Please use this identifier to cite or link to this item: `http://hdl.handle.net/11455/2725`
 標題: 類神經網路在最佳化設計之應用Applications of Artificial Neural Networks in Optimum Designs 作者: 李建宏Lee, Jiann-Horng 關鍵字: 類神經網路;Artificial Neural Network;最佳化;反傳遞網路;逆傳遞網路;結構;Optimization;Counter Propagation Neural Network;Back Propagation Neural Network;Structural 出版社: 機械工程學系 摘要: 本文針對一些最佳化的問題，應用建構的類神經網路模式，預測目標函數和限制函數的值，並配合使用數學規劃法，以求得最佳化問題的最佳解。在建構類神經網路模式的過程中，本文運用逐步縮減設計空間的策略以減少分析或實驗所需的點數並提昇類神經網路預測的準確度，以得到較佳的預測結果。 設計空間縮減的方式，是依據設計變數的數目來決定各設計變數縮減的比例，並在縮減後之設計空間內利用中心組合設計(CCD)的佈點方式來擷取訓練範例，然後將訓練範例代入反傳遞類神經網路(CPN)或逆傳遞類神經網路(BPN)進行類神經網路訓練與測試。並利用類神經網路的預測結果配合商用軟體DOT/DOC中的連續線性規劃法及分枝界限法(Branch and Bound)分別進行第一階段最佳化設計。由第一階段最佳解的分佈情況，可決定第二階段設計空間縮減的方式，並在縮減後之設計空間內建構新的逆傳遞類神經網路逐步找尋最終之最佳解，並與數學式或有限元素分析所求得的最佳化設計結果比較。This thesis deals with the optimum design problems using artificial neural networks. The artificial neural networks are constructed to predict values of the objective and constraint functions. The optimum solutions are found by using mathematical programmings and neural networks. To yield accurate neural networks, the strategy of reducing design space is used to construct more accurate neural networks. The way of reducing design space depends on the number of design variables. The central composite design (CCD) is used to create training patterns in the reduced design domain. The neural networks are then trained and tested using these data. After constructing the neural networks for objective as well as constraint functions, the optimization solver DOT/DOC is utilized to solve the first-stage optimum design problem. After obtaining the initial optimum solution, next neural network models are then constructed in an even smaller design space around the initial optimum point. The final optimum solution is found by using these newly established neural network models. The obtained optimum solutions using neural networks are compared with those obtained by finite element analyses or experimental data. URI: http://hdl.handle.net/11455/2725 Appears in Collections: 機械工程學系所

TAIR Related Article