FORWARD MODELING OF MARINE ELECTROMAGNETIC SURVEY USING ARTIFICIAL NEURAL NETWORKS

Marine electromagnetic (EM) survey is an engineering endeavor to determine the
location and dimension of hydrocarbon reservoirs which are particularly situated under
the sea floor. Forward modeling is one of the important step in processing the data of
a marine EM survey. By this modeling, the distribution of resistivity values along the
sea bed could be mapped and the location and dimension of the associated hydrocarbon
layer could be predicted.
As an alternative to the established methods in conducting forward modeling, in this
research two types of artificial neural networks are employed to determine the possibil�ities of them as forward models for marine EM survey. The networks are a multi-layer
perceptron (MLP) network and a radial basis function (RBF) network. The motivation
of this work is to find out the possibilities of these networks as forward models for
marine EM survey.
To achieve the research goals, a set of synthetic data must be generated using a
simulation software. These data are used to train and test the MLP and RBF networks
until they attained a sufficient property of generalization in modeling marine EM survey
data. To validate the correctness of the models, a reverse method of forward modeling
has been employed, which is the inversion process.
Occam's inversion has been specifically used to validate the neural networks' for�ward modeling. It is found that artificial neural networks, specifically MLP and RBF
networks, have a possibility to become forward models for marine electromagnetic sur�vey. In addition, this research found that the forward modeling by RBF network is better
than the corresponding one by MLP network.