In spite of the technologic development happened at seismic prospection, and the significative amount of data with seismic two-dimensional (2D) and three-dimensional (3D) surveys, some process in the seismic interpretation task like the first break picking, remains in a manual version, that still needs an intuitive human intervention. This dissertation purpose, fill in the seismic processing with the intention to look for an efficient method to enable the computational simulation of the human visual system behavior, through decision process automation involved in first break picking in a seismic trace; looking at to preserv the interpreter intuitive knowledgement to more complex tasks, where your knowledgement will be better profitable. Neural networks, the most important implementation of neurocomputing systems, were initially developed by neurobiologists as computer models of the neural system in the brain. They differ from conventional computation techniques in their ability to adaptively discriminate or learn through repeated exposure to examples, their tolerance to data component failure and their robustness in the presence of high noise levels. This computing technology provide some techniques that can reduce the labor intensive aspects of the first break picking, maintaining the quality and reliability of the results. The method here presented is an application of an artificial neural network computational process, known as feedforward multilayer perceptron trained with the error back-propagation algorithm; from the establishment of a convenient neural network architecture and learning set that make possible its application over seismic data. This method is a computational simulation of seismic interpreter decision intuitive process for first break picking in seismic traces. The applicability, efficiency and limitations of this approach will be appraised in synthetic data obtained starting out the ray theoretical method.
|
