類神經網路於高層建築設計風載重案例式專家系統之應用

The Application of Artificial Neural Networks in a Case-Based Design Wind Load Expert System for Tall Buildings

陳冠廷(Kuan-Ting Chen)

指導教授 ： 王人牧

淡江大學/工學院/土木工程學系碩士班/碩士(2008年)

https://doi.org/10.6846/TKU.2008.00243

摘要

人類自古以來在解決問題過程中遇到困難時，最常使用之方法即為從過去相近案例中學習解決問題的經驗，並以此經驗去解決目前所遭遇之問題。此方法即為案例式推理。專家系統是以領域專家長期累積的解題經驗，經歸納整理後提供他人該領域之專家意見或建議。而案例式專家系統可搜尋出與使用者欲解決問題最接近案例，並提供其解決方法或經驗。類神經網路模仿人腦神經元模式去模擬或預測複雜領域之結果。透過過去合理之案例訓練神經網路以達到學習之效果，再模擬或預測使用者欲求解問題之解答。
本研究將案例式推理和類神經網路應用於風工程領域，建構出一套應用於高層建築設計風載重之案例式專家系統。藉由此系統可提供使用者取得指定建物模型風洞實驗之風力頻譜和風力係數。若是系統中並無相同之案例，將從既有案例中推斷出最相近模型建物之風力頻譜。模擬風力頻譜部份，則是透過類神經網路將指定案例相似性最高之前若干件案例納入神經網路訓練，並模擬出指定案例之風力頻譜。
由於近年來網際網路的蓬勃發展，因此將網路資訊技術整合進此案例式專家系統，便可使大眾更容易取得專家之意見。其網路資訊主要技術為，資料庫MS SQL Server及案例式推理引擎CBR Works之結合，以及透過FLASH、Java Server Page把使用者輸入資訊、CBR Works所搜尋出之案例，及MATLAB Web Server執行輻狀基底函數神經網路後之模擬頻譜展示至網頁上。
而輻狀基底函數神經網路不論是模擬訓練或驗證數據之順風向或橫風向風力頻譜，其模擬結果除了低頻部份之誤差較大以外，其餘部份均在可接受範圍內。在模擬現今風洞試驗尚未有資料之案例的風力頻譜時，僅能以人為觀察其模擬結果是否合理。

關鍵字

風力頻譜 ； 案例相似性 ； 專家系統 ； 類神經網路 ； 輻狀基底函數神經網路

並列摘要

Since ancient times, the most common way to solve problems for human beings, who encounter difficulties, is making use of the problem solving experience of similar situations, and applies it to the current problem. The description given above is case-based reasoning. An expert system that offers users advices and suggestions of a specific domain is a system based on accumulated and deducted expert experiences. And a case-based expert system searches for the most similar case to provide solutions or problem solving experiences to the current problem. Artificial neural network is an approach to simulate or predict the results of complex domain by using similar (but highly simplified) models of the biological structures found in human brain. Training ANN with existing cases with reasonable answers, it can simulate or predict the results of problems which people who want to know.
This research constructed a case-based design wind load expert system for high-rise buildings using case-based reasoning and ANN. Users can acquire the design wind spectrums and coefficients of specific target buildings. If the same case is not in the system, it settles on the most similar one from the existing cases and offers the wind coefficients and spectrums of the similar case. For more accurate design wind loads of the target building (not one of the existing cases), pre-trained ANN learned from a group of similar ones in the case database can simulate the wind spectrums of the target building.
Due to the flourishing of the Internet, integrating web-based techniques into this case-based expert system provides more convenient consultations to the public. And the web-based techniques include combining the database MS SQL Server with the case-based engine CBR Works, and using FLASH and Java Server Page to display users’ inputs, the searching outcome of CBR Works, and the RBFNN(Radial Basis Function Neural Network) simulated wind spectrum, executed by MATLAB Web Server.
Despite the low-frequency area has higher error, both alongwind or acrosswind RBFNN simulated spectrums have little error comparing with the real data. As for using RBFNN simulating those models which aren’t under wind tunnel test yet, we could estimate that the simulation is reasonable by the opinion of experts.

並列關鍵字

Wind Force Spectrum ； Case Similarity ； Expert System ； Artificial Neural Network ； Radial Basis Function Neural Network

參考文獻

3. 陳姵君，2007，“利用類神經網路預測建築物在干擾效應下之設計風載重”，淡江大學土木工程學系碩士班論文。

23. 張錚，2002，“MATLAB教學範本程式設計與應用”，知城數位科技股份有限公司。

1. John D. Holmes, 2001, “Wind Loading of Structures”, SPON PRESS.

9. A.C. Khanduri, C. Bédard, T. Stathopoulos, 1997, “Modeling wind-induced interference effects using backpropagation neural networks”, J. Wind Eng. Ind. Aero., Vol. 72, pp. 71-79.

10. Peng Huang and Ming Gu, 2005, “Experimental study on wind-induced dynamic interference effects between two tall buildings”, Wind and structures, Vol. 8, No. 3, pp. 147-161.