Predictive Deconvolution Based On Maximization Of Non-Gaussianity

In oil seismic exploration, multiple suppression is always one of the key and difficult problems. The existence of multiples can interfere or even damage the image formation of the true boundary of the offset section. Therefore, it is important to identify and weaken or remove the multiples reliably in seismic data processing.Predictive deconvolution is one of the common used multiple suppression methods. The traditional predictive deconvolution (TPD) is based on the second order statistics, and assumes that the multiples and primaries are orthogonal. However, the real seismic data usually do not satisfy this assumption.In this paper, we use the property that the multiples and primaries are non-Gaussian distributed, propose an improved predictive deconvolution based on maximization of non-Gaussianity (PDMNG), and apply it to multiple suppression. Compared with TPD, PDMNG does not need the assumption that the multiples and primaries are orthogonal. Results of artificial synthetic data experiments show that, using PDMNG is better than using TPD in multiple suppression.However, since the real seismic data is usually time-space variant, using PDMNG usually can not get good results in real seismic processing. Based on the time-space variant property of the real seismic data, we apply different predictive filters to different time sections, and propose a time-space variant predictive deconvolution based on maximization of nongaussianity (TSV-PDMNG). Results of real seismic dataset show that, using TSV-PDMNG can remove the multiples from the seismic dataset very well.In this paper, we also propose a multichannel predictive deconvolution based on maximization of nongaussianity (MC-PDMNG), based on the lateral continuity of real seismic dataset. Results of real seismic dataset show that, compared with single channel predictive deconvolution, MC-PDMNG can remove more multiples, but the numerical stability of the algorithm is weak.