建築結構強震紀錄之類神經網路參數識別

Artificial Neural Network Parametric Identification for Strong Motion Records of Building Structures

林旺春(Wang-Chuen Lin)

指導教授 ： 鍾立來 ； 王安培

中原大學/工學院/土木工程學系/碩士(2006年)

https://doi.org/10.6840/cycu200600566

摘要

現今地震災害頻傳，預測地震之科技也尚未成熟，所以只能依照法規進行耐震設計以防患未然，建築結構的動態行為一直是耐震設計與分析中所考慮的重要因素之ㄧ。近年來發展了各種系統識別方法與模式，其中參數系統識別模型的建立，如ARX模型。為利用整批處理的參數推測法，不適用於非線性與時變性系統。而單純使用類神經網路建立輸出入值之關係，其權重並無法有效表達動力參數。
本研究採用遞迴參數推測(recursive parameter estimation)方法，發展出線上系統識別，即遞迴最小平方法(RARX)模型。此模型在設定初始共變異數矩陣(covariance matrix)時，往往因設定不理想，導致識別效果不佳。因此將類神經網路與RARX模型做一結合，採用單層神經元，定義其類神經權重值為估測動力參數，藉由無須設定共變異數矩陣，識別非線性時變系統。
將類神經網路與RARX模型做一結合，衍生出RARX-ANN模型。最後利用加裝阻尼器之鋼構架試驗與台東消防分隊大樓，分析非線性時變行為，由識別結果發現，因塑性行為增加與主結構破壞，會導致頻率遞減與阻尼比突增之現象，可做為一參考指標。

關鍵字

類神經網路 ； 參數系統識別 ； 遞迴參數推測方法

並列摘要

The earthquake disaster occurs frequently nowadays, but science and technology for the prediction of earthquakes is not well developed. The structures dynamic behavior always is one of the most important factors while designing and analyzing the seismic resistance. In recent years, there are lots of various system identifications methods developed, such as ARX model. ARX model is a parametric method using the batch and not suitable for the non-linear and time-varying system. Only using artificial neural network to establish the relationship between inputs and outputs, its weights could not express the dynamics parameter effectively.
The purpose of this research is to use recursive parametric estimation method to develop the on-line system identification, which is called RARX model. An initial covariance matrix has to be assigned and this initial covariance matrix will affect the identification result. Therefore, we use the neural network to modify the RARX model. The single layer neuron network is selected in this study, and the network weights are defined as the dynamic parameters. Consequently, a nonlinear time-varying system can be identified, and the initial covariance matrix is not necessary.
We derive the RARX-ANN model by combing the RARX model and artificial neural network. Experiment of steel frame with damper and Fire Department building in Taitung are taken as two examples for analyzing the nonlinear time-varying behavior in this study. According to identification results, damage of main structure and increase of plastic behavior will cause the frequency decreases and damping ratio increases. The results of this paper may offer as a reference for earthquake research later on.

並列關鍵字

artificial neural network ； recursive parametric estimation method ； parametric system identification

參考文獻

【20】 林昭葳，「結構加裝圓棒形加勁消能器之動力分析及試驗驗證」，碩士論文，國立台灣大學土木工程學研究所，臺北，2005。

【1】 McVerry, G. H., “Structural Identification in the Frequency Domain from Earthquake Records,” Int. J. of Earthquake Engineering and Structural Dynamics, Vol. 8, pp. 161-180(1980).

【2】 Beck, J. L. and Jennings, P. C., “Structural Identification Using Linear Models and Earthquake Records,” Int. J. of Earthquake Engineering and Structural Dynamics, Vol. 8, pp. 145-160(1980).

【3】 Udwadia, F. E. and Kuo, C. P., “Non-Parametric Identification of a Class of Nonlinear Close Coupled Dynamic Systems,” Int. J. of Earthquake Engineering and Structural Dynamics, Vol. 9, pp. 385-409 (1981).

【4】 Masri, S. F. and et al., “Non-Parametric Identification of a Class of Nonlinear Multidegree Dynamic System,” Int. J. of Earthquake Engineering and Structural Dynamics, Vol. 10, pp.1-30(1982).