透過您的圖書館登入
IP:3.17.184.90
  • 學位論文

數學規劃模式能量函數建立之探討

Energy Functions For Mathematical Programming Models

指導教授 : 古思明
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


本研究探討利用類神經網路動態系統求解數學規劃問題中能量函數的建 立。研究中除了回顧五個相關網路模式之外,並根據文獻[4]中的所謂二 階段能量函數的觀念建立組合的能量函數,且將其階段切換方式由時間改 為由決策變數的位置來完成。研究中還引用了零工工廠排程問題為例,分 別運用懲罰函數法、Rodriguez-Vazquez模式、乘數法、與前述的組合能 量函數等四種方法來建立能量函數,並驗證比較其求解績效。由實例的驗 證結果我們發現Rodriguez-Vazquez模式優於懲罰函數法,而乘數法與組 合能量函數則更優於此二種方法,且對於懲罰參數的大小較不敏感。

並列摘要


This research focuses the energy functions arising from the application of artificial neural networks to solve mathematical programming problems. In particular,we combine penalty function method and multiplier method to formulatea mixed energy function,which was motivated by the two-phase model in the literature. Unlike the phase switching by a proper time in original two-phasemodel, we modify phase switching method by the location of decision variables.We take job shop scheduling problem to compare the the performance among energyfunctions- penalty function,Rodriguez-Vazquez model,multiplier method, and mixed energy function. We find that Rodriguez-Vazquez model is superior to penaltyfunction method. Furthermore, multiplier method and mixed energy function are better than the former two functions and are insensitive to the values of thepenalty paramrters.

延伸閱讀