標題: 應用類神經網路模擬鋼筋混凝土房屋結構容量震譜
Application of Artificial Neural Network Model for Simulating
作者: 陳彥伶
Chen, Yen-Lin
洪士林
Hung, Shih-Lin
土木工程學系
關鍵字: ATC-40;容量震譜;類神經網路;耐震性能設計法;性能目標;基因演算法;ATC-40;capacity spectrum;artificial neural network (ANN);performance-based seismic design;genetic algorithm (GA)
公開日期: 2009
摘要: 結構物之功能與型態日趨複雜化,傳統耐震設計已不能滿足各方面需求,於是發展出耐震性能設計法。根據結構性能需求可將其功能特性區分為五項,分別為安全性、耐久性、經濟性、環境性與適用性。近年來,房屋結構耐震性能設計法趨於設計主流,開始有人提出在滿足耐震性能目標"生命安全"下,同時考慮滿足目標"最小成本"要求的結構多目標最佳化設計。本研究之目的乃以類神經網路(ANN)模式來模擬鋼筋混凝土房屋結構之容量震譜,以作為在考量安全與維護成本之多目標基因演算法(GA)鋼筋混凝土房屋結構最佳化用。整體設計原則係依據美國ATC-40耐震性能設計法進行考量,以容量震譜法進行設計評估,其中引入位移指標對結構的耐震性能進行控制概念。本研究首先利用SAP2000建立719棟五層樓到十層樓的房屋結構容量震譜為ANN模式之案例,其中638棟為訓練案例,其餘為驗證案例。最後結合GA結構多目標設計進行測試,探討容量震譜之ANN模式所模擬的結構物之容量震譜對房屋結構功能性設計的影響。由研究結果證實,ANN可有效模擬3D RC房屋結構之容量震譜。而當訓練案例加入變異後確實會影響GA最佳化結構設計產生的性能點,且加入變異的容量震譜若提高,則最後最佳化結果的性能點也偏高,造成不保守情況產生。
The requirement-function of building structures is getting multifarious; hence conventional seismic design approach cannot completely satisfy all requirements. Newly, performance-based seismic design is developing and turns into a novel design scheme. The functionality of building structures can be categorized into different categories, such as safety, durability, economic, environmental, and utilizable. Recently, the approach of performance-based seismic resistance design of buildings gets more paying attention. Besides satisfying seismic resistant for safety purpose, constructing with minimum cost is also an important issue for multi-objective optimization of buildings. The aim of this work is applying artificial neural network model to simulate the capacity spectrum of concrete building structures. The simulating results are then used for a genetic algorithm (GA) model for optimization design of reinforced concrete buildings under multi-objectives, the safety and maintenance cost. The complete design principle is based on US ATC-40 performance-based seismic design code. First, 719 cases of 5-storey to 10-storey buildings capacity spectrums are created with SAP2000 for ANN model. 638 cases are used for training and the rest are used as verification cases. Finally, the ANN model is integrated with GA multi-objective optimization design to verify the affect of the simulated results generated via ANN model. The verification results confirmed that ANN can effectively simulate the capacity spectrum of 3D RC building structures and the final design through GA model is acceptable. Also the effect of noise in data is investigated and shows that noise may influence the performance of the solution searched by GA.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079616506
http://hdl.handle.net/11536/42228
Appears in Collections:Thesis


Files in This Item:

  1. 650601.pdf