【作者】 李金霞；

【导师】 邱公伟；

【作者基本信息】 福州大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 工业生产过程中，时滞过程是比较难控的对象，对时滞工业过程控制方法和机理的研究一直是国内外过程控制界的热门课题。各国学者除了采用一些常规控制方案，如改进型PID控制，采样PI控制，Smith预估控制，大林控制算法等进行时滞过程的控制之外，都在寻找更有效、更实用的控制方法。本文在对用传统方法控制时滞系统进行综述的基础上，指出近年来发展起来的预测控制在时滞系统的控制上有良好的前景。预测控制分为IMC，MPC，DMC，GPC四类，其中DMC具有建模方便，算法简单，对模型要求低的特点，使其特别适合于工业现场控制中推广应用。时滞系统往往带有一定的非线性，这对预测控制的应用造成一定的困难，本文指出将神经网络引入预测控制，构成智能预测控制系统对改善控制性能有明显的效果。尤其是单神经元控制，结构简单，计算方便，实时性强，有较好的鲁棒性与自适应性，把它引入到预测控制中可以得到比较好的实现。近年来发展起来的预测控制对时滞系统有较好的控制效果，它提供了较好的跟踪性能，但是在抗扰性能上不能令人满意，而且对于具有非线性的对象适应性较差。把DMC控制与PID控制结合起来构成复合结构，来解决抗扰性能与跟踪性能双优的问题，又把神经网络控制引入进来以适应非线性对象，就有可能构造出先进的预测控制系统。本文在DMC－PID串级控制的基础上，提出了一种改进方法，即引入神经元PID控制。该方法解决了PID参数的整定问题，提高了DMC的抗干扰性能，达到跟踪性能与抗扰性能双优，并提高了预测控制对非线性对象的控制能力。最后通过对控制系统的设计与仿真试验，与单一预测控制进行比较，证实了该方法的可行性与有效性。更多还原

【Abstract】 In industrial process ,it is more difficult to control the time-lag system. The research of control method and mechanism of time-lag system is always a popular topic in domestic and foreign process control circles .The scholars from all corners of the world not only employ themselves on some traditional programs such as modified PID control ,sampling PI control Smith Predictor ,Dahlin algorithm etc , but also look for more efficient and more practical control methods .Based on the summarization of traditional control methods in dealing with time-lag system , in this dissertation ,we bring out that predictive control has good prospect for dealing with time-lag system .Predictive Control can be divided into four types :IMC (Internal Model Control ), MPC (Modem Predictive Control ), DMC (Dynamic Matrix Control ) and GPC (Generalized Predictive Control ). Among them , DMC has advantages of convenient modeling, simple computing ,and better robust . Therefore DMC applies to industrial field control specially. Time-lag systems always have some non-linearity ,which makes the application of Predictive control some difficult . In the dissertation ,the author points out : introducing Neural Network into Predictive Control to compose intelligent control system can improve the control capacity notably .Especially single neuron ,has the advantages of simple structure convenient computing ,better real-time ability and stronger robust and adaptability .So it is realized easily to introduce single neuron into predictive control . Predictive Control ,which has developed in recent years ,can control time-lag system efficiently .Its trackability is better ,but its immunity is not satisfied. Moreover , its adaptability to non-linear system is not good .The cascade control system consisted of DMC and PID is provided with combinated structure. Dual optimal problem of trackability<WP=4>and immunity could be solved. In addition, the suitability to non-linear system could be improved via introducing NN into cascade system. An advanced predictive control system might be compose by above methods .Based on the DMC-PID cascade system, an improvement method is researched in this dissertation, that is, neuronPID is introdued into cascade system . This method can solve the problems of PID parameters’ tuning , and improve the immunity . Dual optimal problem of trackability and immunity could be realized .Simulation experiments compared with single predictive control approve this method’s feasibility and validity.The ability of dealing with the non-linear system is raised.更多还原