Integration of Fuzzy Kohonen Clustering Networks (FKCNs) with Neural Networks for Well Log Properties Estimation

Abstract

In this paper we propose a Fuzzy Kohonen Clustering network (FKCN) to estimate the porosity log from seismic attributes in an oil field in south-west of Iran. There are two main reasons why we have used Kohonen self organizing map which is an online method. First, in Geophysics generally we are dealing with a large volume of data which can deteriorate the performances of Neural Networks. Since this technique, unlike off line methods, gets data one by one, it helps Neural Networks to work properly by reducing mathematical complexity. Second, this method allows our model to be adapted as new wells are drilled. Here we have applied this clustering technique to obtain a more accurate result to classify seismic data. Results indicate that, integrating FKCN with neural network can be a strong predictor of reservoir characteristics through its application on seismic data. Performance exploration of neural networks with or without appliance of such technique has been studied comparatively. The outcome of applying this method on conventional neural network such as PNN or RBF proposes a considerable improvement in porosity log estimation compared to the same neural networks when no such technique is performed.