Function Fitting for Control Source Electro-Magnetics Data Using Elman Neural Network

The problem of function fitting for certain geophysical problem such as Control Source Electro-Magnetic (CSEM) can be solved using a partially recurrent network called Elman Neural Networks (ENN). ENN is one of the subclasses of partial recurrent neural networks. A Recurrent Neural Network (RNN) is an important class of neural networks where connections between units form a directed cycle. The Elman network differs from conventional neural network structure, in that it has addition layer (context layer) with feedback connection from the output of the hidden layer to its input. This feedback path allows Elman networks to recognize and generate temporal patterns, as well as spatial patterns. ENN has an advantage of having a low probability of being affected by external noise. Also, it can be trained to act as an independent system simulator. This study presents an application of ENN in function fitting for CSEM data. The synthetic training data has been generated using Computer Simulation Technology (CST) software. As a preliminary study, the data set was selected carefully representing a no hydrocarbon reservoir CSEM simulation. The trained Elman network shows an encouraging good fitting with MSE as low as 0.000275.