以基因演算法同步決定一維構造之各層速度與厚度

Abstract

一維速度構造是地震研究之重要基礎資料，舉凡快速定位、波形反演等工作皆有賴於此，於是需要以可靠且有效率的演算法建立可信賴的速度構造。全域搜尋的演算方式雖然最為直覺，也最為可靠，但在資料量龐大的地球物理研究中，大量計算資源需求往往成為實際應用上的最大窒礙。現存的應用程式中，則大多使用線性計算的逆推方式。然而，這種方式除了有著容易受到先驗知識影響的缺點，且難以同時決定各層之厚度與速度。因此，本研究採用基因演算法，期望以此非線性的逆推方式，有效率地建立可靠的一維速度構造。本研究取用臺灣嘉義地區1991年至2006年地震的P波與S波資料，並以福傳程式語言撰寫程式來分析。結果顯示地震之分布密度會嚴重影響演算結果的可靠程度，但在地震資料充足之區域，基因演算法確實可在穩定的演算下大幅節省計算資源與時間，快速找出速度不連續面之深度。另外，藉由演算結果的比較，推測CHY測站底下存在迥異於周遭的區域構造。然而，因為基因演算法帶有隨機過程，難以確保得到最佳解與否，因此本研究建議應該將基因演算法的結果做為初始模型，進一步結合別的逆推方式做最佳化為宜。

Layered velocity model is important basic information for seismology. Global search and calculus-based inversion are commonly used to determine layered models. Global search offers accurate solutions but usually takes much time calculating. On the contrary, calculus-based inversion saves time but usually comes with the local minimum problem. Genetic algorithm (GA), which has been used in many fields, probably is an appropriate method to balance between accuracy and time saving. The purpose of our research was to determine velocities and thicknesses of layered velocity model simultaneously with GA method. In this study, P phase and S phase arrivals were used. Fortran program codes were written and tested with earthquake data of Chia-Yi area, Taiwan from1991 to 2006. The results showed that the velocity models calculated by GA are strongly affected by the density distribution of earthquakes. However, with enough data, our approach balanced well between efficacy and efficiency. The results suggest that GA could be a good way to obtain reliable initial velocity models for geophysical studies.