【作者】 杨锡运；

【导师】 宋之平； 徐大平；

【作者基本信息】 华北电力大学（北京） ， 热能动力工程， 2003， 博士

【摘要】 DCS在我国火电机组中虽已广泛应用，但其控制策略均采用常规PID控制。因火电机组控制对象的复杂性，其设备和工作原理涉及多个领域，多个变量，动态特性具有非线性、大滞后和时变等特点，这种基于单回路固定模型的传统控制策略限制了控制品质的提高，因此开发出适合火电厂过程的先进控制策略具有重要意义。本文深入研究了多变量热工对象的先进控制策略问题，同时对键图模型先进控制进行了初步探索，主要内容可归纳为：针对非线性多变量对象，提出了基于模糊神经网络的非线性控制策略，使非线性引发的解耦模型无法实现和线性控制器品质差的问题，得到有效解决。文中首次提出了一种模糊神经网络学习算法：移动小论域法，设计了规则少的模糊神经网络非线性在线新控制器，并提出基于分散预测补偿和基于神经网络非线性补偿器的两种模糊神经网络多变量控制方案， 针对大时滞多变量对象，设计了基于内模解耦的先进控制方案。内模控制含有Smith预估和逆控制思想，再融合解耦技术，可获得大时滞多变量对象高的控制品质。文中给出基于对消法的无时滞和有时滞两种多变量对象的时域内模解耦方案；又基于预测内模设计法，首次提出基于单值MAC的内模解耦多变量控制算法，包括输入无约束和输入有约束两种情况。从提高实时性角度出发，对大时滞多变量对象，本文还提出了在线计算复杂性低的预测函数先进控制策略。文中给出一阶时滞对象预测函数与规范串联前馈解耦技术相结合的多变量透明控制方案，又首次提出预测函数的多变量直接控制算法，并研究其在迟延平衡和迟延不平衡两种类型多变量对象上的解耦特性。为拓宽热工过程先进控制策略的研究领域，本文还对目前国际上新兴的键图模型控制进行了初探。首次将键图模型引入国内热力系统动力学领域，并直接研究基于键图模型的新型控制方法。文中成功首创了火电厂自然循环锅炉蒸发系统键图模型，汽包水位的动态仿真曲线证实，键图模型精度高；提出的基于无因果划键图模型的定性和定量信息混合控制方法，具有鲁棒性和精确性兼顾的好控制品质。最后将上述先进控制策略进行了应用仿真。非线性多变量球磨机对象和具有大惯性、大时滞耦合的汽-汽交换器多变量再热汽温对象上仿真结果证实，本文的研究对丰富火电厂多变量对象的控制具有重要理论意义和实用价值；键图新控制算法的仿真则为热工过程先进控制和控制理论发展提供了一个全新思路。更多还原

【Abstract】 Although DCS is widely applied into power plant, its control strategy still depends on conventional PID theory. Due to the high complexity of fossil-fueled power plant, which involves equipments of multi-field and multivariable with the characteristics of nonlinear, large delay and time varying, the control performance is somewhat stunted by conventional control strategy based on single-loop, fixed parameter model. Therefore, it is important to develop advanced control strategies for thermal process control. In this dissertation advance control strategies of multivariable thermal process has been explored, and a control method based on bond graph model has also been discussed preliminarily. Facing nonlinear, multivariable process, a fuzzy neural network control strategy is presented. It not only realizes decoupling model with nonlinear, but also overcomes the low quality of linear controller. Both a learning algorithm of fuzzy neural network named small dynamic universe and a fuzzy neural network controller (FNNC) with simplified rules are firstly proposed. Then multivariable control scheme of FNNC with nonlinear decoupling based on distributed predictive compensation or neural network compensator is given. For the multivariable process with large delay, an internal model decoupling control structure is provided. An internal model control, which implies the principle of Smith estimator and inverse control, combines decoupling technology to achieve good performance for multivariable process with large delay. According to the cancellation method, IMC for multivariable process with or without large delay are designed individually. Besides, the MAC multivariable control algorithm based on prediction approach is proposed, which includes no input constraint condition and input constraint condition. In order to improve real-time performance of the algorithm, for large delay multivariable process a predictive functional control (PFC) with low complexity of calculation is proposed. A transparency control that incorporates P criterion decoupling into predictive functional control for first order plant of large delay is developed. Furthermore, a multivariable PFC is firstly presented and control performance is tested and studied on delay-balanced system and delay-unbalanced system. For enlarging the scope to advanced control of thermal process, on the other hand, the dissertation also preliminarily research on bond model control, which is a new control field in the world. A bond graph model is firstly introduced into thermaldynamics <WP=6>in our country and a control algorithm based on bond graph model is developed. Bond graph model of vaporization system of natural circulation boiler is successfully built and a simulation curve of the drum level demonstrates the high precision of the model. Further, a hybrid qualitative and quantitative control method based on bond graph of no causality stroke is put forward and can achieve a control performance of both robustness and precision. Simulation results of above advanced control strategy are also carried out. Tests on ball mill system with nonlinear coupling and reheat steam temperature with large time constant, large delay using steam-steam exchanger show that multivariable control strategies developed have important theoretical and practical advantage for process control of power plant. The bond graph control opens a new window to research on control theory and thermal process control.更多还原