Neural Network and Least Squares method (ANN-LS) for Depth Estimation of Subsurface Cavities CASE OF STUDIES: GARDANEH ROKH TUNNEL, IRAN

Natural cavities in limestone bedrock (Karst regions) cause sink holes in upper deposits. Detection of cavities is one of the most frequently cited applications of gravity survey. Sink-holes are seen very frequently in many parts of Iran. They are mainly produced due to a substantial decrease in the underground water table. They are vital and riskyproblems particularly when they exist in urban, road and industrial areas. We aim to detect and model the sink-holeswhich generate from Karstic holes in the basement using gravity method. The gravity interpretationmethods usually face a non-linear optimization problem with multi-variable, multi-objective function extremum, multi-solution and so on. Therefore, it is necessary that the more stable and efficient algorithms is used in the geophysical inversion. Artificial neural network has been used in the gravity data inversion (ANN). However, for high-dimensional, multipeakfunction problems in gravity anomalies data inversion, the effect using ANN method is not good, and it easy tofall into the local minimum. In this paper, we propose ANN and least squares method (LS) to solve gravity anomalies data parameter optimized inversion. This method is applied to synthetic data with andwithout random error. Finally, the validity of the method is tested on a field example from Karst region in Chahar mahal Bakhtiyari Province, IRAN

Keywords:

Simple Causative Sources , gravity data , natural cavities , ANN-LS method , Gardaneh Rokh tunnel , Iran