Optimization Method Research On Seismic Wavelet Extraction

Seismic wavelet estimation is base of seismic impedance inversion, processing of seismic data deconvolution, three-term AVO inversion, offset, feature extraction, geophysical interpretation and seismic forward model. The extracted seismic wavelet is good or not will affect the forward model, inversion results and seismic data interpretation. In order to obtain a more precise seismic wavelet has a profound significance. Always, seismic wavelet is mixed, in frequency domain concentrates in subsection. In the condition of the noise is stronger, extraction seismic wavelet needs to use the appropriate band-pass filter effectively filtered out some noise frequencies, which will improve the accuracy of the extracted seismic wavelet. Because the extracted seismic wavelet length is unknown and the reflection coefficient sequence is non-Gaussian, needs to use higher order statistics approximated the length of seismic wavelet. Under the convolution model forms seismic records, using moving average (MA), autoregressive moving average (ARMA) model to establish the objective function, then gets seismic wavelet by solving it.first, using the SM (Steiglitz-Mcbride) method under the ARMA model extracts seismic wavelet, the simulation shows that at the zero-phase case and strong noise can greatly extract seismic wavelet, but under mixed phase extracts seismic wavelet became worse; then, using the characteristics of higher order statistics retention phase, due to the synthetic seismic data shortly and length of extracted seismic wavelet uncertainty, using cumulant matrix equation method to solve cannot get satisfactory results, so only uses high order cumulant spectrum extracted seismic wavelet phase,and combines with SM extracted the amplitude, restructuring the seismic wavelet, also calls SM-HOCS algorithm extrcated seismic wavelet; Finally, under MA model established the objective function, using band-pass filter on the synthetic seismic data filters some noise and using higher order statistics roughly determines the extracted wavelet length which gets well search range for the improved genetic algorithm to extract the seismic wavelet, simulation gets good results in the mixed phase, relatively low signal to noise ratio of the actual situation and in processing seismic data.Analysing the lack of real-coded genetic algorithm optimization for the bounded domain of high-dimensional function, a new improved real-coded genetic algorithm proposes. Avoiding premature convergence and population drift adds chaos theory acting on few of the same individuals, which increased diversity of the population; in order to enhance the effectiveness of the search direction, using conjugate direction method as the mutation direction and utilizing the idea of particle swarm optimization for crossover. On several typical function simulation and extracted seismic wavelet in seismic data, the results show that the improved algorithm is not only fast convergence,robustness, also can be optimized with higher precision.