應用倒傳遞神經網路模擬非線性遲滯動力行為

Abstract

結構物在受到劇烈的外力作用或產生較大變形時，會有非線性及非彈性的行為發生。而在上述狀況下，結構的恢復力(Restoring force)有遲滯(Hysteretic)現象及勁度、強度上的退化產生。文獻中針對描述結構遲滯回復力行為的研究，已有許多不同模式被發展完備，但這些模式也造成當合併求解運動方程式時，有相當大的困難。本文嘗試用類神經網路中倒傳遞(Backpropagation, Bp)神經網路，在時域分析下模擬含遲滯回復力之結構運動行為。文中採取一般常用的Bouc-Wen遲滯回復力模式，合併運動方程式所產生的資料以作為類神經網路訓練之案例，根據本研究對單自由度結構在擬地震下之模擬結果，發現能有頗為良好的近似成效。於是，利用類神經網路的模擬方式便可取代一般傳統分析遲滯模型的方法，直接從已知的結構反應資料中，找出結構系統的性質，進而分析在各種外力作用下此結構的動力反應值。

Structures under severe loading or large deflection would have inelastic and nonlinear behavior. Hence, hysteretic and deterioration of stiffness and strength in structural restoring force is presented. Recently, the study of structural hysteretic behavior had been expanded in several different mathematical models in literature. However, the nonlinear behavior for these models is still hard to solve. This work presents an application of backpropagation (Bp) neural network to analyze the structural behavior with hysteretic restoring force in time domain. The general hysteretic model, Bouc-Wen model, are employed to generated training instances for Bp network. A SDF system with five different hysteretic models with different external loading have been investigated in this work. The results show that the Bp network can identify the property of structures and simulate the nonlinear behavior. Sensitive analysis is also utilized in this work to analyze the correlation between input and output parameters. Two different external loading and used to verify the performance of Bp network. The results demonstrate that the nonlinear behavior of the system also can be identified even the loadings are new inputs for the system.