應用調適性類神經模糊推論系統從事鋼筋混凝土短柱之最佳化設計

本論文主要利用調適性類神經模糊推論系統從事鋼筋混凝土短柱之最佳化設計。首先應用遺傳演算法從事鋼筋混凝土短柱之最佳化設計，已知設計條件為設計軸力、中性軸深度、混凝土抗壓強度與鋼筋降伏強度、柱之無支撐長度與軸向鋼筋號數，以國內鋼筋混凝土設計規範為依據，考慮軸向力與彎矩聯合作用、鋼筋淨間距與鋼筋比的限制，建立遺傳演算法之限制式。遺傳演算法之目標函數為混凝土與鋼筋之最低總造價，設計變數為：柱斷面的尺寸，拉力鋼筋根數與壓力鋼筋根數，在執行遺傳演算法過程中程中可同時得到軸向力之偏心距。獲得多組最佳化設計之資料後，將這些資料分為三組：訓練資料、驗證資料與測試資料，作為調適性類神經模糊推論系統訓練、驗證與測試之用。訓練調適性類神經模糊推論系統，本論文應用減法聚類演算法先將資料分成幾個聚類，依聚類性質建立各輸入變數之歸屬函數及得到最少數之模糊規則。數值結果顯示本論文之調適性類神經模糊推論系統所使用之模糊規則數很精簡，介於2至7，而且測試資料之推論系統輸出值與目標值兩者之相關係數皆大於0.94，對複雜之柱設計而言精確度不錯。

關鍵字

調適性類神經模糊推論系統；鋼筋混凝土短柱；遺傳演算法；減法聚類演算

並列摘要

This thesis aims to optimally design reinforced concrete short columns by using the adaptive neuro-fuzzy inference system. Using a genetic algorithm, This thesis first works on the optimal design of reinforced concrete short columns. Given conditions are the factored axial load, neutral axis depth, compressive strength of concrete and yield strength of steel, length of the column and the size of the size of steel bars. The constraints are built based on the domestic reinforced concrete engineering design code, by considering the strength requirements of combined axial load and bending, the clear spacing between longitudinal bars and reinforcement ratio. The objective function is to minimize the total cost of steel and concrete and design variables are the column size and number of the tensile reinforcement and compressive reinforcement. The eccentricity of the axial load can be obtained while the genetic algorithm is in the process. A number of optimal data are collected and then divided into three groups: the training set, validation set and testing set, for the use of the adaptive neuro-fuzzy inference system. The data will be classified by the subtractive clustering algorithm in order to build the suitable membership functions for each inputs and the minimum fuzzy rules. Numerical results show that the number of fuzzy rules is very concise, ranging from 2 and 7. The correlation coefficients between targets and the outputs of adaptive neuro-fuzzy inference system are all greater than 0.94. The accuracy is considered to be good for the complicated column design.