本文應用倒傳遞網路模式以發展類神經網 路的線性最佳化設計,其主要步驟為(1)數學公 式化(2)網路訓練及測試(3)縮小全域搜尋及最佳 化,收斂到最佳值為止。本文所提出的方法,可 免除傳統simplex法當設計變數及拘束條件大量 增加時的繁複數值計算,其亦和simplex法相同的 得到全域最佳解。網路的繼續發展及改良,將 有助於提高最佳值的精確度。文中以五個線性 雙變數函數的最佳化設計問題為例進行模擬與 分析,以驗證所提的類神經網路線性最佳化設 計方法。 The paper presents an approach of Artificial Neural Network (ANN) in Linear Programming (LP) for engineering design. Back Propagation Network (BPN) is applied to this approach with 2-0-1 structure. One can divide this method into three primary steps, that are mathematical formulation, training & testing of network, and global optimum search. A technique of reducing searching area has been introduced for accelerating convergence. This approach can straightly produce a considerable satisfying solution to avoid the large numerical computation and analysis in conventional Simplex Method. This developed algorithm is based on the N- dimensional design space. For the clear expression, several 2-D design examples are presented to demonstrate the method and the feasibility of the proposed approach. It is noted that the method can be applicable to higher dimensional design problems.
中國機械工程學會第十屆學術研討會控制組論文集=Proceedings of the Tenth National Conference of the Chinese Society of=Mechanical Engineers，頁141-147