| 摘要: | 在風工程的領域中有相當多的研究項目，例如:環境風場、建物顫振、橋梁抗風…等，而建築設計風載重也是淡江大學風工程研究中心研究的項目之一，目前針對建築設計風載重專家系統以及類神經網路在建築設計風載重的應用上已有初步的研究分析與成果，然而卻沒有將這些研究內容完全整合起來，建置成一個功能完整的建築設計風載重專家系統。
本研究希望整合目前研究完成的類神經網路風力頻譜、風力係數以及氣動力資料庫，建構一個完整的建築設計風載重專家系統。藉由此系統可以讓使用者在輸入目標建物的各項資訊後，可經由氣動力資料庫搜尋風洞實驗室做過的相似案例資料，再利用輻狀基底函數神經網路預測其目標建物之風力係數及風力頻譜，並將案例資訊及預測資訊同時輸出供使用者比較，比較完成後還可根據預測出來的風力係數推導出適合之設計風載重供使用者參考。
本研究並嘗試以資料探勘的方式取代舊有的案例式推理。過去案例式推理是根據少數的實驗數據，再依照系統建造者所設定的參數評估出較相似的案例，雖然較簡潔易理解但主觀意識較高。因此本次採用資料探勘便是期望使用者可以有更多系統功能可選擇，藉以獲得更符合使用者所需要的資訊。
過去專家系統往往附加許多的程式來強化其效能，在數值計算以及圖形輸出上也都為各自獨立的程式，但這麼一來難免增加系統管理上的複雜度。如果有程式未開啟還會影響系統造成錯誤，因此本系統為了減少程式的使用量，本研究之系統使用MATLAB 來整合數值計算以及圖形輸出的功能，資料庫也只採用MS SQL Server，並以MATLAB Server Pages 取代MATLAB Web Server 來做為執行輻狀基底函數神經網路後之模擬頻譜展示網頁。
藉著MATLAB Server Pages 可以直接將風力頻譜及設計風載重的曲線直接在網
頁上模擬出來讓使用者可以更清楚明瞭，此外MATLAB Server Pages 撰寫網頁本身是由JSP 與Java 語言為主，因此對於網頁平台的開發可以統一使用JSP 語言來進行，對於系統的開發與管理來講更加便捷。
In wind engineering, there are a considerable number of research topics, for example: environmental wind field, structure fluttering, wind resistant design of bridges, etc., and building design wind load is also a research topic in the Wind Engineering Research Center at Tamkang University(WERC-TKU). Some research results with respect to wind load expert system and artificial neural network wind spectra prediction have been accumulated, but these studies have not been fully integrated into the building design wind load expert system of WERC.
This study is intended to integrate artificial neural network (ANN) and aerodynamic database to construct a design wind load expert system for buildings. The system allows
the user to input the target building’s information and searches through the aerodynamic database for similar cases, then, uses ANN to predict the target building’s wind
coefficients and wind spectra. The case information and forecasts can be output for comparison, and more complete wind load projections can be made based on the predicted wind coefficients and wind spectra.
The reported research tries to use data mining to replace case-based reasoning. Case-based reasoning is based on a small number of experimental data and preset parameters to select similar cases. Although relatively simple and easy to understand, it is very subjective. Therefore, data mining is used to provide more selections of functions and to gain more user desired information.
Previous expert systems often employed a number of additional programs to enhance their performance in numerical calculation and graphical output. However, this inevitably increases the complexity of system management. If the supporting programs are not properly running on the server, the expert system produces errors in the middle of a run. In order to reduce the use of additional programs, this research used MATLAB to provide both numerical and graphical capabilities. Also, the aerodynamic database was solely
implemented with MS SQL Server, and MATLAB Server Page was used instead of MATLAB Web Server to execute ANN simulation functions and to draw wind spectra and loading curves. In addition, MATLAB Server Page is written by JSP and Java. Therefore, web platform development can be uniformly performed using JSP, which is convenient for system development and management. |
