Adaptive Multiple Attenuation Based On Blind Source Separation

It has been widely acknowledged that the research of multiple attenuation algorithms has become one of the most important subjects in the area of seismic exploration. This paper focuses on the research of adaptive multiple attenuation techonology based on constrained blind source separation method. The author presents some effective methods to improve current constrained blind source separation method and thus to solve some existing issues on multiple attenuation algorithms.Constrained blind source separation method has received more and more attention in the area of blind source separation. The main focus of this subject is to convert priori knowledge on sources to constraints and solve blind source separation problem based on corresponding constrained optimization algorithms and blind source separation algorithms. Current constrained blind source separation algorithms which are designed for adaptive multiple attenuation is mainly based on sparsity of seismic data. This paper proposes two methods to improve existing constrained blind source separation algorithms.In order to address the issues such as damages to primaries etc., this paper incorporates continuity of seismic data as constraints into algorithms and thus proposes a new algorithm of continuity constrained independent component analysis. During the realization phase of this algorithm, the author researched into many methods and issues such as multiple attenuation modeling, wavelet difference elimination method, mixed matrix and under-determined blind source separation etc. Experimental results on synthetic data and real data show that the algorithm this paper presents is better than current constrained blind source separation algorithms.In addition to the research above, this paper proposes a method based on continuity in frequency domain to solve under-determined blind source separation problem. This method makes use of constraints provided by prediction error filter in frequency domain, and converts previous under-determined problem to determined problem. Experimental results on synthetic data prove that this method is more effective than current constrained blind source separation.