Hybrid Nonlinear Statics and Its Distributed Computing Implementation

Estimation of surface-consistent residual statics is a crucial factor limiting the imaging quality. However, for seismic data acquiring at complex surface and has low signal-to-noise ratio, the traditional linear residual statics is usually ineffective. The most prominent problem is that the traditional method is easy to fall into local optimum, and it is difficult to solve complex residual statics. The residual statics problem is essentially a nonlinear global optimization problem with multi-extreme values, hence, nonlinear algorithm should be introduced. The nonlinear residual statics technique can automatically jump out of the local optimal solution, which has a strong global optimization ability. Monte Carlo is one of the nonlinear methods mostly adopted in nonlinear statics. But there are still many other nonlinear methods can be employed to further improve the optimization ability. Meanwhile, the “embarrassingly parallel” feature of the nonlinear algorithm make it hard to realize distributed parallel computing. Mainstream commercial software can only deal with 2D datasets or on a single node with large memory. This paper demostrates a hybrid nonlinear statics method, which is effective for the global optimazation of residual statics. In addtion, the high-performance distributed computing employing Spark distributed computing architecture is realized in this paper.