Please use this identifier to cite or link to this item: `http://hdl.handle.net/11455/1925`
 標題: 可處理離散和混合變數之演化策略法Discrete and Mixed-Variable Evolution Strategy 作者: 陳信昌Chen, H.C. 關鍵字: 演化式最佳化;RNES;重組率;突變率;離散變數;混合變數 出版社: 機械工程學系 摘要: 摘要比起其他演化式最佳化方法，RNES在適應值的計算上相對的簡單，個體的適應值計算方式是基於個體在各目標下的排名和擁擠程度而定。但RNES在一些參數的設定使用或機制的選用上仍有改善空間，因此本文將嘗試修改、新增演化過程中的參數與機制，希望達到改善演化效率，提供穩定的求解能力。這些修改包括了對重組方式的新增、重組率與突變率的引用、外部菁英族群數量的控制機制改善等，同時也進行題目的測試與效能評估。經過不同類型、目標數的題目測試後，新增或修改後的RNES都有不錯的效率改善與求解能力。實際上的最佳化問題設計變數大多是受到數量與範圍拘限的離散變數，因此本文還針對RNES設計了三種處理離散變數的方法，另外也引用在其他演化式方法中使用的離散變數處理方法卜瓦松分佈亂數方法共四種方法，使得RNES除了可以處理實數變數之外，還可以同時處理離散變數或是混合離散、整數、連續變數等問題。本文以數個不同複雜程度的題目進行測試，測試的結果顯示，RNES配合本文所提出的三種處理離散變數方法在處理具離散變數或混合變數的最佳化問題上都有不錯的求解能力與效率。AbstractCompared with other evolutionary algorithms the fitness computation in multiobjective solver RNES is relatively simple. The fitness is computed based on the ranks and the crowding status of the individual. But the parameter settings and the evolutionary operators still have rooms to improve. This thesis tries to eliminate the drawbacks of RNES to increase the efficiency and capability of finding better solutions. These efforts include adding new recombination operators, introducing mutation probability and recombination probability and simplifying clustering operation. Some problems are used to test modified RNES and the results are satisfactory.Many real-life optimization problems contain discrete variables and constraints. In addition to previous improvements this thesis also introduces three methods to treat discrete variable problems. Besides those three methods developed in this thesis one method from other paper using random number of Poisson distribution to treat discrete variables is also tested. The RNES with these discrete variables treating methods can solve not only continuous variable problems but also mixed-variable problems. Several test problem with different characteristics are used to test the methods proposed in this thesis. In general the outcomes show the methods proposed indeed can solve those problems efficiently. URI: http://hdl.handle.net/11455/1925 Appears in Collections: 機械工程學系所