Multi-Layer Perceptron Neural Network for Air Wave Estimation in Marine Control Source Electromagnetic Data

Abstract

Marine Control Source Electro-Magnetic (MCSEM) survey is a technique for remote identification of sub-sea floor structures of the earth's interior using Electro-Magnetic (EM) signals. Air wave signal is major problem associated with the data recorded by this technique in shallow water environment. The air wave signals are parts of the EM signals that propagate from EM source via the atmosphere and induced along air/sea surface. These air wave signals has the ability to limit and mask the electromagnetic response of a subsurface resistive body so that signals from subsurface, possibly containing valuable information about a resistive hydrocarbon reservoir is hardly distinguishable. This paper presents the application of a feed forward multi-layer perceptron neural networks model for estimation of air waves in MCSEM survey data based on offset and sea water depth values. The proposed model has 3 hidden layers with sigmoid activation function, an output layer with purelin transfer function and Levenberg-Marquardt (trainlm) as the training function. Simulated airwave data for ten sea water depths from 1000m to 100m at interval of 100m were used as the training data. Coefficient of multiple determination and Mean Square Error (MSE) obtained from the multi-layer perceptron model and the estimation with multiple linear regression model are compared. Preliminary results demonstrate that multi-layer perceptron neural networks are a viable technique for the estimation of air waves in MCSEM data.