2D Inversion of Gravity Data Using Artificial Neural Networks in Order to Detect of Subsurface Cavities

Abstract

We aim to estimate the depth of subsurface cavities from gravity data by a new method of 2D inversion through a Multi Layer Perceptron(MLP) neural network.Infact, this method is an intelligent way to interpret gravity data and gain an estimation of depth and shape. The MLP neural network was trained for two main models of cavities: sphere and cylinder in a domain of radius and depth. We tested different MLP’s with different number of neurons in the hidden layer and obtained the optimum value for number of neurons in the hidden layer. Then it was tested in present of 10% noise(S/N=.1), and also tested for real data. It presented good results for depth estimation of subsurface cavities. Keywords: gravity, 2D inversion, subsurface cavities, artificial neural networks, Multi Layer Perceptron