實數編碼利基遺傳演算法應用於河川不恆定流模式阻力參數率定之研究

詹明修(Ming-hsiu Chan)

指導教授 ： 蔡丁貴

國立臺灣大學/工學院/土木工程學研究所/碩士(2004年)

https://doi.org/10.6342/NTU.2004.01966

摘要

遺傳演算法(genetic algorithm) 係由Holland在1960年代模仿自然界「物競天擇，適者生存」進化現象而發展出來的一種搜尋法則。本文修改Goldberg 與 Richardson(1987)提出之二位元編碼利基遺傳演算法，發展一實數編碼利基遺傳演算法以搜尋多峰多谷函數之全域最小值。為了表現本演算法的優異性，本文比較本文演算法、實數編碼遺傳演算法、二位元編碼利基遺傳演算法及二位元編碼遺傳演算法之收歛結果。收斂結果顯示：本文發展之實數編碼利基遺傳演算法是四種方法中最佳的。
然後本文利用實數編碼利基遺傳演算法自動優選「全流域河川不恆定流模式」(賴經都, 1999)中最佳阻力參數。本文首先進行模式阻力參數之率定工作，然後再進行模式之驗證工作。本文所欲推求最佳化參數之河系為淡水河全流域河系，依河道特性分為17個河段(reach)。每個河段有一個阻力參數，共有17個參數需率定。目標函數為令觀測水位與模式計算水位之均方根差(root-mean-square error)為最小；其限制式則為「全流域不恆定流模式」及阻力參數之上下限。
研究結果顯示：率定所得的阻力參數，可使模式之計算水位與觀測水位相當吻合且每次率定得到的阻力參數值均十分相近。以另一時段觀測資料進行模式驗證時，發現計算水位與觀測水位亦十分接近。可見本文發展之實數編碼利基遺傳演算法的確是一客觀且能正確找出全域最佳解的自動化參數率定方法。

關鍵字

實數編碼利基遺傳演算法 ； 遺傳演算法

並列摘要

Based on imitation mechanics of “the natural selection and the survival of the fittest”, genetic algorithm was developed by John Holland in the 1960s. In this paper, a real-valued-coding niche genetic algorithm modifying the binary-coding niche genetic algorithm (Goldberg and Richardson, 1987) is developed for searching global minimum value of a multiple-peak function efficiently. To demonstrate the advantage of this developed algorithm, we compare present methods with real-valued-coding genetic algorithm, binary-coding niche genetic algorithm and binary-coding genetic algorithm. The results show that the real-valued-coding niche genetic algorithm developed in this thesis is the best among the four methods mentioned above.
The second purpose of this thesis is to automate the calibration process of CCCMMOC model (Lai, 1999) by incorporating a real-valued-coding niche genetic algorithm. We first calibrate the resistance coefficients of CCCMMOC and then verify them. The target river system is Tamsui River system which is divided into 17 reaches. Each reach corresponds to a resistance coefficient. The objective function is root-mean-square error of the difference between the measured and calculated water levels at all observation stations. The constraints are 1-D unsteady flow model (CCCMMOC), and the upper and lower limits of resistance coefficients.
The results are promising because the resistance coefficients calibrated by present algorithm with different initial population points converge to the same values. On the other hand, the simulated results of water levels are in good agreement with the observed whether in the calibration case or the verification case. Therefore, the real-valued-coding niche genetic algorithm developed in this thesis is a good automatic calibration method which can find the global optimum accurately and objectively.

並列關鍵字

real-valued-coding niche genetic algorithm ； genetic algorithm

參考文獻

3. Becker, L. and Yeh, W.W-G. (1972). “Identification of parameters in unsteady open channel flows,” Water Resources Research, 8(4), pp. 956-965.

4. Becker, L. and Yeh, W.W-G. (1973). “Identification of multiple reach channel parameters,” Water Resources Research, 9(2), pp. 326-335.

7. Du, D. -Z., and Paralos P. M. (1998). Handbook of combinatorial optimization, 3, Kluwer Academic Publishers.

8. Goldberg, D. E. (1989). Genetic algorithms in Search, Optimization, and Machine Learning, Addison-Wesley Publishing Co., Reading, MA.

9. Grefenstette, J. J. (1960). “Optimization of Control Parameters for Genetic Algorithms,” IEEE Transactions on System, Man, and Cybernetics, SMC-16(1), pp122-128.