A New Type of Neural Network For Reservoir Identification Using Geophysical Well Logs

We have developed a neural network algorithm of radial basis network (RBN) type for geoscience applications. This new probabilistic neural network (PNN), referred to as “gravity-capturing neural network,” employs multidimensional even distance and introduces the resultant force competition mechanism for the output layer. When used for geological pattern recognition with well-logging data, it avoids misjudgment due to a magnitude jump of a single parameter and can extract complex and hidden formulas from laboratory and field measurements more efficiently. A field case study of reservoir identification with geophysical well logs is presented to demonstrate the advantages of this neural network over the conventional PNN in such classification applications.