【作者】 王哲；

【导师】 温淑焕； 刘福才；

【作者基本信息】 燕山大学 ， 模式识别与智能系统， 2009， 硕士

【摘要】 广义预测控制是80年代产生的一种新型计算机控制方法,是预测控制中最具有代表性的算法之一。它一出现就受到了国内外控制理论界和工业界的重视,成为研究领域最为活跃的一种预测控制算法。但由于广义预测控制中控制律的计算需要求解Diophantine方程和逆矩阵,计算量大,因此,很不适合工业过程中一些要求快速性高、实时性强的系统。针对这一问题,本文在参考国内外相关文献的基础上,对广义预测控制快速算法进行了研究,主要研究内容如下:首先,在广义预测控制基本算法的基础上,从广义预测控制与神经网络、与模糊控制等的结合方面,综述了广义预测控制的各种快速算法,并对广义预测控制的特点和参数选择进行了介绍。其次,针对广义预测控制算法中计算量大的问题,提出了一种改进的广义预测控制快速算法,在基本的广义预测控制算法中引入约束矩阵,从而将广义预测控制中的矩阵求逆部分变为标量形式进行求解,大大减小了计算量,并在此算法中引入Elman神经网络,可以使广义预测控制算法能够很好的应用在非线性系统控制中。最后,基于混沌系统非线性强、多变量耦合等特点,提出了利用带时变遗忘因子的递推最小二乘进行参数辨识的方法,得出混沌系统的模型;同时在广义预测控制算法中引入柔化输入信号,从而减小了计算量,实现了快速性,使得广义预测控制算法在混沌系统中有很好的应用效果。更多还原

【Abstract】 Generalized Predictive Control (GPC), a new type of computer control method birthing in 1980’s, is one of the most representative algorithms. It is researched actively and has received increasing attention in the field of control and industry. This paper studies the fundamental principles of predictive control and the new development of GPC. The calculation of control law in generalized predictive control needs to solve Diophantine equations and inverse matrix. Because of the mass calculation the applications in high rapid and strong timely system are limited. According to these problems and on the basis of the massive domestics and foreign literatures, several fast generalized predictive control algorithms are proposed and some simulation studies are carried out for the typical industrial models. We study the algorithm of fast generalized predictive control. The main research results have been shown below:Firstly, combining with neural network and fuzzy control, we summarize all kinds of fast algorithms of GPC on the basis of the basic algorithm, and we also introduce GPC’s characteristic and how to choose parameter.Secondly, for the mass calculation of generalized predictive control algorithm, an improved generalized predictive control algorithm is proposed. It introduces constraint matrix into basic generalized predictive control algorithm which change the matrix inversion part into scalar form in order to get the solution, thus reduce the mass computation greatly. We introduce the Elman neural network into this algorithm as well as in order to use this algorithm for the nonlinear system.Finally, Based on strong nonlinear and multi-variable coupling features of chaos system, we use recursive least squares with forgotten factor to identify parameters and get chaos system model. Adding the soft coefficient to input signal, we can reduce calculation, achieve faster system response and get GPC algorithm used in chaos system with better application results.更多还原