模擬退火系統於震測速度選取之最佳化之研究

Abstract

我們提出一個運用模擬退火演算法來解決反射震測訊號處理中速度選取 (velocity picking) 最佳化的問題。速度分析是基於在一個2D的速度影像 (semblance image) 裡挑選出time-velocity pairs，time-velocity pairs形成的曲線 (polyline) 表示速度與時間之間的關係函式。傳統的震測訊號速度選取是由有經驗的地球物理專家透過觀察semblance image及seismic survey來進行選取，需要花費不少時間。我們提出整個速度選取問題可以被公式化成為最佳化問題。我們使用peak-climbing clustering來識別semblance中極大peaks之所在。在peaks排列集合而成的有限解集合中，利用模擬退火演算法找到全域最佳解。從實驗中，使用模擬震測資料以模擬退火演算法找到最佳化的stacking velocity與時間的對應關係，再作normal moveout correction (NMO) 修正及stacking，得出stacked traces，可以反映真實地層的原貌 (in time)。我們的研究結果將有助於速度選取最佳化及自動化，並有助於震測訊號的解釋。

We propose a method of automatic velocity picking in reflection seismic using simulated annealing. Velocity picking is to pick the time-velocity pairs representing the relation of stacking velocity and time based on 2D semblance image. Conventionally seismic velocity picking was made by geophysical experts through looking at the semblance image, which is time consuming. Assuming chosen velocity pickings are on the peaks of the semblance image, velocity picking can be formulated into a combinatorial optimization problem. We get the maximal peaks on semblance by peak-climbing clustering, and we use the simulated annealing to find the global optimum in the large finite set of peaks permutations. Then we apply the obtained stacking velocity to do the normal moveout correction and stacking. Our method can automate and optimize the velocity picking and improves the seismic interpretation.