【作者】 陈太聪；

【导师】 韩大建；

【作者基本信息】 华南理工大学 ， 结构工程， 2003， 博士

【摘要】 近几十年来,随着系统概念的逐渐普及,旨在研究系统建模、分析和综合的控制理论相应在工程技术的多个学科领域得到了越来越广泛的应用。不同于经典控制理论,现代意义上控制理论的分析对象已由单输人单输出系统扩展到多输入多输出的系统结构,相应的控制方法也由简单的开环控制向闭环反馈控制以及更高层次的自适应控制发展。作为当前控制理论的研究热点,自适应控制包括模型参考自适应控制和自校正控制两个分支。前者于20 世纪60 年代建立,采用自适应机构来克服由系统模型非确定参数引起的系统输出的非确定性,相当于高级的闭环反馈控制,是自适应控制思想的初级发展形式,在结构工程领域的抗震抗风与振动控制中已有较多应用。后者于70 年代提出,通过在线系统辨识,估计系统模型参数,进而修改控制参数,以使系统适应环境的变化,这一涉及系统本质的控制思路是自适应控制的高级发展形式,在结构工程领域仍鲜见专门研究。作为自适应控制思想的理论补充和应用研究,本文首次采用自校正机制的自适应控制思想对结构工程问题进行研究,得到了非确定性参数分析的具一般意义的若干新方法,并将其应用于复杂工程问题——大跨度斜拉桥结构的施工过程控制,获得了理想的效果。不同于一般的自适应控制研究仅着眼于系统参数的辨识工作,本文力求从系统建模的一开始即对系统参数重要性进行分析评价,从而建立起有限完备观测和有限完全控制系统,进而进行系统主要参数的辨识研究。研究过程中,通过应用人工智能手段和随机结构概念,得到了具一般性的结构参数重要性(灵敏度和非确定性)分析方法以及时变随机结构的参数连续识别方法。本文旨在针对复杂结构工程的控制问题,建立一套具一般意义的、自包含的非确定性参数分析理论,以实现自校正机制的自适应控制思想。全论文共6 章。其中第2 章(结构系统参数灵敏度分析新方法研究)、第3章(结构系统参数非确定性分析新方法研究)和第4 章(时变随机结构系统的参数连续识别新方法研究)内容统编为非确定性参数分析的理论研究篇;第5 章(崖门大桥工程概况和大跨度斜拉桥的确定性多阶段施工仿真分析研究)和第6 章(崖门大桥的自适应施工控制实践)内容统编为工程应用篇。本论文的理论创新和主要工作如下: 1. 首次系统地将自校正机制的自适应控制思想应用于结构工程领域,通过结构参数重要性分析和时变随机结构的参数连续识别研究,建立起了一套具一般意义、自包含的非确定性参数分析方法。2. 在结构系统的参数灵敏度分析中,针对在一定区域内变化的一般性非确定参数,采用人工智能手段——人工神经网络(三层感知器模型)的泛化映射机制近似模拟结构参数和结构响应间的非线性关系,进而由神经网络的结构参数构造更多还原

【Abstract】 During the last several decades, with the worldwide acceptance of system concept, the system control theory focusing on the studies on system modelling, analysis and synthesis has been successfully applied in various engineering fields. In the modern control theory, which is different from the classical one, the SISO system as the control object has been extended to the MIMO system and respectively, the simple open-loop control theory as the control method has been developed to the close-loop feedback control theory and recently to the adaptive control theory. As an advanced control theory, the adaptive method has arisen many interests in current system control researches. The adaptive method, according to different control ideas, can be mainly classified into two categories: model-reference adaptive method and self-tuning adaptive method. As for the model-reference adaptive method which was set up in 1960s, use is made of adaptive device to conquer the system output uncertainty brought by the uncertain system parameters. From that meaning, the model-reference adaptive method can be regarded as a high-level type of close-loop feedback method and as a primary type of adaptive method. In the field of structural engineering, many applications of this method have been found in the structural vibration, earthquake and wind responses control practices. As for the self-tuning adaptive method which was proposed in 1970s, use is made of system identification algorithm to estimate the uncertain system parameters in a real-time way, so that the estimated and adjusted system model can adapt itself to the varying environment. Since such a method deals with the nature of the controlled system, the self-tuning adaptive method is then regarded as an advanced type of adaptive method and being widely used in many engineering fields in recent years. However, in the field of structural engineering, there still lacks of special study on this adaptive method. As a theoretical supplement to the adaptive control theory and a corresponding engineering application study, for the first time, this dissertation investigates how to apply the self-tuning adaptive idea into structural system control problems. In the investigation, several general-meaning theoretical methods for analyzing uncertain structural parameters are proposed. And finally, the proposed self-tuning control idea is successfully applied into the control practice of complex engineering problem —construction process control of long-span cable-stayed bridge. Generally, in a common adaptive control study, the main content may be a parameter estimation of the system model. The present study, however, focuses on the system modelling, as well as the parameter estimation. Hence, an application of adaptive idea is divided into two steps: firstly, to evaluate the different importance of different system parameters so as to set up a finitely complete observation and finitely complete control system model considering the parameter importance; secondly, to estimate those parameters of the system model as accurately as possible. In this study, the artificial intelligent tool and stochastic structure concept are applied to develop general-meaning theoretical methods for evaluating structural parameter importance (i.e. sensitivity and uncertainty) and sequentially estimating parameters of time-variant stochastic structure. The main purpose of this dissertation, in a word, is to establish a general-meaning and self-contained set of uncertain parameter analysis theory to realize the self-tuning adaptive control idea in complex structural engineering problems. The whole dissertation is composed of six chapters. Among them, Chapter 2 (new method for parameter sensitivity analysis of structure system), Chapter 3 (new method for parameter uncertainty analysis of stochastic structure system) and Chapter 4 (new method for sequential parameter estimation of time-variant stochastic structure system) are included into one part named theoretical studies; Chapter 5 (general situation of the Yamen Bridge and deterministic simulation analysis of cable-stayed bridge construction process) and Chapter 6 (non-deterministic parameter analysis practice in the self-tuning control of the Yamen Bridge construction process) are included into another part named engineering applications. The theoretical innovations and main achievements of this dissertation can be summarized as follows: 1. The self-tuning adaptive control idea is first introduced into structural engineering problems. Through studies of structural parameter importance analysis and sequential parameter estimation of time-variant stochastic structure, a general-meaning and self-contained set of uncertain parameter analysis theory is established. 2. In the parameter sensitivity analysis of structure system, the general uncertain parameters varying in a certain domain are concerned. Use is made of artificial intelligent tool —Artificial Neural Network (ANN) —to approximate the nonlinear mapping mechanism between system parameters and system outputs, and then, a uniform formula composed of ANN’s parameters is derived to approximate the first-order sensitivity indices. As compared to the common-used finite-difference approximation method, the proposed method can provide more comprehensive andmore reliable sensitivity indices. While compared to the analytical method, the proposed method shows advantages of uniformed program coding and good feasibility. 3. Fast training algorithm of ANN is well known to be an essential premise assuring the wide applications of ANN. Concerning the disadvantages observed in the standard Levenburg-Marquardt (LM) algorithm which is mostly common-used in training Multi-Layer Perceptron, the present study proposes a Modified LM algorithm, in which the decay rate of training parameter varies adaptively during the ANN training process and as a result, the training time can then be efficiently cut down to less than half of that required in the standard LM algorithm. Such a highly efficient Modified LM training algorithm greatly increases the feasibility of applying the ANN method into the parameter sensitivity analysis, and also the following parameter uncertainty analysis. 4. In parameter uncertainty analysis of stochastic structure system, the parameters with a random field distribution are concerned. The ANN is embedded into the Monte Carlo (MC) digital simulation as a surrogate of the deterministic finite element solver. Then a so-called MC-ANN method is proposed for uncertainty analysis of stochastic structures. By use of the proposed MC-ANN method, the statistical features of random system outputs corresponding to the different random parameters can be derived much quickly and much efficiently. And the second-order importance indices of the parameters can then be obtained. As compared to the direct MC method, the proposed MC-ANN method shows a much higher computational efficiency up to several times. While compared to the analytical Stochastic Finite Element Method (SFEM), such as the first-order SFEM which has a high computational efficiency, the proposed MC-ANN method shows a much higher computational accuracy. Based on the combination of quick mapping from the ANN and a convincing solution from the MC method, the proposed MC-ANN method provides a promising and practical technique for accurately analyzing a stochastic structure. 5. In the parameter uncertainty analysis of a stochastic structure system, the discretizaiton of parameter random field is known to be one important premise of a stochastic structure analysis. Concerning the common-used local average discretization, a matrix type of Gaussian-integration method for applying the local average discretization is presented. With its clear concept and uniform operation, this new dicretization method shows an advantage of programming convenience.6. In the parameter uncertainty analysis of stochastic structure system, generally, Gaussian pseudo-random sequences generated from pseudo-random generators are directly applied into the MC sampling for representation of the discretized parameter random field. However, it is observed that errors inherent in the pseudo-random sequences to mean values, deviations and covariances may result in non-negligible errors in the representation. Concerning this issue, a calibrated MC sampling method is presented to shift, rescale and orthogonalize the Gaussian pseudo-random sequences so that the first-and second-order statistical features of the discretized parameter random field can be precisely represented with the calibrated MC samples. 7. In the parameter estimation analysis of structure system, concerning the common case of time-variant stochastic structure system, such as segmental construction of bridge structure, the author proposes a new sequential linear/nonlinear estimation formula by use of Markov’s process assumption and maximum a posteriori criteria, based on the consideration of construction consistency and the treatment of system rebuild. In two special correlation cases, the proposed estimation method is equal to the common-used least square estimation method and the Kalman filter estimation method, respectively. As a result, the proposed method can then be applied in a more general way. The parameter-updating and parameter-predicting abilities of the proposed estimation method in time-variant stochastic structure system can then be used efficiently to conduct time-variant structure system control problems, such as bridge segmental construction control process. 8. As an engineering application of the proposed theoretical studies, finally, the self-tuning adaptive control idea is applied into the construction control practice of Yamen Bridge project. The bridge is a prestressed concrete cable-stayed bridge with two pylons, single cable-plane and a main span of 338m. Actually, this bridge is the longest cable-stayed bridge of the same type in Asia. During the construction process, a girder segment is cast in-situ on movable carriages. Before the application of the self-tuning adaptive idea, it should be noted that deterministic simulation analysis of cable-stayed bridge construction process needs to be set up in advance, regarding that the deterministic simulation analysis is the inner mechanism of the construction process system and the basis of all the above uncertain parameter analysis. Concerning the prestressed concrete cable-stayed bridge which is common-used in China, in the present study, a new method of using element’s equivalent load increments is proposed to solve the important problem in construction simulation —estimating the creep and shrinkage effects of concrete. As compared to the routine更多还原