《基于改进型神经网络pid控制算法的烟气脱硫控制》由会员分享，可在线阅读，更多相关《基于改进型神经网络pid控制算法的烟气脱硫控制（90页珍藏版）》请在金锄头文库上搜索。
1、太原理工大学 硕士学位论文 基于改进型神经网络PID控制算法的烟气脱硫控制 姓名：李伟发 申请学位级别：硕士 专业：控制理论与控制工程 指导教师：田建艳 20090401 太原理工大学硕士研究生学位论文 I 基于改进型神经网络 PID 控制算法的烟气脱硫控制 摘 要 PID 类型的控制技术在现代过程控制中仍然占有主导地位， 但是经典的 PID 控制是基于准确模型的， 且要求系统特性变化与控制量之间是线性映射 关系，对于参数变化、干扰众多的控制系统，难以获得满意的控制效果。 神经网络 PID 控制是将神经网络应用于 PID 控制并与传统 PID 控制相结合 而产生的一种改进型控制方法，是对传统的
2、 PID 控制的一种改进和优化。 但是基于传统算法的神经网络 PID 控制自身存在训练时间长、收敛速度慢 以及局部极小等问题而限制了其在现代工业过程中的应用。因此，本文采 用 LM 算法的神经网络，建立具有预测输出的辨识器，构建了具有预测模 型的新型神经网络 PID 控制，此控制具有很强的逼近能力、收敛速度快且 能够有效地避免局部极小值。 在石灰石-石膏湿法烟气脱硫工艺中， 吸收塔浆液PH值的测量和控制是 影响脱硫率和终产物石膏品质的关键因素，脱硫反应方向很大程度上取决 于吸收液的PH值。PH值越大，SO2的溶解度越大，越有利于SO2的吸收，然 而，PH值的增大却不利于石灰石的溶解，且容易形成
3、塔内结垢造成堵塞。 因此对PH值的测量和控制具有十分重要的意义。 本文针对PH值变化过程的高度非线性、时滞性以及各种不确定性，常 规PID无法达到满意的控制效果， 实际脱硫工程中经常采用手动控制的方式 对其进行控制的现状， 构建了具有预测模型的改进型神经网络PID控制对其 太原理工大学硕士研究生学位论文 II 进行控制。 具体的研究内容包括以下几个方面： （1）深入某电厂生产现场，熟悉掌握了石灰石-石膏湿法烟气脱硫的工 艺，并了解了当前生产现场控制中存在的主要问题。 （2）根据某电厂的运行参数，并根据化学剂量平衡关系，在假设吸收 塔内浆液具有均一的浓度和温度分布的前提下，建立了被控对象（PH
4、值） 与控制量（石灰石浆液供浆量）之间的关系模型。 （3） 针对目前普通 PID 控制算法存在的问题， 采用了一种改进型的 PID 控制算法。并通过 Matlab 仿真，对所改进的 PID 算法进行了仿真试验，获 得了较好的控制效果，但改进型 PID 控制算法还是存在超调量大、稳定时 间长、振荡剧烈、适应性不强等缺点。 （4）针对改进型的 PID 控制算法所存在的这些缺点，构建了一种带预 测模型的改进型神经网络 PID 控制算法。首先采用传统 BP 算法对其进行 学习， 仿真结果表明传统 BP 算法不适合作为此控制系统的学习算法。 其次 利用 LM 算法对所建立的控制系统进行学习，通过 Mat
5、lab 仿真，将控制结 果与改进型 PID 的控制结果相比较，结果表明所建立的基于 LM 算法的带 预测模型的改进型神经网络 PID 控制算法在自适应性、抗干扰能力和控制 品质等方面较改进型 PID 控制算法均有显著的提高。 关键词：PID，BP 神经网络，LM 算法，湿法烟气脱硫，PH 值 太原理工大学硕士研究生学位论文 III THE FLUE GAS DESULFURIZATION CONTROL BASED ON CONTROL ALGORITHM OF IMPROVED NEURAL NETWORK PID ABSTRACT The PID control technology is
6、 one of the most commonly used technology of the modern process control, Because the traditional PID control is based on accurate model, and need the change of system characteristic is linear mapping relationship with control variable, so it is difficult to obtain satisfied control effect to the con
7、trol system having changed parameters and numerous interference. The neural network PID control is a improved control method that applied neural network to PID control and combined with the traditional PID control, it is the improvement and optimization of traditional PID control. Because the traini
8、ng time of network was long, convergence speed was slow and local minimum value couldnt be avoided to the neural network PID based on traditional algorithm, therefore, it was used that neural network based on LM algorithm, and then the identification implement has been constructed with the character
9、istic of predictive output, at last, a new kind of neural network PID control was formed. It had better approximation ability, and fast convergence speed and could avoid the local minimum value efficiently. 太原理工大学硕士研究生学位论文 IV The measurement and control of absorber slurry PH value is the key factor
10、effecting desulphurization rate and gypsum quality in the limestone-gypsum wet flue gas desulphurization, the direction of desulphurization reaction depends on absorption liquid PH value to a large extend. The larger the PH value, the larger SO2 solubility, the more beneficial to SO2 absorption. But
11、 large PH value is not beneficial to limestone dissolution, and is easy to form the scaling of absorber and lead to blocking. So, the measurement and control of absorber slurry PH value have very important significance. To the thing that the change of absorber slurry PH value is a process having hig
12、hly nonlinear, time-delay and various uncertainty variation, the difficult to meet the requirement of control by conventional PID controller, and the method of control we are using is the hand brake lever bracket control today. A new kind of neural network PID with the characteristic of predictive m
13、odel is applied to it. The main research contents of the thesis are as follows: (1) Going deep into production field of one power plant, studing and mastering the process of limestone-gypsum wet flue gas desulphurization, and understanding the main problems of current control.in production field. (2
14、) Assumed that the concentration and temperature of absorber slurry is equal, the realationship between PH value and flow volume of limestone slurry was established, according to operation parameters of one power plant, and balance relationship of chemical dose. 太原理工大学硕士研究生学位论文 V (3) To the problems
15、 of ordinary PID control at present, a improved PID control algorithm is used. By Matlab simulation, it was tested, and obtains favorable control effects. But the improved PID control algorithm had many disadvantages, firstly, its overshoot is very big, secondly, its stability time is very long, and
16、 third, its adaptability is not strong and so on. (4) To the disadvantages of improved PID control algorithm, a new kind of neural network PID control algorithm with the characteristic of predictive model is constructed. Firstly, traditional BP algorithm was applied to study it, the results showed that traditional BP algorithm is not suitable for its learning algorithm. Secondly, LM algorithm was applied t
