螞蟻演算法最佳化類神經網路於土壤液化評估之研究

Ant Colony Optimization Algorithms Optimizes Neural Networks on Prediction of Soil Liquefaction

李元宏(Yuan-Hung Li)

指導教授 ： 左天雄

國立臺灣大學/工學院/土木工程學研究所/碩士(2009年)

https://doi.org/10.6342/NTU.2009.01221

摘要

大地工程所面對的問題存在著許多不確定與不規則的非線性特質，其因果關係錯綜複雜，不是以簡單的數學公式就可以展現出來。一般對土壤液化(liquefaction)的預測都是以經驗公式來判斷是否合於評估法的安全係數來做為判斷的標準。然而在各種的評估法中所考慮的土壤液化強度參數卻可能不是可以用簡單的經驗法則來計算，因為參數間彼此可能為一種高度複雜且相互影響的關係連接著。因此我們有必要去尋找一種可以做非線性連結的系統來做為分析土壤液化的工具。而類神經網路系統中，多層次的網路架構就是為了提供處理複雜非線性問題的能力。欲建立系統的反應機制，我們藉由輸入、輸出的觀測資料以建立輸入與輸出訊號間的最佳映射關係。
本案例研究搜集了歷年世界各地地震之現地液化觀察資料共208組，分為146組訓練資料(70 %)及62組測試資料(30 %)分別放入所架構之倒傳遞類神經網路系統。並用螞蟻演算法來最佳化網路架構，找出土壤輸入參數與現地液化資料間之最佳化連結。最後將輸出結果與傳統的簡易CPT液化評估法，如Robertson法、Shibata法作一誤判率之比較。結果無論是在液化、非液化或整體誤判率皆能有更為優異的表現。

關鍵字

類神經網路 ； 螞蟻演算法 ； 土壤液化

並列摘要

Geotechnical problems have many unconfirmed and irregular nonlinear characteristics. The causalities are complex and can not be displayed by simple mathematic formula. The prediction of soil liquefaction is usually estimated by formula of experience to decide whether it correspond to safety coefficient. The soil parameters may not be calculated only by simple experience rules because there are complicated relationships between the parameters which are mutual effect. We have to find a tool with nonlinear system to estimate the soil liquefaction problems. The multilayer structure of artificial neural networks is use to deal with the complex nonlinear problems. We establish the response mechanism by observing the input-output pairs.
The study collects 208 observing data of earthquakes over the world. Classifying 70% to training set and 30% to testing set by random. Training the data using back- propagation neural networks which is optimized by ant colony optimization algorithms to find the best network parameters. Comparing the results of net output with the results of experience formula, the results of networks have better display.

並列關鍵字

Neural Networks ； Optimization ； Soil Liquefaction

參考文獻

[39] 羅華強（2001），”類神經網路—MATLAB的應用”，清蔚科技股份有限公司。

[1] Dai, H., and Macbeth, C.,(1995) ”Automatic picking of seismic arrivals in local earthquake data using an artificial neural network,” Grophys. J. Int. 120, pp.758-774.

[2] Ellis, G. W., Yao, C., Zhao, R., and Penumadu, D.,(1995) “Stress-strain modeling of sands using artificial neural networks,” Journal of Geotechnical Engineering, Vol. 121, No. 5, pp.429-435.

[3] Goh,A.T.C.(1993).“Seismic Liquefaction Potential Assessed By Neural Networks,”Journal of Geotechnical Engineering,Vol.120,No.9, pp,1467-1480.

[4] Hebb D.O.,(1949),“The Organization of Behavior,”A Neuropsychological Theory. New York : Wiley.