摘要
本計畫的主要研究工作有二個部分，第一個部分是整理早期的地震資料以建置一個統一的地震目錄。第二個部分是，鯉魚潭水庫壩體結構的系統識別。
第一個部份的研究工作，首先是整理台灣地區的地震觀測史及不同時期的地震目錄和資料，再利用觀測期重疊的地震規模資料，推導出一些規模轉換公式並驗證其可信度，然後將不同時期的地震規模全數轉換為近震規模(ML)及震矩規模MW。本研究在整理早期的地震資料時，曾去除24個因為錯讀S波到時而誤定為台灣地區的地震，也整理了規模6.0以上的地震之斷層面解，以及四個早期地震的災害分布圖及一些災害狀況像片，並特別仔細檢驗四個大規模地震的規模可信度，及將其中一個(1920年花蓮東方外海的地震)重新定位。最後，本研究計畫將整理後的地震目錄、斷層面解、以及地震災害分布圖和災害狀況像片，包裝在一個植基於地理資訊系統的「台灣地區地震搜尋系統」裡。
第二個部份的研究工作，首先是應用類神經網路及水庫的強震資料來建立鯉魚潭水庫壩體的動力行為模式，其中時間域方法NARX神經網路訓練誤差小，可以精確模擬水庫壩體的動力行為模式，並研判壩體在921地震中沒有明顯的損壞；而頻率域方法則因訓練誤差較大，無法以預測誤差進一步研判壩體是否在921地震中有明顯的損壞。此外，本研究也應用隨機子空間識別法識別大壩之自然頻率、模態阻尼比等動態特性參數。應用921地震前後二個地震資料識別得到之水庫壩體自然頻率雖有差異，但因水庫壩體的輸入除了地震力之外還有壩體水位所造成的壓力，而水位對壩體頻率應有某些程度的影響，因此尚須壩體水位資料以進一步分析系統以判定魚潭水庫壩體是否可能損壞。

Abstract
This project has two major objectives. The first is to establish a uniform earthquake catalogue for Taiwan, while the second is to conduct structural system identification for Li-Yu-Tan dam in Miao-Li County.
The history of Taiwan earthquake observation and various earthquake catalogues and data collected were reviewed. Then some empirical formulas for magnitude conversion were derived, with the data set which appeared in different catalogues, and their effectiveness were checked. Finally, a uniform earthquake catalogue with local magnitude ML and moment magnitude MW was established by these conversion formulas. In the process, 24 earthquakes were deleted from the catalogue because they were accidently recorded to have occurred in Taiwan due to the misinterpretation of S arrivals. Four major earthquakes were re-examined for their magnitudes and one of them was relocated as well. The catalogue also includes fault plane solutions of earthquakes with magnitude equal to or greater than 6. The damage distribution maps and some damage pictures of four historical and earlier period earthquakes were also edited. The uniform earthquake catalogue, including fault plane solutions of earthquakes with magnitude equal to and greater than 6, and the damage distribution maps and damage pictures were finally integrated into a GIS based “Taiwan earthquake data search system”.
In the second task, the dynamic behavior model of Li-Yu-Tan dam using artificial neural network and strong earthquake data was first established. Through the time-domain method, NARX neural network has minimal training error and can accurately simulate the dynamic behavior of Li-Yu-Tan dam. According to time-domain method results, Li-Yu-Tan dam was diagnosed to have suffered no obvious damage during the Chi-Chi earthquake. The training error of artificial neural network is larger in frequency-domain method, and therefore it is difficult to diagnose whether or not Li-Yu-Tan dam suffered damage during the Chi-Chi earthquake. Besides, the stochastic subspace identification method was also applied in this study to identify modal natural frequencies and damping ratios using earthquake data before and after the Chi-Chi earthquake. There was little discrepancy between the identified natural frequencies before and after the Chi-Chi earthquake. Since the input forces of the dam include earthquake force and water pressure, water levels should have a certain influence on natural frequencies of the dam. Therefore, water level data is still necessary to further diagnose whether or not Li-Yu-Tan dam was damaged during the Chi-Chi earthquake.