Research On Reflection Pattern Analysis Method Of Multidimensional Seismic Signals

The processing and interpretation of seismic exploration signal are the main means for finding underground oil and gas reservoirs.Different lithological parameters(especially elastic parameters),different reservoir characteristics(whether oil or gas)and different structural features(such as fractures,caves)can lead to the change in seismic reflection signal.People observe a specific reservoir seismic reflection signal,looking for the difference between the specific reservoir and other reservoirs,and make similar but not identical seismic reflection signals into a class.In order to emphasize the similarity of the same type of seismic reflection signal and the difference of different types of seismic reflection signals,we define the characteristics or phenomena of a type of seismic reflection signals which are different from other types of reflection signals as the pattern of the seismic reflection signals.The reflection pattern analysis of seismic signals can discover the common features of the same kinds of seismic reflection signals and the differences between different kinds of seismic signals,so as to identify different underground geological features.The existing reflection pattern analysis methods of seismic signal are mostly based on the single and post-stack seismic data characteristics to carry out one-dimensional seismic signal reflection pattern analysis,so that the analysis can not make full use of the geological reflection information,is not conducive to the fine explanation.Therefore,based on the pre-stack reflection feature and stratigraphic neighborhood information,the analysis method of seismic reflection pattern is studied from the perspective of multidimensional signal analysis in this thesis.The problems such as pre-stack reflection feature extraction,horizon interpretation error,formation thickness variation and data noise are analyzed,aiming at improving the stability and accuracy of seismic reflection mode analysis,and reducing the multi-solution of geological characteristic analysis,so as to obtain deep understanding of underground reservoir and provide help for exploration of complex oil and gas reservoirs.The main work and innovations of this thesis include the following five aspects:(1)The superiority of pre-stack seismic signals in analyzing different structural characteristics and reservoir characteristics is demonstrated,and the two-dimensional Gabor feature based pre-stack reflection recognition methos is proposed.The algorithm uses the good time-frequency and spatial response of the two-dimensional Gabor kernel to form effective feature of pre-stack reflection signal.The unsupervised and supervised prestack reflection pattern analysis is performed with SOM(Self-organizing Map)and TMR(Texture Model Regression)algorithm respectively.For solving the difficulty of dimensionality reduction and clustering analysis caused by high-dimensional Gabor features,SpPCA(Sub-pattern Principal Analysis)algorithm is used to reduce feature dimension for improving computational efficiency.The testing on real seimsic data demonstrates the advantages for identifying caves and predicting reservoir favorable area.(2)In order to analyze the difference in pre-stack reflection pattern caused by the continuity of the formation,the computing method of pre-stack texture attributes is proposed to solve the problem of inaccurate of texture attributes beacuse of waveform stacking effect.Pre-stack texture attributes can not only describe the discontinuity of different reflection points,but also represent the AVO(Amplitude Variation with Offset),anisotropy and media homogeneity.Multi-direction pre-stack texture attributes describe the variations of wavefrom from different aspects.Combined with the SOM clustering algorithm,the pre-stack texture based pattern analysis method is proposed.The study shows that the pre-stack texture attributes can describe the lateral variation of strata more accurately than the post-stack texture attributes,and the classification of pre-stack textures can effectively describe different seismic reflection configurations pattern.(3)To reduce the influence of horizon interpretation error on the analysis of seismic reflection pattern,the adaptive phase K-means algorithm is proposed.In the algorithm,the adaptive phase distance is proposed as a measure of waveforms similarity,which calculates the waveforms similarity under the optimal phase matching.Thus the proposed distance can suppress phase variations caused by the interpretation horizon.By taking knowledge from the traditional K-means algorithm,this thesis use EM(Expectation Maximization)algorithm to optimize the objective function optimization of the proposed algorithm and utlize the best matching phase to update the centroids.The test of synthetic data and physical model data show that the algorithm has good convergence,and finally obtain consistent waveform phase.The resulting map can effectively reveal the geological features in physical model.(4)For solving the problem of inconsistency of seismic reflection signal dimension beacause of the variations in formation thickness,the dynamic sub-window matching(DSWM)distance is proposed to measure the similarity between two unequal length seismic signals,according to the stratigraphic sedimentary characteristics and the reflection characteristics of seismic signals.DSWM can comprehensively use the top and bottom horizons of the target layer to analyze seismic reflection pattern,whereas fixed time windows can easily bring some information of other layers or discard some features of the target layer,resulting in undesirable classification.DSWM finds a serious of matched sub-windows for computing the distance between signals,and the dynamic programming algorithm is used to obtain the global optimal sub-windows in this thesis.By combining the proposed distance with K-nearest neighbor(KNN)classification algorithm,a supervised waveform classification algorithm called KNN+DSWM is developed,which can be used to quantitatively identify the spatial distribution of geological bodies based on the information from drilling wells.The testing results show that the KNN+DSWM method can adapt to the change of formation thickness and has some robustness to the horizon interpretation error.(5)Aiming at the influence of noise on the seismic reflection pattern analysis,the neighborhood information of seismic signals is introduced into the reflection pattern analysis process,and multi-waveform classification(MWFC)and RegFCM(Regularized Fuzzy C-means)algorithms are used for seismic reflection pattern analysis.The MWFC algorithm takes the neighborhood waveform as input and thus is not sensitive to the noise in seismic data.Considering the influence of the fault and the boundaries of the geological body on the classification,the multi-window method is used to extract the multi-waveform.The RegFCM algorithm introduces neighborhood constraints in its objective function and enhances the continuity of the classification results.By analyzing the influence of different parameters on classification results,a reasonable parameter selection scheme is given.Through the synthetic data test and the actual data application,the algorithm is proved to have the ability of noise suppression,and the continuous and reliable analysis results can be obtained.Finally,the performance and applicability of the two algorithms are compared and analyzed.