电气与信息学院自动化专业外文翻译(毕业设计用)-锅炉蒸汽温度模糊神经网络的广义预测控制.doc

毕业设计（论文）外文翻译 neuro-fuzzy generalized predictive control of boiler steam temperature锅炉蒸汽温度模糊神经网络的广义预测控制 本 科 电气与信息学院 自 动 化 讲 师 ： 2008年4月20日 学生姓名学历层次所在院系所学专业指导教师教师职称完成时间（本文档前部分为中文，后部分为英文部分，后部分英文部分为pdf转化为word版本，不清晰之处，可参考本人上传的英文pdf版本原文，可以免费下载英文pdf版本(下载地址：/p-34794851.html)，以人格担保。可先下载英文pdf版本后，再下载中文部分即本文档）锅炉蒸汽温度模糊神经网络的广义预测控制 摘要：发电厂是非线性和不确定的复杂系统。现代电厂在运作上的，为确保高效率和高负荷的能力，可靠的控制过热蒸汽温度是必要的。本文提出了一类在非线性广义预测控制器的基础上的模糊神经网络（ nfgpc ）。所提出的非线性控制器适用于控制一台200 mw电厂的过热蒸汽温度。从实验的移植和仿真移植上获得比传统的控制器好得多的性能。 关键词：模糊神经网络;广义预测控制;过热蒸汽温度。1 引言 这种持续不断的电厂和电力站复杂系统的特点是非线性、不确定性和负载扰动。蒸汽发电的过程中锅炉-汽轮机温度过热是一个重要的过程，蒸汽加热后，进入涡轮驱动发电机。控制过热蒸汽温度不仅是在技术上具有挑战性，但在经济上也是十分重要的。 图 1锅炉过热器和蒸汽生成过程。 从图1看出 ，产生的蒸汽从锅炉汽包通过低温过热器后进入辐射型屏。水变成喷涂的蒸汽，以控制过热蒸汽的温度。适当的控制电厂过热蒸汽温度是极其重要的，可以确保整体效率和安全性。蒸汽温度太高是不可取的地方，因为过热它可损害和高压力汽轮机，太低也不行，因为它会降低电厂效率的。减少温度波动内过热也是重要的，因为它有助于减少在单位内机械应力造成的微裂纹，延长单位秩序寿命，并减少维修成本。作为gpc的推导应该尽量减少这些波动，它是众多的控制器是最适合实现这一目标的。 多变量多步自适应调节已适用于控制过热蒸汽温度在150吨/ h锅炉 ，提出了广义预测控制以控制蒸汽温度。非线性长程预测控制器基于神经网络发展是以控制主蒸汽温度和压力。控制主蒸汽压力和温度的基础上，非线性模型的构成是非线性静力常数和非线性动力学。 模糊逻辑是把人类的经验透过模糊规则表现出来。然而，设计模糊逻辑控制器是有点时间消费，由于模糊规则，往往得到的试验是错误的。在此相反，神经网络不仅有能力近似的非线性职能与任意精度，他们也可以有受过训练的实验数据。该模糊神经网络（ nfns ）最近开发的优势，模型的透明度，模糊逻辑，和学习能力的神经网络。该nfns已被用来发展自我控制，因此，一个有用的工具，可以发展国家非线性预测控制。从nfns可以考虑到作为一个网络构成的几个地方， 其中每一项包含一个局部线性模型，在非线性预测控制的基础上nfns可以制订与网络，使用各自的地方线性模型把当地所有的广义预测控制器（gpc ）的设计，。按照这一办法，在非线性广义预测控制器的基础上， nfns ，或简单地说， 该模糊神经网络的广义预测控制器（ nfgpcs） 推导出在这里。建立控制器，然后应用控制过热蒸汽温度的200 mw机组。实验所得的数据，用来训练nfn模型植物，并从哪个地方gpcs组成部分的nfgpc ，然后设计。 拟议的控制器的测试首先就模拟这个过程中，在申请前，它控制火电厂。2模糊神经网络模型 考虑以下的一般单输入单输出非线性动态系统： （1） 其中f ？ 是一个平稳的非线性函数，例如，一个泰勒一系列扩张的存在，e( t）是一个零的意思和是差分算子，ny, nu, ne和d分别是延迟了该系统已知的命令和时间。 让当地的线性模型的非线性系统（1）在作业点为o( t）是由以下控制自回归综合移动平均线（ carima ）模型： a其中，a(z1)= a(z1)，b(z1)和c(z1)是多项式在z - 1 落后的转向。请注意，该这些系数多项式函数的转向点为o（t）。非线性系统（1）分割成为几个作业区域，如每个地区可以近似当地的线性模型。自nfns是一类在本地的联想记忆网络的知识存储，他们可以应用到模型这一类非线性系统。示意图该nfn结果表明，在图2 。b-样条函数作为隶属函数在nfns由于以下原因。第一， b-条功能可随时在指定的秩序的基础功能和数目内，第二，他们是界定在一个范围内的支持和输出的基础上功能始终是积极的，即i.e., jk(x) = 0, x / jk, j 和 jk(x) 0, x jk, j ， 第三，根据职能的基础上形成一个分割的团体，i ,e k(x) 1, x xmin, xmax.第四，输出的基础上的功能可以得到由一个复发的方程。 图2模糊神经网络。 隶属函数的模糊变量的衍生从模糊规则可以得到由该变量的基础上的职能。作为一个例子，考虑该nfn显示在图2 ，构成以下模糊规则： 如果操作条件（ x1是积极的小，和xn 是消极的大型） ， 那么输出是由当地carima模型i:这里n是输入x和p的传染媒介维度，在nfn中重量工具的数字，它是给出的， 这里ki和ri是依据作用的命令和各自内在结的数字。单变量的b-多槽轴的依据作用以前描述了也适用于多维分布的依据作用，nfn的产品是是被定义的亢奋长方形。 3神经模糊的网络广义预测控制gpc通过使以下方式从而减到最小而获得作用 10， 这里qj和j分别为的衡量要素的预言错误和控制计算， yr (t + j)是jth前面参考道路，d是极小的费用、n和m分别为最大费用预言错误和控制的。 控制从nfgpc的计算是被衡量的总和的从gpc的地方p控制器获得的控制。 这里ui (t)是控制ith区域，i (x)是在(4)最早的方式。 注意在的重量nfgpc与那是相同的在nfn中塑造的过程。从交换gpc控制器之间 nfgpc介入模糊的逻辑，它可以被解释为没有，仅仅作为一个模糊控制器，还可以作为一位模糊的监督器。这种控制可以是光滑的，如果重量i (x)是适宜的建立，从nfn (6)和控制(8)，指定的j 通过(7)可以被重写如下： 由于使用不平等，价值函数首先被简化。从下面 化简 从(10)式暗示被衡量的被摆正的的总和可以是价值函数j.重写的一个最高界面通过(9)给出 这里 通过(11)式显示本质上是作为那使j.减到最小那使减到最小的ji本。 从(12)，一套地方广义预测控制器得到 nfgpc构成部分。 这里 和 通过计算前面优化m步控制，和使用后退前面第一步控制被实施，通过原则10，给 其中 可以简化为 4过热蒸汽温度的神经模糊塑造和有预测性的控制 让是过热蒸汽温度和， 过热蒸汽水对高温过热装置的。 可以通过由二次式样接近 11 ： (17) 式是线性模型，然而，仅一个地方模型为选择的工作点。 因为装载是独特的事，它被用于选择在nfn的地方之间分裂。 基于这种方法，如图3所显示，使用操作员的经验， 被划分成五个地区，看待装载200mw作为一个高度， 180mw作为中等上流，作为的160mw作为中等低落的方式、140mw和120mw如低。 为间隔时间30s，估计的线性本机塑造使用了使用了显a (z1)在nfn在表1.使用了显示。图三本地模式的隶属函数 通过nfgpc的时延d，也是极小的，被设置到30 s， 通过查控制nfgpc的m的作用来表现， m的价值选择与相对天际n设置了对的相对地大价值10. 对于小m，闭环反应是缓慢而合理的。通过发现是得到形式m= 6。因此，通过改善，当进一步增加m， nfgpc以后被用于控制实际能源厂。 级联控制计划的用途是广泛控制过热的蒸汽温度。 前馈控制，与蒸汽流动，并且作为输入的气体温度，可以是应用的提供对大变化的一个更加快速的反应在这两个方面。 实际上，前馈道路仅仅是被激活的，当这些可变物上的重大的变化时，这些控制方案也防止的更加快速的动力学植物，即，蒸汽水的阀门和混合的水或者的蒸汽， 影响植物，即，更加缓慢的动力学的高温过热装置。 全球性非线性nfn模型在表1提出的nfgpc计划是在下面的图4。 图四nfgpc控制过热蒸汽温度与前馈控制在nfn里面只暗示二个地方控制器激活了每次，确定由装载信号通过会员资格作用。 考虑负载变动，其中装载从140mw增加到在1%和2.5%/min.之间的速率195兆瓦20分钟和减退到160兆瓦每60分钟. 从fig.5，过热蒸汽的控制温度由nfgpc达到，作为温度两个的f向上和向下装载。改变在7c.之内。 这个结果与比较下的结果在一个380兆瓦单位被测试，中止干扰。 相反，在波动过热蒸汽温度使用是大幅度运用pi控制器，如fig.6所显示。图五nfgpc过热蒸汽温度控制图六过热蒸汽温度控制的梯级pi控制器 作进一步例证，能源厂被模仿在表给的nfn模型1，和是受控的由nfgpc，常规线性gpc的控制和落下的pi控制器，当装载改变160mw到200mw。 常规线性gpc控制器是为设计的地方控制器“medium”操作区域。 结果在fig.7， 显示表里，正如所料，最佳的表现得到nfgpc的方式，因为它被设计根据一更加准确模型。 这由常规线性跟随gpc控制器。 常规的表现pi控制器是最坏的，表明它无法令人满意的控制过热蒸汽温度下大负载变动。 这也许是控制的原因。 图七比较研究nfgpc ，传统的线性gpc的，和级联pi控制器 实际上，控制u (t) 的gpc通常是通过计算 (8)得到的。 然而，如果u (t)超出物理极限控制器，控制器饱和发生。 限制由二次规划算法的被合并优选的价值函数受作动器极限支配的 gpcs。 让在上的变化蒸汽水阀门的频率被限制： 考虑到nfgpc的表现限制给(18)为装载干扰40兆瓦在图8显示，蒸汽水控制正常0,1。 这个拘束的控制信号nfgpc似乎是有期望作控制器极限。 当u超出限制，新的控制信号得到，期望控制会超出极限。 图八比较限制，优化nfgpc和制约因素有限nfgpc表现5 结论通过一个200mw电厂的塑造和控制使用的神经模糊的方法在本文被提出。nfn包括五个地方模型。 网络产品是本机模型插值法b-多槽轴依据作用给的会员资格。nfgpc同样地被修建，是被设计的在nfn的carima模型。 nfgpc适用于与光滑的非线形性的过程它充分的操作范围可以被分成几个线性操作地区。 提出的nfgpc因此为控制提供一个有用的选择非线性电厂，以前使用传统方法是受控的，比较困难的。 journal of control theory and applications 2007 5 (1) 83-88 doi 10.1007/s11768-005-5258-6 neuro-fuzzy generalized predictive control of boiler steam temperature xiangjie liu 1, jizhen liu 1, ping guan 2 ( 1.department of automation, north china electric power university , beijing 102206, china; 2.department of automation, beijing institute of machinery, beijing 100085, china) abstract: power plants are nonlinear and uncertain complex systems. reliable control of superheated steam temper- ature is necessary to ensure high efficiency and high load-following capability in the operation of modern power plant. a nonlinear generalized predictive controller based on neuro-fuzzy network (nfgpc) is proposed in this paper. the proposed nonlinear controller is applied to control the superheated steam temperature of a 200mw power plant. from the experiments on the plant and the simulation of the plant, much better performance than the traditional controller is obtained. keywords: neuro-fuzzy networks; generalized predictive control; superheated steam temperature 1 introduction continuous process in power plant and power station are complex systems characterized by nonlinearity, uncertainty and load disturbance 1, 2. the superheater is an important part of the steam generation process in the boiler-turbine system, where steam is superheated before entering the turbine that drives the generator. controlling superheated steam temperature is not only technically challenging, but also economically important 3. from fig.1, the steam generated from the boiler drum passes through the low-temperature superheater before it enters the radiant-type platen superheater. water is sprayed onto the steam to control the superheated steam temperature in both the low and high temperature superheaters. proper control of the superheated steam temperature is extremely important to ensure the overall efficiency and safety of the power plant. it is undesirable that the steam temperature is too high, as it can damage the superheater and the high pres- sure turbine, or too low, as it will lower the efficiency of the power plant. it is also important to reduce the temperature uctuations inside the superheater, as it helps to minimize mechanical stress that causes micro-cracks in the unit, in or- der to prolong the life of the unit and to reduce maintenance costs. as the gpc is derived by minimizing these uctua- tions, it is amongst the controllers that are most suitable for achieving this goal. the multivariable multi-step adaptive regulator has been applied to control the superheated steam temperature in a 150 t/h boiler 3, and generalized predictive control was received 14 october 2005; revised 14 october 2006. proposed to control the steam temperature 4. a nonlinear long-range predictive controller based on neural networks is developed in 5 to control the main steam temperature and pressure, and the reheated steam temperature at sev- eral operating levels. the control of the main steam pressure and temperature based on a nonlinear model that consists of nonlinear static constants and linear dynamics is presented in 6. fig. 1 the boiler and superheater steam generation process. fuzzy logic is capable of incorporating human experi- ences via the fuzzy rules. nevertheless, the design of fuzzy logic controllers is somehow time consuming, as the fuzzy rules are often obtained by trials and errors. in contrast, neu- ral networks not only have the ability to approximate non- linear functions with arbitrary accuracy, they can also be trained from experimental data. the neuro-fuzzy networks (nfns) developed recently have the advantages of model transparency of fuzzy logic, and learning capability of neu- ral networks 7. the nfns have been used to develop self- this work was supported by the natural science foundation of beijing (no. 4062030), national natural science foundation of china (no. 50576022, 69804003), scientific research common program of beijing municipal commission of education (km200611232007). 84 x. liu et al. / journal of control theory and applications 2007 5 (1) 83-88 tuning control 8, 9, and is therefore a useful tool for de- veloping nonlinear predictive control. since nfns can be considered as a network that consists of several local re- gions, each of which contains a local linear model, nonlin- ear predictive control based on nfns can be devised with the network incorporating all the local generalized predic- tive controllers (gpc) designed using the respective local linear models. following this approach, the nonlinear gener- alized predictive controllers based on the nfns, or simply, the neuro-fuzzy generalized predictive controllers (nfg- pcs) are derived here. the proposed controller is then ap- plied to control the superheated steam temperature of the 200mw power unit. experimental data obtained from the plant are used to train the nfn model, and from which lo- cal gpcs that form part of the nfgpc is then designed. the proposed controller is tested first on the simulation of the process, before applying it to control the power plant. 2 neuro-fuzzy network modelling consider the following general single-input single-output form a partition of unity, i.e., j j mk (x) 1, x xmin , xmax . and fourth, the output of the basis functions can be obtained by a recurrence equation. nonlinear dynamic system: y (t) = f y(t - 1), , y(t - n y), u(t - d), , u (t - d - n u + 1), e(t - 1), , e(t - n ) +e(t)/d, e fig. 2 neuro-fuzzy network. the membership functions of the fuzzy variables derived from the fuzzy rules can be obtained by the tensor product of the univariate basis functions. as an example, consider the nfn shown in fig.2, which consists of the following fuzzy rules: if operating condition i (x 1 is positive small, , and xn is negative large), then the output is given by the local carima model i: yi (t) = ia y (t - 1) + + a 1 i ina yi (t - n a) (1) + i0b du i(t - d) + + binb +e i(t) + + cinc du i(t - d - n b) where f . is a smooth nonlinear function such that a tay- lor series expansion exists, e(t) is a zero mean white noise and d is the differencing operator, n y_, n _ , n _ and d are re- spectively the known orders and time delay of the system. let the local linear model of the nonlinear system (1) at the operating point o(t) be given by the following controlled auto-regressive integrated moving average (carima) model: u e (z_ -1 )y(t) = z -d b (z -1 )du(t) + c(z- 1)e(t), (2) e (it - n c) or i戀 (z -1 )yi (t) = z -d db i(z -1 )u i(t) + c i(z -1 )e i(t), (3) where a (z-1 i ) , b i(z -1 ) and c i(z -1 ) are polynomials in the backward shift operator z-1 , and d is the dead time of the plant, u i(t) is the control, and e i(t) is a zero mean inde- pendent random variable with a variance of s 2. the multi- variate basis function a i(x k) is obtained by the tensor prod- ucts of the univariate basis functions, n k=1 ai = maki (x k), for i = 1, 2, , p, (4) where a(z- )1 nomials in z-1 = da(z-1 , ) b(z-1 ) and c(z-1 ) are poly- , the backward shift operator. note that the coefficients of these polynomials are a function of the op- erating point o(t). the nonlinear system (1) is partitioned into several operating regions, such that each region can be approximated by a local linear model. since nfns is a class of associative memory networks with knowledge stored lo- cally 7, they can be applied to model this class of nonlin- ear systems. a schematic diagram of the nfn is shown in fig.2. b-spline functions are used as the membership func- tions in the nfns for the following reasons. first, b-spline functions can be readily specified by the order of the basis function and the number of inner knots. second, they are defined on a bounded support, and the output of the basis j function is always positive, i.e., m k(x) = 0, x / l j and m (x) 0, x l k j-k j-k , l j , l j. third, the basis functions where n is the dimension of the input vector x, and p, the total number of weights in the nfn, is given by, n k=1 p = (r i + k i), (5) where k i and r i are the order of the basis function and the number of inner knots respectively. the properties of the univariate b-spline basis functions described previously also apply to the multivariate basis functions, which is de- fined on the hyper-rectangles. the output of the nfn is, p yi ai y = i=1 p i=1 = yi a i. (6) ai p i=1 x. liu et al. / journal of control theory and applications 2007 5 (1) 83-88 neuro-fuzzy network generalized predic- tive control the gpc is obtained by minimizing the following cost n 85 3 = e p i=1 p i 2 _ j=d (a ) q j yi (t + j) - y r(t + j)2 i function 10, j = e n i=1 2 _ j=1 m + p (a ) l jdu i(t + j - 1)2 = j y r (a i) j i, 2 q (t + j) - y (t + j)2 (11) j=d m l du(t + j - 1) 2, j i=1 where + (7) = e j=1 ji n j=d q j yi (t + j) - y r(t + j) 2 l jdu i(t + j - 1) . 2 where q j and l j are respectively the weighting factors for the prediction error and the control, y (t + j) is the jth r step ahead reference trajectory, d is the minimum costing horizon, n and m are respectively the maximum costing horizon for the prediction error and the control. the con- trol computed from the nfgpc is the weighted sum of the control obtained from p local gpc controllers: p i=1 du(t) = a idu i(t), (8) + m j=1 equation (11) shows that minimizing ji (12) is essentially the same as that of minimizing j . from (12), a set of local gen- eralized predictive controllers is obtained, which forms part of the nfgpc. the local gpc 10 is given by, du (t) = (g tq g + l ) i i i i i -1 g itq iy r(t + 1) -f idu i(t - 1) - s i(z -1 )y i(t), (13) where yr (t + 1) = ry (t + 1), ry (t + 2), , ry (t + n ) t, du i(t) = du i(t), du i(t + 1), , du i(t + m - 1) t, du i(t - 1) = du i(t - n b), du i(t - n b + 1), , du i(t - 1) , t ) , , sin si si (z -1 ) = s 1i( z -1 ) ,s 2i( z -1 (z -1 t ) . where du i(t) is the control in the ith region, a i(x) is defined previously in (4). note that the weights in the nfgpc are identical to that in the nfn that models the process. since switching between local gpc controllers in the nfgpc involves fuzzy logics, it can be interpreted not only as a fuzzy controller, but also as a fuzzy supervisor. the control can be smooth if the weights a i(x) are suitably selected. from the nfn (6) and the co