Fuzzy genetic approach for modeling of the critical submergence of an intake


KOCABAŞ F., Unal B., Unal S., FEDAKAR H. İ. , GEMİCİ E.

NEURAL COMPUTING & APPLICATIONS, vol.23, 2013 (Journal Indexed in SCI) identifier identifier

  • Publication Type: Article / Article
  • Volume: 23
  • Publication Date: 2013
  • Doi Number: 10.1007/s00521-012-1241-6
  • Title of Journal : NEURAL COMPUTING & APPLICATIONS

Abstract

The vertical distance between the water level and upper level of intake is called submergence. When the submergence of the intake pipe is not sufficient, air enters the intake pipe and reduction in discharge occurs. The submergence depth at which incipient air entrainment occurs at a pipe intake is called the critical submergence (S-c). It can also cause mechanical damage, vibration in pipelines and loss of pump performance. Therefore, the determination of the S-c value is a significant problem in hydraulic engineering. To estimate the S-c values for different pipe diameters, experimental works are conducted and results obtained are used for modeling of critical submergence ratio (S-c/D-i). In this study, a fuzzy genetic (FG) approach is proposed for modeling of the S-c/D-i. The channel flow velocity (U), intake pipe velocity (Vi) and porosity (n) are used as input variables, and the critical submergence ratio (S-c/D-i) is used as output variable. The 44 data sets obtained by experimental work were divided into two parts and 28 data sets (approximately 64 %) were used for training, and 16 data sets (approximately 36 %) were used for testing of models. The experimental results were compared with FG, an adaptive neuro-fuzzy inference system (ANFIS) and artificial neural networks (ANNs). The comparison revealed that the FG models outperformed the ANFIS and ANN in terms of root mean square error (RMSE) and determination coefficient (R-2) statistics for the data sets used in this study. In addition to RMSE and R-2, which are used as main model evaluation criteria, mean absolute error is used to evaluate the performance of models.