【作者】 张坚；

【导师】 周洪煜；

【作者基本信息】 重庆大学 ， 热能工程， 2003， 硕士

【摘要】 现代电站的生产过程中存在过热汽温、再热汽温、锅炉水处理、负荷调节等许多大迟延、大滞后、特性时变的对象，它们中有些还是具有强非线性特性的对象或多变量耦合系统，采用常规的PID控制手段很难取得良好的控制效果。若应用现代控制理论中的自适应控制、最优控制、解耦控制、预测控制等控制手段，则需要建立被控对象的数学模型，而且往往控制系统的计算量大、实时性差。这些缺点使其很难满足实际生产过程的需要从而极大地限制了其在现代电站中的应用。随着发电机组向大容量、高参数方向发展及各种新型发电方式的出现，电站中各生产环节的特性越来越复杂而对其控制品质的要求却越来越高，急需新的控制技术来对其进行有效的控制。神经网络具有表示任意非线性关系和学习等能力,通过恰当选择网络层次和隐层单元数能够以任意精度逼近任意连续函数及其各阶导数。为时变、非线性对象的动态特性的辩识提供了简单而有效的一般性的方法，解决了时变、非线性对象控制中的瓶颈问题。因此基于神经网络的各种先进控制技术是解决现代电站中控制难题的一条有效途径。本文紧密结合我国电厂的实际情况，以解决电厂实际运行中存在的控制问题为出发点，抓住火电厂热工控制系统普遍存在的大滞后和特性时变的特点，在已有的神经网络控制算法的基础上，针对这些算法中存在的不足进行进一步研究并提出了基于Elman网络的隐式广义预测控制、基于改进的Elman网络的自适应预测函数控制、基于混合神经网络的非线性自适应预测函数控制、基于神经网络的多变量解耦自适应预测函数控制及基于模糊神经网络的模型参考自适应预测控制等多种改进方案。这些改进方案既具有自适应控制和预测控制等现代控制理论的优点，又无需被控对象的数学模型，而且控制效果良好、适应性广，满足了非线性对象及多变量耦合系统等特性不同的生产环节对控制系统的要求。使用MATLAB语言对采用上述控制方式的过热汽温、再热汽温、给水系统、锅炉水处理、负荷调节等不同对象的控制系统进行仿真实验，结果表明了这些控制技术的有效性。更多还原

【Abstract】 There are many plants that have the character of time-varying,large delay,large inertia in the process of production of modern power station such as the superheated steam temperature,the reheated steam temperature,the water treatment of boiler and the load regulation.Some of them have the strong character of nonlinear and some of them are multivariable coupling systems It is very difficult to obtain good effect of control by conventional PID control.It is necessary to set up the mathematics models of controlled plants if applying control means of modern control theory such as self-adaptive control,optimal control，decoupling control and predictive control.And these control systems have large calculation and bad character of real time in general.These disadvantages make them be not able to meet the need of real production process and limit the application of them in modern power station.Now the generator units are developing towards large capacity and high parameters.Many new kinds of generating electricity means are coming into being.The characers of production units of power station become more and more complex but the demand to quality of control becomes more and more strict.The new control means are in bad need to control them effectively.Neural network has the ability of learning and expressing any nonlinear relation.It can approximate any continuous function and their any order derivatives with any precision if it has the correct layers of networks and the correct number of hidden units.So the identification of dynamic character of time-varying,nonlinear plants has a kind of simple and effective means.The critical problem of control of time-varying,nonlinear plants is resolved.So all kinds of advanced control means based on neural network is an effective way that resolve the control problem of modern power station.This dissertation is dedicated to solve the practical control problems in power station under the existence of thermal control system characterized by large delay and time varying.Aiming at the present disadvantages of the neural network control algorithm,some improved means are researched and put forward.They are implicit generalized predictive control based on Elman network,self-adaptive predictive function control based on improved Elman network,nonlinear self-adaptive predictive function<WP=6>control based no hybrid neural network,multivariable decoupling self-adaptive predictive function control based on neural network and model reference self-adaptive predictive control based on fuzzy neural network.These means do not need the mathematic model of controlled plants but own the merits of self-adaptive control and predictive control of modern control theory .And the control effects are good and the adaptive character of them are extensive.They can meet the demand of production units with different characters of nonlinear plants and multivariable coupling systems.The simulation experiments of different plants such as superheated steam temperature, reheated steam temperature,water-supplying system,water treatment of boiler,load regulation are done in MATLAB.The results show that these control means are effective.更多还原