M. Gohari-Moghadam - National Hard Rock Research Center, ۱۵۷ Havaboard Cross, Shiraz, Iran
N. Samani - Department of Earth Sciences, Shiraz University, Shiraz ۷۱۴۵۴, Iran-Department of Civil Engineering, University of Toronto, Toronto ON Canada M۵S ۱A۴
B. Sleep - Department of Civil Engineering, University of Toronto, Toronto ON Canada M۵S ۱A۴

Type curve matching techniques are traditionally applied to determine the transmissivity (T) and storage coefficient (S) of infinite homogeneous, isotropic confined aquifers from constant rate pumping test data. In this paper as a new alternative, the potential of an Adaptive Neuro-Fuzzy Inference System (ANFIS) has been investigated to approximate the match point coordinates and hence the aquifer parameters. The proposed ANFIS model has been trained for N drawdown measurements. The Lin and Chen (2006) approach and the Theis (1935) well function were used to generate (N-1×10205) training set. Performing Principal Component Analysis (PCA) makes possible the training of the proposed ANFIS model using the first principal component (PC1) as input. PC1 describes 99.911% of the training matrix variation. Consequently the dimension of the input matrix reduces to (1×10205) regardless of the number of drawdown measurements. During training, an automated supervised hybrid-learning algorithm calculates the parameters of the proposed ANFIS model. We found that 11-Gausian membership functions could approximate the well function with the desired accuracy in terms of the Relative Root Mean Square Error (RRMSE) of computed and target vectors of match point coordinate. Further, as an advantage to Artificial neural networks (ANN) we found that the ANFIS model could train well with a reduced number of training patterns. Finally, the proposed ANFIS model has been verified by 2000 synthetic drawdown sets and its reliability validated by two real pumping test drawdown records. it is found that the proposed ANFIS models estimates aquifer parameters better than graphical and ANN methods in terms of greater accuracy, shorter training time, and simpler model structure.

Adaptive Neuro-Fuzzy Inference System; Principal Component Analysis; Confined aquifer, Parameter identification