Document Type : Original Research Paper

Authors

• T Azari ¹
• N Samani ²

¹ Ph.D., Department of Earth Sciences, Sciences College, Shiraz University, Shiraz, Iran

² Professor, Department of Earth Sciences, Sciences College, Shiraz University, Shiraz, Iran

Abstract

In recent years, the artificial neural networks (ANNs) are used as an alternative to the conventional type curve matching techniques for the determination of aquifer parameters. In this paper two multilayer perceptron networks (MLPNs) are developed for the determination of leaky confined aquifers parameters. Leakage into the aquifer takes place from either the upper aquifer through the confining aquitard or the storage in the confining aquitard. The first and second networks are trained for the well functions of leaky aquifers (a) without and (b) with storage in the confining aquitard, respectively. By applying the principal component analysis (PCA) on the adopted training data sets the topology of both networks are reduced and their efficiency increased considerably. In contrast to the existing networks the topology of developed networks is fixed to (2×10×2) regardless of number of records in the pumping test data. The networks generate the match point coordinates for any individual pumping test data set. The match point coordinates are incorporated with Hantush-Jacob (1955) and Hantush (1960) analytical solutions and the aquifer parameter values are determined. The performance of the MLPNs is evaluated by three sets of real field data and their accuracy is compared with that of type curve matching techniques. The proposed MLPNs are recommended as simple and reliable alternatives to previous ANN methods and the type-curve matching techniques.

Keywords

• Aquifer parameter estimation
• Aquitard
• Artificial Neural Network
• Principal component analysis (PCA)
• Levenberg–Marquardt (LM) training algorithm