台灣位於環太平洋地震帶上,以致於地震發生頻繁,且其中不乏具 破壞性之地震,因此有關地震的相關研究就顯得特別重要。本研究利用 類神經網路將地震測站所記錄之數據與當地地質資料模式化,從中獲得 一相對較佳模式,用以推估台灣24 個地震劃分區域之地表尖峰加速度值。 模式輸入之參數,在地震數據方面包括震源深度、震央距離、與地震規 模,在地質資料方面則包含剪力波數與標準貫入度,合計共五項輸入因 子。研究結果顯示有四個區域即雲林縣、南投縣、嘉義縣與嘉義市,其 類神經網路模式所推估之結果超出建築法規所制定之設計值,需要加以 特別注意。另外本研究也建立了水平地表尖峰加速度值與震源距離之關 係式,由於引進地質特性,所以本研究所得結果應比前人之研究更具信 賴度,這些計算結果應可提供相關研究或工程人員後續發展之參考。
Ground strong motions are frequently occurred in Taiwan as this island is located in the Circum-Pacific seismic zone, and thus the topic related to earthquake problems is shown to be of great importance. In this study, the approach of neural network is adopted to model seismic records from checking stations with the effect of local geological conditions. A relatively better model is obtained for estimating peak ground acceleration at twenty-four seismic subdivision zones in Taiwan. The input parameters including five items, which are focal depth, epicenter distance, and local magnitude of earthquake records, as well as shear wave velocity and standard penetration test value from onsite geological surveys. The studied results show that four regions including Yunlin county, Nantou county, Chiayi county, and Chiayi city needed to be more precautious as each of these areas has a higher neural network estimation of peak ground acceleration than that of design value in the seismic building code. In addition, a relationship between horizontal peak ground acceleration and focal distance is developed, which may exhibit a more reliability than previous studies as it did consider geological characteristics in the model. These computational results may provide useful information for relevant engineering works and latter researches.