蒸汽温度系统神经元网络控制方法的研究

【作者】 杨晓飞；

【导师】 陈铁军；

【作者基本信息】 郑州大学 ， 控制理论与控制工程， 2002， 硕士

【摘要】 在火电厂热工生产过程中，整个汽水通道中温度最高的是过热蒸汽温度，蒸汽温度过高或过低，都将给安全生产带来不利影响，因此，必须严格控制过热器的出口蒸汽温度，使它不超出规定的范围。汽温调节对象是一个多容环节，它的纯迟延时间和时间常数都比较大，干扰因素多，对象模型不确定，在热工自动调节系统中属于可控性最差的一个调节系统。目前热工过程控制中，传统的控制方法如PID等得到了广泛的应用，但这种方法在系统负荷稳定时能够取得很好的控制效果，而在系统负荷有较大波动时，则难以稳定及时的对系统进行控制。随着智能控制理论的深入研究，智能控制为火电厂热工过程自动控制提供了新的方法。 首先，本文在深入了解火电厂热工过程生产工艺的基础上，仔细分析了影响蒸汽温度变化的各种扰动因素。蒸汽温度对象的动态过程和对象模型十分复杂，扰动因素诸多，比如蒸汽流量的变化，燃烧工况的变化，锅炉给水温度的变化、进入过热器蒸汽热焓的变化等，而且这些因素还可能互相制约。在研究影响蒸汽温度变化的三个主要扰动因素即蒸汽流量、烟气流量及其温度和减温水流量变化时系统动态特性的基础上，采用机理分析方法，建立了蒸汽温度系统的仿真数学模型。 然后，根据神经元控制理论，在探讨了基本的神经元控制算法的基础上，分析了神经元增益对控制性能的影响，根据仿真实验及实践证明，对于开环放大倍数较大的被控对象，它能起到衰减神经元的控制作用，消除系统响应超调和振荡的不利影响；对于开环放大倍数较小的被控对象则能增强神经元的控制作用，加快系统的响应速度，这就要求神经元增益具有自适应能力，借助于PID控制的思想，将PID算法用于调整神经元控制器的增益，利用实际输出和期望输出之间的误差及误差变化等信息进行线性组合来决定神经元增益，提出了PID自适应调节增益神经元控制算法，同时，给出了以输出误差平方为性能指标的神经元控制算法。本文设计了蒸汽温度系统的控制方案，为克服诸多干扰因素，采取串级控制系统，其中主控制器采用PID自适应调节增益神经元控制器，副控制器采用常规PI控制器。结合现场实际情况，通过仿真试验，将PID自适应调节增益神经元控制系统和常规PID控制系统作了对比，分别在加入内环蒸汽流量干扰、外环烟气温度干扰、对象参数发生变化等情况下进行了比较，结果证明PID自适应调节增益神经元控制器有着更好的控制效果，在抗干扰和适应工况变化方面都要优于常规的PID控制器。 针对PID自适应调节增益神经元控制器在系统响应速度方面的不足，本文研究了PID自适应调节增益神经元控制器的两种改进算法，具有积分分离作用的神经元 ws 智能控制器和具有神经元增益为一逐步衰减函数的智能控制器，详细分析了衰减函数 的定义及其参数对控制性能的影响，给出了神经元增益为一逐步衰减函数的智能控制 器的算法，通过仿真试验，表明具有衰减函数增益的神经元智能控制器较常规卜＊ 串级控制系统和nD 自适应调节增益神经元控制系统在动态过程的平稳性和快速性 方面有较大改善，并且具有控制算法调整方便、设计简单等优点。然后，针对神经元犁套”“一’“’——””———””’——””’““——’”””’”——”““’—“”’———””’”’””—‘””’—“””’””—·－———一圈 基本控制算法，通过仿真实验，研究了神经元增益、学习速率、初始权值等控制参数7 对系统控制性能的影响，对控制器性能改善和初始参数的选择具有抬导意义。 最后，本文研究和设计了基于模式识别的仿人变周期智能控制器，并将其与常规 PID控制系统、具有衰减函数增益的神经元控制系统作了仿真比较，仿真结果表明， 基于模式识别的仿人变周期智能控制器在系统响应的快速性方面优于兵有衰减函数增 益的神经元控制器和常规PID控制器，但在系统动态过程的平稳性方面不如具有衰 减函数增益的神经元控制器。综上所述，本文针对火屯厂过热器蒸汽温度系统，着重 作了控制算法的研究，并对各种算法进行了仿真实验，仿真结果表明，具有衰减函数 增益的神经元控制系统有着较好的控制效果，并且它的算法设计简单，有较好的实川 价值。更多还原

【Abstract】 The super-heated steam temperature is the maximal temperature in the whole steam channels at the process of thermodynamic engineering in power plant. If the steam temperature is too high or too low, it will bring on dangerous factors. We must control the super-heated steam temperature of the output of the super-heated implement to an expected range. The vapor object is a multi-container tache. It has pure lag and many disturbances. Its constant time is much bigger and its object model is not confirmable. It is the most difficult control system in the process of thermodynamic engineering. The traditional control method such as PID controller is implemented widely in the process of thermodynamic engineering now. But they work well only when the systemic load is steadily and can not work well when the systemic load is changed with a wide margin. With studying of the intelligent control theory, it provides a new control method for the process of thermodynamic engineering in power plant.At first, the author analyzes many disturbance factors in vapor temperature on the basis of the produce technics in power plant. The vapor object has complicated dynamic process and intricate object model. It has many disturbances. Such as vapor flux, burning medium, water temperature, vapor temperature entrance into the super-heated implements. These factors may affect on each other. On the basis of analyzing the three main disturbance factors in vapor temperature and the main dynamical characteristic of the super-heated steam temperature control system in power plant, this paper builds the vapor temperature math model for simulation by use of mechanism analyzing.Based on the single neuron control theory, this paper analyzes the neuron plus influence on control performance on the foundation of the basic neuron control algorithm. Simulation results prove that for the control objects which have a more large opened-loop plus, it can weaken the control effect of the neural controller and abate the system responseoverall plus and oscillation, for the control objects which have a less smaller opened-loop plus, it can strengthen the control effect of the neural controller and quicken the system response. These request the neuron plus to have self-adjusting ability. If we can apply the regular PID algorithm and the error between practice output and expected output and its radio of its changing to decide the neural controller plus, we can design a self-adjusting plus single neural controller based on regular PID algorithm. At the same time, we also provide neuron control algorithm about the performance target based on the output error square function. The paper designs the control project about the super-heated steam system. To overcome the disturbance factors, we apply series control system to the steam system. The main controller adopts the self-adjusting plus single neural controller and the minor controller adopts the conventional PI controller. As compared with the neural controller and the conventional PID controller is simulated. The experiment shows that the effect of the neural controller is better and it is superior to the conventional controller on the condition of joining the inner loop vapor flux disturbance and outer loop smoke temperature disturbance and other conditions.Aimed at the problem the self-adjusting plus single neural controller based on regular PID algorithm has a slow response velocity, the author analyzed two improved algorithm of the neural controller. One is separate integral neural controller, another is attenuation function neural controller. The author analyzed some parameter influence on control performance and defined the attenuation function and its algorithm. The simulation comparisons between two neural intelligent controllers show that the attenuation function neural controller is superior to the self-adjusting plus single neural controller in response velocity, and its algorithm is simple and easy. Then, aimed at the basic neuron control algorithm, simulation experiments were made. The auth更多还原