Studies on structural identification and inteligent control system of seismic response

本研究的主要结果可以总结如下：1。提出了一种利用扩展卡尔曼滤波（EK-WGI）识别多自由度结构滞回退化模型结构参数的方法。数值结果表明，即使在严重非线性的情况下，该方法也能很好地预测结构的线性和非线性行为。然而，估计的参数并不十分令人满意，特别是对结构响应不敏感的非弹性参数。提出了一种通过学习相关动态系统实现自动关联的方法，该方法用系统动力学模型表示。为了证明这一逻辑问题的正确性，进行了一个单自由度线性系统辨识的数值试验。在SDOF模型骨架精确的情况下，对线性SDOF系统进行了稳定辨识。然而，在采用一个更复杂的SD模型的例子中，我们观察到，尽管正确地识别了水平分量，但每个环节的权重值往往会收敛于其他值，甚至发散。似乎需要将一些数学设备集成到SD模型中。在这方面，神经网络的关联技术可以作为一种手段。为了获得有规律的定向学习，有效地利用结构学习作为参数和系统辨识问题的一种替代方法。提出了一种全局和局部的迭代过程，以获得循环网络学习中稳定和快速的突然性。提出了一种利用自适应控制理论合成最优控制器的方法。最优控制的设计过程包括结构辨识和控制器综合。提出了加权局部迭代迭代线性滤波平滑器（WILFS）用于结构参数的估计。基于该算法所识别的物理模型，易于合成最优控制器。通过计算机仿真，验证了WILFS方法对最优控制的有效性。少

The main results of this study may be summarized as follows :1.A metod was presented to identify the structural parameters on a hysteretic and degrading model for multidegree of fredom structure by the Extended Kalman filter (the EK-WGI method). Numerical results show that the present method predicts the linear and nonlinear behavior of the structure vary well even for the case withsevere nnlinearity. However, the estiated parameters are not quite satisfactory, particulary for inelastic parameters which are insensitive to the structura respnse.2.A undating technique to autoassociate by learning a concerned dynamic system, which is represented by a system dynamics odel, was proposed. To justify the ogical problem, a numerical test was carried out for identification of a SDOF linear system. In example in which the skelton of a SD mode is precise, stable idetification was made on a linear SDOF system. However, in example in which a ore complicated SD model is employed, it was observed tha … More t values of weights n each link tend to converge other vaues than exected or even diverge, althugh level components are identified properly.3.It seems that some mathematical device needs to be integrated into the SD model. As this regard, neural network's atoassociate techniques might serve as the means. In order to obtain a regularly directed learning, effective use of the structural learng was cnsidered as an alternative methd for parmeter and system identificatin problems. A global and local iteratin procedure was proosed to obtan stable ad fast cnvergency in the learning of recurrent networks.4.A method to synthesize an ptimal controller by using adaptive control theory was presented. The design procedure of the ptimum control consists of an identifiation of the structure and a synthesis of the controller The iterated linear filter-smoother with weighted local iteration (WILFS) was proposed to use in the estimation of the structural parameters. The otimal controller was easy to be synthesized based on the hysical model identified by this algorith. By carrying out comuter simulations, the usefulness of WILFS method for the optimum control was veified. Less