ANN Methods For Seismic First Break Picking And Its Parallelization Study

Tomographic Static Corrections is used in data processing of complex geology with good effect. On the other hand, it's application is based on the exactness of first break picking. With the development of the seismic prospecting to the region with more complex surface, such as the desert and the mountains, the old methods of first break picking can't meet the requirement. So, the research of more precise and adaptable new method is needed.ANN(Artifical Neural Network) is a cross subject, which can complete difficult intelligent analysis and process effectively nonlinear, fuzzy and uncertain problems with simple mathematic methods. Because of its massive parallel processing, distributed storage, self-adaptation and error allowance, ANN has been widely applied in biology, electronics, computers, mathematics and physics.In this paper, based on the research and analysis of the theory of the classical BP NN (Neural Network), The author has developed the dichotomy initialization of weights, auto-optimization of the network structure, self-adaptation rate of training and random-optimization operator methods. Farther, the FNN(Fuzzy Neural Net -work) based on the pi-sigma module was achieved. The ANN methods have been approved available in seismic first break picking by applications in theory data. The result of seismic first picking is satisfied by using the modified BP NN and the FNN. On the other hand, the seismic data, especially the three-dimensional seismic data, is enormous. The serial arithmetic can't meet the requirement of the seismic data processing, so the parallel arithmetic is indispensable. We have designed the modified BP NN parallel arithmetic based on the research of instinctive parallelity in BP NN and the MPI (Message Passing Interface), and realized the parallel processing for the seismic first picking.