ağ modelinde da hem de doğrusal olmayan ağ modelinde yapılan tüm tahminlerde Levenberg-Marquardt algoritmasından daha başarılı tahminlerde bulunmuştur. Güneş enerji santralinin üretimi 214,75 ortalama kare hatası değeri ile 10 nöronlu ağ ile tahmin edilmiştir. Üniversitenin elektrik tüketimi ise 435,82 ortalama kare hatası değeri ile 5 nöronlu ağ ile tahmin edilmiştir. Ayrıca elektrik üretimi ve tüketimi regresyon analizi ile de tahmin edilmeye çalışılmıştır. Regresyon analizi ile tahmin edilen elektrik üretiminin ortalama kare hatası değeri 831,3238 olarak bulunmuştur.
For İzmir Bakırçay University to achieve its goal of being an ecologically and economically self-sufficient university, it is crucial that electricity production and consumption be accurately predicted and estimated. İzmir Bakırçay University aims to meet its energy needs with renewable energy sources. Unlike fossil fuel power plants, renewable energy sources are not energy sources whose production can be controlled. Because of this situation, the production must be estimated. Accurate estimation of electricity production and consumption is of great importance not only for the university but also at the national level. Predicting the production and consumption correctly contributes to correctly adjusting the amount of electricity to be found in the grid and reducing the loss. Making this estimation hourly is of great importance for applications such as smart buildings, smart campuses and smart grids, as well as enabling more precise adjustment of the amount of electricity to be found in the grid. The electricity generation of the currently installed solar power plant between July 1, 2019, and October 31, 2021, and the electricity consumption between the same dates were estimated using different estimation methods. Meteorological parameters such as temperature, global solar radiation, cloudiness, and precipitation were used to estimate electricity production. For the prediction made with artificial neural networks, nonlinear autoregressive network and nonlinear autoregressive network models with external input are used. These models used the Bayes optimization algorithm and Levenberg-Marquardt algorithm as training algorithms. As a result of the experiments, the nonlinear network with external inputs made more successful predictions with both the Bayes optimization algorithm and the Levenberg-Marquardt algorithm compared to the nonlinear network. The Bayes optimization algorithm, which is used as a training algorithm, also made better predictions than the Levenberg-Marquardt algorithm in all predictions made both in the nonlinear network model with external input and in the nonlinear network model. The generation of the solar power plant was estimated with a 10-neuron network with a mean square error value of 214.75. The university's electricity consumption was estimated with a 5-neuron network with a mean square error value of 435.82. In addition, electricity production and consumption were tried to be estimated by regression analysis. The mean square error value of the electricity production estimated by the regression analysis was found to be 831,3238
For İzmir Bakırçay University to achieve its goal of being an ecologically and economically self-sufficient university, it is crucial that electricity production and consumption be accurately predicted and estimated. İzmir Bakırçay University aims to meet its energy needs with renewable energy sources. Unlike fossil fuel power plants, renewable energy sources are not energy sources whose production can be controlled. Because of this situation, the production must be estimated. Accurate estimation of electricity production and consumption is of great importance not only for the university but also at the national level. Predicting the production and consumption correctly contributes to correctly adjusting the amount of electricity to be found in the grid and reducing the loss. Making this estimation hourly is of great importance for applications such as smart buildings, smart campuses and smart grids, as well as enabling more precise adjustment of the amount of electricity to be found in the grid. The electricity generation of the currently installed solar power plant between July 1, 2019, and October 31, 2021, and the electricity consumption between the same dates were estimated using different estimation methods. Meteorological parameters such as temperature, global solar radiation, cloudiness, and precipitation were used to estimate electricity production. For the prediction made with artificial neural networks, nonlinear autoregressive network and nonlinear autoregressive network models with external input are used. These models used the Bayes optimization algorithm and Levenberg-Marquardt algorithm as training algorithms. As a result of the experiments, the nonlinear network with external inputs made more successful predictions with both the Bayes optimization algorithm and the Levenberg-Marquardt algorithm compared to the nonlinear network. The Bayes optimization algorithm, which is used as a training algorithm, also made better predictions than the Levenberg-Marquardt algorithm in all predictions made both in the nonlinear network model with external input and in the nonlinear network model. The generation of the solar power plant was estimated with a 10-neuron network with a mean square error value of 214.75. The university's electricity consumption was estimated with a 5-neuron network with a mean square error value of 435.82. In addition, electricity production and consumption were tried to be estimated by regression analysis. The mean square error value of the electricity production estimated by the regression analysis was found to be 831,3238
Açıklama
Anahtar Kelimeler
Enerji tahminleme, Yapay sinir ağı, Çoklu regresyon, NARX, Akıllı enerji yönetimi, Intelligent energy management, Multiple regression, Artificial neural networks, Energy estimation
