【作者】 任燕燕；

【导师】 刘长良； 王东风；

【作者基本信息】 华北电力大学 ， 控制理论与控制工程， 2014， 博士

【摘要】 几乎所有的生产系统都是非线性系统,人们常说的线性系统是对系统非线性特性在某种程度上进行忽略或者某种假设条件下近似得到的,这种近似必然会产生误差,影响生产系统的控制效果。生产系统结构越来越复杂,包含的非线性特性也更多样化,简单的线性近似已经不能满足提高系统生产力的要求,所以,非线性系统辨识是大势所趋。到目前为止,没有一种通用的方法可以对不同结构的非线性系统进行辨识,一般是具有不同非线性特性的系统,用不同的辨识方法。模块化的非线性模型结构简单,内部连接方式明了,比较适合生产系统的辨识,是近年来颇受研究者青睐的一种非线性系统辨识常用的模型。模块化模型主要有Hammerstein模型(H模型)、Wiener模型(W模型)及后来出现的Hammerstein-Wiener模型(H-W模型)和Wiener-Hammerstein模型(W-H模型)。热工系统是规模庞大、结构复杂、控制要求较高的生产系统,生产过程中存在着不同程度的非线性,本文从自动控制系统的构成出发,分析了热工控制系统中执行器和检测变送器的非线性特性,进而得知,模块化的非线性模型适合于热工系统典型过程的辨识。本文重点研究了三种模块化非线性模型：Hammerstein模型、Wiener模型以及Hammerstein-Wiener模型。采用粒子群算法及其改进算法优化模型参数,用神经网络理论构造新的模块化模型及推导模型自身学习规则。借助分散控制系统存储的热工过程输入输出数据,将模块化模型的辨识方法应用于热工系统的辨识中。本文主要内容包括：1.提出了一种基于聚类分析的样条函数多项式Hammerstein模型,用粒子群算法寻优模型参数。将该辨识算法应用于热工系统某生产过程的辨识中,仿真结果表明了该样条函数Hammerstein模型辨识算法的有效性,为热工系统辨识提供了一种有效途径。2.提出了两种网络化Wiener模型,两种模型分别用BP网络和RBF网络表示模型的非线性部分,将模型转换成串联的网络结构；两种模型都采用双层优化策略,用BP算法和粒子群算法分内外两层优化模型参数。将这两种方法应用于热工系统两个对象的辨识,CO2浓度系统的辨识结果表明网络化W模型较样条函数H模型效果好,主汽压系统的辨识结果表明网络化W模型有较好的适用性。3.引入量子计算理论,用量子粒子群算法辨识一般指数多项式Hammerstein模型,并将该算法应用于热工系统的辨识。一般指数多项式模型结构简单,计算速度较快；量子粒子群算法较普通粒子群算法,增加了种群的多样性,一定程度上避免了早熟。从循环流化床机组三个对象的辨识结果可以看出,简单多项式H模型可以用于一部分实际系统的辨识,量子粒子群算法一定程度上可以提高指数多项式H模型的辨识精度。4.提出了一种网络化Hammerstein-Wiener模型,研究了一般多项式H-W模型和文中提出的网络化H-W模型的辨识方法,用量子粒子群算法辨识模型参数。分别将两种模型应用于热工系统两个典型环节的辨识,仿真结果表明了H-W模型能较好地表达生产系统的特性。本文主要研究基于智能计算的非线性模型辨识算法及其在热工系统中的应用,希望本文的研究工作能对热工系统的辨识有一定的理论与实践价值,并能对其它生产系统的辨识起到一定的启发作用。更多还原

【Abstract】 Almost all production systems are nonlinear systems. The so-called linear system is obtained approximately under some assumptions or by ignoring to some nonlinear characteristics of the system, while this kind of approximation will inevitably produce errors, which will influence the control effect of production systems. As the structure of production system becomes more and more complex, nonlinear characteristics contained in those systems are also more diverse, as a result, simple linear approximation can no longer meet the requirement of improving system productivity, and therefore, nonlinear system identification has become the general trend. So far, there is no universal method to identify nonlinear systems with different structures, usually, different methods to differernt systems with different nonlinear characteristics.The modular nonlinear model has not simple structure only, but also clear internal connections, so it is popular and widely used by researchers in recent years. Modular models mainly include Hammerstein model (H model), Wiener model (W model), and the succeeding Hammerstein-Wiener model (H-W model) and Wiener-Hammerstein model (W-H model).Thermal system is large-scale production system with complex structure and high demand for control performance, and there are different degrees of nonlinearity in the production process. In this thesis, analyzes nonlinear characteristics of actuators and detection transmitters in thermal control systems, starting from the composition of the automatic control system, and then it is learned that modular nonlinear models are suitable for the identification of typical processes of thermal systems.Three modular nonlinear models are mainly studied in this thesis:Hammerstein model, Wiener model, and Hammerstein-Wiener model. Model parameters are optimized using particle swarm optimization algorithm and its improved algorithm. With neural network theory, new modular models are constructed and the learning rules of the models itself are deduced. Identification methods of modular models are applied to the identification of thermal systems, with the input and output data of thermal processes stored in distributed control systems. The main contents include:1. Spline function polynomial Hammerstein model based on cluster analysis is introduced and particle swarm optimization algorithm is used to optimize model parameters. This identification algorithm is applied to identify a typical link of thermal system, and the simulation results show the effectiveness of this spline function Hammerstein model, which provided an effective way for identification of production systems.2. Two networked Wiener models are introduced, the nonlinear parts of which are represented by BP network and RBF network respectively, and then the models can be converted into series structures. Both of these two kinds of models adopt double-optimization strategy, which optimizes networked models in inner and outer layers with BP algorithm and particle swarm optimization algorithm respectively. These two methods are applied to identification of two objects of thermal systems. The identification results of CO2concentration system show that networked W model outperforms spline function H model, and the identification results of main-steam pressure system show that networked W model has better applicability.3. Identify general index polynomial Hammerstein model with quantum particle swarm optimization (QPSO) algorithm, by adopting quantum computing theory, and apply this algorithm to the identification of thermal systems. Generally index polynomial H model has simple structure and faster calculation speed. Precocity is avoided to some extent, for the diversity of population is increased in QPSO algorithm, compared with general particle swarm optimization (PSO) algorithm. From the identification results of three objects of circulating fluidized bed unit, it can be seen that simple polynomial H model can be used for part of the actual system identification and quantum particle swarm algorithm can improve the identification accuracy of model to some extent.4. One kind of networked Hammerstein-Wiener model is introduced, and two identification methods of general polynomial H-W model and the networked model proposed in this paper are studied, adopting QPSO algorithm to identify the parameters of the models. The two methods are applied to the identification of two typical production links in thermal systems, and simulation results show that H-W models are more capable to express characteristics of production systems.This thesis studies mainly the identification algorithms of nonlinear models and its application in thermal systems based on intelligent computing. It’s wished that the research work in this thesis has some theoretical and practical value for the identification of thermal systems and some enlightening action for the identification of other production systems.更多还原