Using Neural Networks to Evaluate the Effectiveness of a new Seismic Fault Attribute

Reflection seismology is the most commonly used method to obtain a picture of the earth interior. For the oil industry, the final goal of the seismic interpretation is the identification of possible hydrocarbon reservoirs. The detection of geological faults is a basic step in this process. Many attributes derived from the seismic data have been proposed as fault identifiers. Despite the great computational cost of most attributes, no attribute alone is able to solve the automatic fault detection problem. A robust approach to improve the quality of the detection is the so-called meta-attributes. It consists in a non-linear combination of a number of attributes with the use of a supervised neural network. This work proposes a new attribute with low computation cost. In order to evaluate its effectiveness, two meta-attributes are built with the Multi-layer Perceptron network method: one with the network trained with just a traditional attribute and another which also includes the new one. Their training and classification performances are compared.