Publication: Yapay sinir ağları ile akım tahmini: Mahmudiye göleti örneği
Program
Authors
Can, Murat
Advisor
Date
Language
Type
Journal Title
Journal ISSN
Volume Title
Abstract
Yapay Sinir Ağları yöntemi, insan beyni gibi öğrenebilen, öğrendiklerinden tecrübe edinen, mukayese yapabilen ve de sonuçlar çıkarabilen, günümüz teknolojisine en uygun, son 30 yıl içerisinde mühendislik alanında hızla yaygınlaşan ve gelişen, bir simülasyon modelidir. Bu yöntemin günümüzde yaygın olarak kullanılan diğer matematiksel modellerden en önemli ayrıcalığı, modelin gerçek veriler ile sürekli eğitilmesi, veri adedi çoğaldıkça modeldeki mevcut veriler üzerine eklenerek, gerçeğe daha yakın sonuçlar çıkarabilmesidir. Bu çalışmada Yapay Sinir Ağları yönteminin hidroloji alanına da uyarlanabileceğini göstermek amacıyla, su yapılarının mühendislik açısından işlevsel, emniyetli ve ekonomik planlanması için gereken belirli bir zaman sürecindeki akım değerlerini elde etmek için kullanılmıştır. Bu amaçla oluşturulan yapay sinir ağı çeşitli alternatif durumlar için eğitilmiştir. Eğitim yani girdi ve test yani çıktı veri setlerini oluştururken çoklu regresyon modelinden yararlanılmıştır. Sonuçta oluşturulan yapay sinir ağı, halen devlet kurumları tarafından kullanılan lineer regresyon ile akademik ve ihtisas kurumları tarafından kullanılan çoklu regresyon modelleri ile karşılaştırılmıştır. Sonuçlar yapay sinir ağlarının bu alanda rahatlıkla kullanılabileceğini, hata oranlarının mühendislik toleransları içinde kaldığını göstermiştir. İleriki zamanlarda ağların ara katmanları daha iyi çözümlendiği zaman daha başarılı sonuçlar vereceği kuşkusuzdur.
Artificial Neural Networks are simulation models having skills as learning like human brain, acquiring experience, making comparision, deducting and so familiar for updated technology, also highly developed and applicable to engineering field for the last 30 years. The most important exclusivity of this kind of models rather than the other extant mathematical models, is the sustainable training of the model by existing real data values. Corresponding trained model can update itself according to the new data sequences that can be added upto existence for the time being and the models reach to conclusion so close to the actual result. In this study; it is aimed to predicate that Artificial Neural Networks are applicable to hydrologic cases of deriving flow rates for a adequate period of time that is required for planning functional, secure and economical hydraulic structures by engineering point of view. Thereby composing artificial neural network by aiming is trained for different alternative cases. Trained as input data set and Test as output data set are produced whereby multiple regression analysis. As a result of forming Artificial Neural Network is compared both by the lineer regression analysis used for governmental agencies and by the multiple regression analysis used for academical and practice instituties. Eventually the results are pointed out that this models have sufficient reliability for these cases, because of the error margins remain in the limit of engineering tolerances. But when the intermediate layers of these networks are resolved more effectively at a future, it is definite that these models will yield more succesful results.
Artificial Neural Networks are simulation models having skills as learning like human brain, acquiring experience, making comparision, deducting and so familiar for updated technology, also highly developed and applicable to engineering field for the last 30 years. The most important exclusivity of this kind of models rather than the other extant mathematical models, is the sustainable training of the model by existing real data values. Corresponding trained model can update itself according to the new data sequences that can be added upto existence for the time being and the models reach to conclusion so close to the actual result. In this study; it is aimed to predicate that Artificial Neural Networks are applicable to hydrologic cases of deriving flow rates for a adequate period of time that is required for planning functional, secure and economical hydraulic structures by engineering point of view. Thereby composing artificial neural network by aiming is trained for different alternative cases. Trained as input data set and Test as output data set are produced whereby multiple regression analysis. As a result of forming Artificial Neural Network is compared both by the lineer regression analysis used for governmental agencies and by the multiple regression analysis used for academical and practice instituties. Eventually the results are pointed out that this models have sufficient reliability for these cases, because of the error margins remain in the limit of engineering tolerances. But when the intermediate layers of these networks are resolved more effectively at a future, it is definite that these models will yield more succesful results.