2D inverse modeling of residual gravity anomalies from Simple geometric shapes using Modular Feed-forward Neural Network

Abstract

In this paper, we introduce a new method called Modular Feed-forward Neural Network (MFNN) to find the shape factor, depth and amplitude coefficient parameters related to simple geometric-shaped models such as sphere, horizontal cylinder, and vertical cylinder, which cause the gravity anomalies, in 2D cross section. Using MFNN inversion results can determine the shape, depth and radius of a causative body. The design of MFNN consists of 3 similar one layer feed-forward neural networks (FNNs). Each feed-forward Neural Network which is as a module, first train using the back-propagation method for a parameter with synthetic gravity data and then to test the trained networks with new gravity data. The new approach has been tested first on synthetic data from different models using well-trained networks. The results of this approach show that the parameters values estimated by the modular inversion are almost identical to the true parameters. Furthermore, the noise analysis has been examined where the outputs of the inversion produce satisfactory results with 10% of random noise. The reliability of this approach is demonstrated for real gravity field anomalies measured over an iron deposit in Kerman province, Iran. MFNN inversion show the best shape for the underground mass is vertical cylinder with a depth of 21.18 m and a radius of 17.89 m.