关键词: 小波网络, 小波变换, 分段时-频分析, 地震资料反演

Abstract: In the light of the theoretical technique of wavelet networks,we study the method for inversing the time difference of artificial acoustic wave (Δt) by using multi-scale decompositon of wavelet from seismic records. We analyze also the theoretical method of wavelet decomposition,as well as network learning and training from seismic signals. In these studies,we find that for the adjacent seismic channels,the similarity of a smaller section is better than that of the whole channel. Therefore,the information obtained from the adjacent seismic channel can be extrapolated to the other channel by using time-frequency analysis method of wavelet. The results of the afore-mentioned studies,as well as the inversion and calculation of practical seismic data indicate that the structure of the wavelet network is much easier to be determined than that of the artificial neural network,while the convergence of wavelet network is faster than that of artificial neural network. In addition,the high and low frequency information of the adjacent channel can be transformed to the other seismic channel by taking the advantage of better similarity of subsection time-frequency analysis of seismic data,as well as the capability of learning,training and memorizing of the wavelet network. In this way,we may enhance not only the resolution capability but also the reliability of seismic data inversion.

Key words: wavelet networks, wavelet transfer, subsection time-frequency analysis, seismic data inversion