Publication: Yapay Sinir Ağları Kullanılarak Evapotranspirasyonun Tahmin Edilmesi ve Ampirik Metotlarla Karşılaştırılması
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Abstract
Bu çalışmada farklı yapay sinir ağları (YSA) modelleri (çok katmanlı yapay sinir ağları (ÇKYSA), radyal tabanlı yapay sinir ağları (RTYSA), genelleştirilmiş regresyon yapay sinir ağları (GRNN), bulanık yapay sinir ağları (ANFIS) ile ASCE Penman-Monteith yönteminden elde edilen referans bitki su tüketimini tahminleri Zimbabve Bulawayo Goetz ili meteoroloji istasyonu verileri kullanılarak gerçekleştirilmiştir. Zimbabve Bulawayo Goetz ili meteoroloji istasyonundan 1988-2001 yılları arasındaki iklimsel verilerden (sıcaklık, rüzgar hızı, güneş radyasyon ve bağıl nem) faydalanılarak farklı amprik yöntemlere (ASCE Penman-Monteith, Hargreaves, Turc, Priestley Taylor, Ritchie ve Valiantzas) göre günlük evapotranspirasyon (ETo) miktarı hesaplanmıştır. Yapay sinir ağlarında kullanmak üzere 1988-1994 yıllarındaki veri setleri eğitim, 1995-1998 yıllarındaki veri setleri test, 1999-2001 yılları arasındaki veri setleri ise doğrulama olarak alınmıştır. Beş farklı normalizasyon tekniği kullanılmış ve YSA modeli tahmin doğruluğu üzerindeki etkisi de araştırılmıştır. Modellerin performansını belirlemek için belirleme katsayısı (r2), tahmin hatasının standart sapması (RMSE) ve ortalama mutlak hata (MAE) kullanılmıştır. ÇKYSA modeli, güneş radyasyonu, bağıl nem, rüzgar hızı, maksimum ve minimum sıcaklık girdisi olarak kullanılarak denenmiş ve girdi kombinasyonları arasında en iyi performans sağlanmıştır. ÇKYSA'ın Levenberg Marquadt (LM) algoritması ETo tahmininde Conjugate Scale Gradient (SCG), ve Resilient Propagation (RP) algoritmalarına göre daha iyi sonuçlar vermiştir. Deneysel modeller ASCE Penman-Monteith yöntemi ile karşılaştırılmış ve Hargreaves ve Turc modeli ETo tahmininde en iyi performansı göstermiştir. Çalışmadan elde edilen sonuçların Zimbabve'de, sulama suyunun tahmini, su kaynakalarının planlanması ve yönetimi çalışmalarında faydalı olacağı düşünülmektedir. Anahtar Kelimeler: Bitki Su Tüketimi, Yapay sinir ağları, Bulawayo Goetz,Zimbabve
The paper investigated the ability of neuro computing techniques, that are, multi-layer perceptrons (MLP), radial basis function (RBNN), generalized regression neural networks (GRNN), adaptive neuro-fuzzy neural networks (ANFIS GP, SC), in modelling reference evapotranspiration (ETo) for Bulawayo Goetz weather station in Zimbabwe. Firstly, ETo is calculated using ASCE-EWRI Penman-Monteith equation from the available climatic data (mean temperature, wind speed, solar radiation and relative humidity) obtained over a period of thirteen years. The data was used to forecast ETo and as inputs while calculated ASCE PM ETo estimates were used as outputs during artificial neural network(ANN) data processing. The obtained climatic data was then divided into training, testing and validation sets, for use during neuro computing. Five different normalization techniques were used and their impact on the accuracy of ANN model prediction was also investigated. Comparisons were made between the ETo estimates of neuro computing techniques and those of ASCE PM ETo. Determination coefficient (r2), root mean square error (RMSE) and mean absolute error (MAE) statistic parameters were used to evaluate the models' performances. In another seperate application, the ASCE PM ETo and artificil neural networks' estimates were compared against the empirically calculated ETo (Hargreaves, Turc, Priestly Taylor, Ritchie and Valiantz). The MLP (5-4-1) model produced the best results amongst neuro computing techniques. The Levenberg-Marquardt learning algorithm performed better than Conjugate Scale Gradient (SCG) and Resilient Propagation (RP). Hargreaves and Turc methods compared to other conventional methods have the best results and are recommended for use in cases of limited or missing climatic data. Key Words: Evapotranspiration; Empirical Methods; Neuro Computing; Bulawayo Goetz.
The paper investigated the ability of neuro computing techniques, that are, multi-layer perceptrons (MLP), radial basis function (RBNN), generalized regression neural networks (GRNN), adaptive neuro-fuzzy neural networks (ANFIS GP, SC), in modelling reference evapotranspiration (ETo) for Bulawayo Goetz weather station in Zimbabwe. Firstly, ETo is calculated using ASCE-EWRI Penman-Monteith equation from the available climatic data (mean temperature, wind speed, solar radiation and relative humidity) obtained over a period of thirteen years. The data was used to forecast ETo and as inputs while calculated ASCE PM ETo estimates were used as outputs during artificial neural network(ANN) data processing. The obtained climatic data was then divided into training, testing and validation sets, for use during neuro computing. Five different normalization techniques were used and their impact on the accuracy of ANN model prediction was also investigated. Comparisons were made between the ETo estimates of neuro computing techniques and those of ASCE PM ETo. Determination coefficient (r2), root mean square error (RMSE) and mean absolute error (MAE) statistic parameters were used to evaluate the models' performances. In another seperate application, the ASCE PM ETo and artificil neural networks' estimates were compared against the empirically calculated ETo (Hargreaves, Turc, Priestly Taylor, Ritchie and Valiantz). The MLP (5-4-1) model produced the best results amongst neuro computing techniques. The Levenberg-Marquardt learning algorithm performed better than Conjugate Scale Gradient (SCG) and Resilient Propagation (RP). Hargreaves and Turc methods compared to other conventional methods have the best results and are recommended for use in cases of limited or missing climatic data. Key Words: Evapotranspiration; Empirical Methods; Neuro Computing; Bulawayo Goetz.
Description
Tez (yüksek lisans) -- Ondokuz Mayıs Üniversitesi, 2016
Libra Kayıt No: 92084
Libra Kayıt No: 92084
Keywords
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
165
