【作者】 王科平；

【导师】 慈春令；

【作者基本信息】 燕山大学 ， 控制理论与控制工程， 2003， 硕士

【摘要】 广义预测控制(Generalized Predictive Control)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。本文在参考大量国内外文献的基础上，简要的论述了预测控制的基本算法、性质和参数选择。然后基于神经网络，从几个不同的方面对智能广义预测控制进行了研究。首先，在第2章介绍了广义预测控制的基本算法，其中所提的各种算法，均需在线求解丢番图方程和大矩阵求逆，运算量大，不利于实时控制。为此，本文在第3章给出了两种改进方案以降低计算量，解决快速性问题。由于神经网络能够实现任意的非线性映射，具有高度的并行结构、易于实现并行处理、学习和适应能力强等特点，所以可以很好的解决预测控制难以建模和实时性问题，所以本文将广义预测控制和神经网络控制技术相结合进行研究。但是对某些非线性比较强的系统，用神经网络建模，有时会存在较大的训练误差，而且这种误差难以通过对网络的进一步训练而缩小，这样建立的预测模型就会存在较大的建模误差影响控制。为此本文提出了一种改进的算法来降低辨识器与控制器对未建模动态的敏感性，并通过仿真来验证该方法的有效性。本文从控制理论、设计方法到将设计方法应用到四机架冷连轧机张力控制系统的全过程进行研究。针对冷连轧机为一快速、实时性强的系统，提出了基于Elman神经网络误差修正的PI型广义预测控制算法。该算法不仅在鲁棒性方面有了很大的改善，而且能大大减少计算时间，以满足工业控制的要求。最后分别把前面论述的几种方法应用到冷连轧张力控制中，仿真结果表明本文所提方法具有较强的理论价值和应用价值。更多还原

【Abstract】 Generalized Predictive Control(GPC) is , which appeared in 80’s, a new type of computer control method and one of the most representative algorithm. It has received increasing attention in the field of control and industry. Based on lots of literature, this paper introduces fundamental principles of predictive control and the newly developments of GPC.This paper studies the basic algorithm and characteristic of GPC. All the algorithms of GPC in Chapter II have to solve Diophantine equation and the inversion of big matrix online. So real time control is infeasible. In order to solve this problem, two kinds of resolutions are put forward in Chapter III.Based on the traditional predictive control, it has provided new predictive and intelligence control methods mixing predictive and neural network. The research results show that by adopting parallel proceeding structure on a large scale, neural network has high convergence coefficients irrelevant to the number of nodes. For this reason, neural network can resolves the GPC’s problems of modeling and rapidness. But in some nonlinear system there is a error which can’t decrease with the neural network modeling. So a robust GPC model is presented, which can cut down the sensitivity of the identifier and controller, increase the adaptive ability and robustness of the controller.The paper makes the full research from control theory, design method and the application of the theory to four fight cold-rolling mill’s tension control. A PIGPC model based on Elman network is presented in this paper. This algorithm not only can improve the characteristic of robust but decrease the control time, and can max the application of industrial.Finally the application of GPC to the tension control of cold trandom machine is studied. Different design solutions of GPC mentioned above are used to realize expectation.更多还原