2D electrical resistivity imaging based on backpropagation artificial neural network. (May 2019)

• Record Type: Journal Article
• Authors: Setyawan, A; Fikri, M S; Suseno, J E; Najib,
• Is Part Of: Journal of physics. Volume 1217(2019)
• Publication Date: 2019-05
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1217/1/012001
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Dipole-dipole is one configuration design survey in electrical resistivity method which is common to interpreted shallow subsurface base on resistivity parameter, and the model can be developed using inversion formula. Otherwise, we developed some model without empirical mathematic formula; it is called Artificial Neural Network (ANN). ANN is a system which has a pattern like the human brain process to solve the complex problems. The research aims to develop a neural network algorithm using Matlab and compare the result 2D model resistivity between ANN and inversion by Res2Dinv (existing procedure) software. The research was done in Taman Rumah Kita (TRK) where located in Faculty of Science and Mathematics, Diponegoro University. A cylinder was buried in the center of TRK and getting the best architecture of the network and the value of Mean Square Error (MSE) of the output network. The backpropagation artificial neural network was built from many layers such as an input layer, hidden layer, and output layer and developed by Matlab programs. The Network train was tested using synthetic data and field data. The synthetic data was made with forwarding modeling method by Res2Mod software, and the field data was obtained by doing the measurement at the measurement site using dipole-dipole resistivity method. The comparing result models are present the best architecture obtained one input layer with three input units, three hidden layers with each layer has 100 neurons and one output layer that obtained by trial and error process. MSE obtained respectively in observation lines are 0.0210 at Line 1, and 0.0345 at Line 2.