Surface Related Multiple Attenuation Based On Independent Component Analysis

Multiple elimination is very important in seismic data processing. In the processing and interpreting of seismic data, the multiple may effect velocity model, seismic profile,geological structure interpretation and seismic properties abstraction. The main method of multiple attenuation include filtering methods which can deal with simple multiple and wave field prediction and subtraction methods which are good at eliminating complex multiple.The wave field prediction and subtraction methods predict multiple by wave field theory, and then subtract the multiple model by least square filter. The algorithm can not attenuate multiple that crossover the primary. In the paper, the algorithm were improved by higher order statistics. Primary were abstracted by independent component analysis instead of least square filter subtraction. The principle of this paper include three steps: Firstly, multiple predicting. The multiple predicting has the same principle as surface related multiple elimination, which predict multiple by convolution in time domain. Secondly, filtering and mixed multiple matrix preparing. In multiple predicting, a response of wavelet double counted should be eliminated by filter. Thirdly, independent component analysis to abstract the primary. In the step, the algorithm was designed and the two instinct indeterminacy of ICA, undetermined independent components and their amplitude, should be solved. In the process, the data outputted from multiple predicting, input into the filtering algorithm and the filtered multiple was used to predict the higher multiple orders multiple. The ICA mixed multiple matrix can be achieved by the iteration.Aim at testing the algorithms, some model seismic data was employed: the flat seafloor model, the rugged seafloor model and Pluto test seismic model. In the paper, a lot of parameters were tested such as the length of the filter, white noise coefficient, to achieve a good result. In the paper fast fixed point algorithm was selected to seek for the separating matrix, and then abstract and recover the primary. After the model test, all the algorithms should be tested by real seismic data. At last of this paper, a conclusion of multiple predicting, filtering and abstracting the primary were given.