Stancking velocity determination via neural networks

Abstract

State of the  of petroleum seismology data processing requires tedious and expensive interpretation of velocity spectrum. Usually this interpretation is perfarmed in an interactive way on workslatians. Even so, as petroleum seismology goes fram a world of two-dimensional data to a world where three-dimensianal data are a common place, the manual interpretation of velocity spectrum becomes a major bottleneek in the overall process of progressing from raw data to a stacked section. This paper shows an automated approach to the interpretation of veloeily speetra via synthetie neural networks. Three or four independent neural networks are emplayed: one for picking significant times, anather for picking staeking veloeity at the significant times found by the first time network, and one ar two netwarks for lhe validation of pieks selected by lhe first two nefworks. The nefworks need an initial veloeity function in addition to the seismie data. A fast automatie procedure for staeking veloeity pieking is the eonsequence. Rather eomprehensive tests on synthetie and real data indieate lhat this method will probably clear the above 
mentioned bottleneck.