結合類神經網路與遺傳演算法評估地震尖峰地表加速度之研究

Combing Artificial Neural Networks and Genetic Algorithms for Estimating Seismic Peak Ground Acceleration

蘇裕翔(Yu-Hsiang Su)

指導教授 ： 柯亭帆

國立屏東科技大學/工學院/土木工程系所/碩士(2014年)

https://doi.org/10.6346/NPUST.2014.00041

摘要

臺灣位於環太平洋地震帶，強震時常發生也帶來嚴重之災害，因此有關地震的研究一直受到相當之重視。本研究旨在利用計算智慧包括類神經網路及遺傳演算法兩種方法，針對臺灣24個地震分區之歷史記錄資料進行訓練、驗證與測試，再據以找出相對較佳模式，用於推估各地震分區之尖峰地表加速度值。研究結果顯示以地震因子 (地震規模、震源深度、震央距離) 及地質因子 (土壤貫入度、剪力波速) 當做輸入參數比單純以地震因子當輸入參數較好。另外結合類神經網路及遺傳演算法比單獨類神經網路模式所推估值精確度更高。本研究所採用之方法與建立之模式可以推估未設測站之尖峰地表加速度值，並可跟建築法規中之設計標準值做比較以便指出可能超出標準之分區。從24個地震分區之計算結果可以發現有3個分區雲林縣、南投縣、嘉義縣之推估值高於設計標準值，須要加以注意其可能潛在之災害。本研究之結果應可提供相關領域一些有用之參考資訊。

關鍵字

類神經網路 ； 遺傳演算法 ； 尖峰地表加速度 ； 地質鑽探

並列摘要

Since Taiwan is located at the circum-pacific seismic zone, earthquakes frequently occur and cause serious disasters sometimes. Therefore, earthquake research has become a main issue and drawn much attention. By the use of computational intelligence, including genetic algorithms and neural networks, this study aims to train, adapt and simulate historical record of twenty-four seismic subdivision zones in Taiwan, so as to find out relatively better mode for estimating peak ground acceleration of seismic subdivision zones. According to the research results, compared to utilizing seismic factors as input parameters, it shows it is better to adopt both seismic factors (earthquake magnitude, focal depth, epicenter distance) and geological factors (Standard Penetration Test value, shear wave velocity) as inputs simultaneously. In addition, it is better to combine with neural networks and genetic algorithms for estimated values than applying merely neural networks. The methods and models created in this study can predict peak ground acceleration value in stations not having been checked. And, in comparison with design values in the seismic building code that indicates, we may confirm the district that may exceed the standard. From 24 earthquake’s zones’ estimation results, we can find three regions—Yunlin county, Nantou county and Chiayi county, need to be more precautious owing to each of these areas has a higher peak ground acceleration value than the design value in the seismic building code. Consequently, The research result should contain useful information as reference to related fields.

並列關鍵字

Artificial Neural Network ； Genetic Algorithms ； Peak Ground Acceleration ； Geological Drilling

參考文獻

6. 林鈺棠，2012，以強震數據與地層特性資料建構類神經網路模式之研究，碩士論文，國立屏東科技大學，土木工程系，屏東。

30. Chen J., Lu M., Chen J., 2012, “An Adaptive Genetic Algorithm Based on Multi-Population Parallel Evolutionary for Highway Alignment Optimization Model”, Construction Technology, Vol. 41, No. 11, pp. 49-52.

32. Goldberg, D. E., 1989, Genetic Algorithms in Search, Optimization and Machine Learning. Addison Wesley Publishing Company.

34. Hebb, D. O., 1949, The Organization of Behavior: A Neuropsychological Theory, New York, Wiley.

35. Hopfield, J. J., 1982, “Neural Networks and Physical Systems with Emergent Collective Computational Abilities,” Proceeding of the National Academy of Scientists of the United States of America-Biological Sciences. Vol. 79, No. 8, pp. 2554-2558.