Bi̇yoloji̇k Atiksu Aritma Tesi̇sleri̇nde Karbon Ve Azot Gi̇deri̇m Veri̇mleri̇ni̇n Tahmi̇ni̇ Amaciyla Yapay Si̇ni̇r Ağlarinin Kullanimi

Abstract

ÖZBiyolojik azot giderimi gerçekleştirilen atıksu arıtma proseslerinin (AO prosesi) modellenmesi amacıyla Aktif Çamur Modeli No. 1 (ASM1) kullanılagelmişse de bu modelde ihtiyaç duyulan girdi parametrelerinin tahmin edilmesi çok zaman almaktadır. Bu çalışma kapsamında, ASM1 kadar detaylı girdi verisi gerektirmeyen geri beslemeli yapay sinir ağlarının ( USE OF ARTIFICIAL NEURAL NETWORKS AS A TOOL TO PREDICT CARBON AND NITROGEN REMOVAL EFFICIENCIES IN BIOLOGICAL WASTEWATER TREATMENT PLANTS ABSTRACTAlthough Activated Sludge Model No. 1 (ASM1) was used for modelling biological nitrogen removal processes, estimation of input parameters required to run this model necessitates complicated laboratory analyses. In this study, the performance of Backpropagation Artificial Neural Networks (BPANN), which requires considerably less numbers of input parameters, in predicting chemical oxygen demand (COD), total Kjeldahl nitrogen (TKN), and total nitrogen (TN) removal efficiencies was tested. For this purpose, four activation functions were employed in BPANN. Results suggested that COD, TKN, and TN removal efficiencies in AO processes can be accurately estimated using BPANN, with the highest learning and prediction capacity when Sinc function is employed. The mean square errors (MSEs) with Sinc-BPANN were calculated as 2.5010 -4 for COD removal efficiency, 4.1510 -4 for TKN removal efficiency, and 2.6510 -4 for TN removal efficiency. Therefore, the Sinc-BPANN is concluded to be an efficient tool for estimating nonlinear nature of COD, TKN, and TN removal efficiencies in AO processes using considerably less numbers of input parameters.