自動(dòng)化專(zhuān)業(yè)外文翻譯---鍋爐蒸汽溫度模糊神經(jīng)網(wǎng)絡(luò)的廣義預(yù)測(cè)控制（英文）

1、Journal of Control Theory and Applications 2007 5 (1) 83–88 DOI 10.1007/s11768-005-5258-6Neuro-fuzzy generalized predictive control of boiler steam temperatureXiangjie LIU 1, Jizhen LIU 1, Ping GUAN 2(1.Department of Aut

2、omation, 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.

3、 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. Anonlinear generalized predictive controller based on

4、neuro-fuzzy network (NFGPC) is proposed in this paper. The proposednonlinear controller is applied to control the superheated steam temperature of a 200MW power plant. From the experimentson the plant and the simulation

5、of the plant, much better performance than the traditional controller is obtained.Keywords: Neuro-fuzzy networks; Generalized predictive control; Superheated steam temperature1 IntroductionContinuous process in power pla

6、nt and power station arecomplex systems characterized by nonlinearity, uncertaintyand load disturbance [1, 2]. The superheater is an importantpart of the steam generation process in the boiler-turbinesystem, where steam

7、is superheated before entering theturbine that drives the generator. Controlling superheatedsteam temperature is not only technically challenging, butalso economically important [3].From Fig.1, the steam generated from t

8、he boiler drumpasses through the low-temperature superheater before itenters the radiant-type platen superheater. Water is sprayedonto the steam to control the superheated steam temperaturein both the low and high temper

9、ature superheaters. Propercontrol of the superheated steam temperature is extremelyimportant to ensure the overall efficiency and safety of thepower plant. It is undesirable that the steam temperature istoo high, as it c

10、an damage the superheater and the high pres-sure turbine, or too low, as it will lower the efficiency of thepower plant. It is also important to reduce the temperaturefluctuations inside the superheater, as it helps to m

11、inimizemechanical stress that causes micro-cracks in the unit, in or-der to prolong the life of the unit and to reduce maintenancecosts. As the GPC is derived by minimizing these fluctua-tions, it is amongst the controll

12、ers that are most suitable forachieving this goal.The multivariable multi-step adaptive regulator has beenapplied to control the superheated steam temperature in a150 t/h boiler [3], and generalized predictive control wa

13、sproposed to control the steam temperature [4]. A nonlinearlong-range predictive controller based on neural networksis developed in [5] to control the main steam temperatureand pressure, and the reheated steam temperatur

14、e at sev-eral operating levels. The control of the main steam pressureand temperature based on a nonlinear model that consists ofnonlinear static constants and linear dynamics is presentedin [6].Fig. 1 The boiler and sup

15、erheater steam generation process.Fuzzy logic is capable of incorporating human experi-ences via the fuzzy rules. Nevertheless, the design of fuzzylogic controllers is somehow time consuming, as the fuzzyrules are often

16、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 betrained from experimental data. The neuro-fuzzy networks

17、(NFNs) developed recently have the advantages of modeltransparency of fuzzy logic, and learning capability of neu-ral networks [7]. The NFNs have been used to develop self-Received 14 October 2005; revised 14 October 200

18、6.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 Commissi

19、on of Education (KM200611232007).X. LIU et al. / Journal of Control Theory and Applications 2007 5 (1) 83–88 853 Neuro-fuzzy network generalized predic- tive controlThe GPC is obtained by minimizing the following costfun

20、ction [10],J = EN ?j=d qj[? y(t + j) ? yr(t + j)]2+M ?j=1 λj[Δu(t + j ? 1)]2, (7)where qj and λj are respectively the weighting factors forthe prediction error and the control, yr(t + j) is the jthstep ahead reference tr

21、ajectory, d is the minimum costinghorizon, N and M are respectively the maximum costinghorizon for the prediction error and the control. The con-trol computed from the NFGPC is the weighted sum of thecontrol obtained fro

22、m p local GPC controllers:Δu(t) =p ?i=1 αiΔui(t), (8)where Δui(t) is the control in the ith region, αi(x) isdefined previously in (4). Note that the weights in theNFGPC are identical to that in the NFN that models thepro

23、cess. Since switching between local GPC controllers inthe NFGPC involves fuzzy logics, it can be interpreted notonly as a fuzzy controller, but also as a fuzzy supervisor.The control can be smooth if the weights αi(x) ar

24、e suitablyselected. From the NFN (6) and the control (8), J given by(7) can be rewritten as:J = EN ?j=d qj[p ?i=1 αi(? yi(t + j) ? yr(t + j))]2+M ?j=1 λj[p ?i=1 αiΔui(t + j ? 1)]2. (9)The cost function is simplified firs

25、t using the Cauchy in-equality. Since[p ?i=1 αi(? yi(t + j) ? yr(t + j))]2? pp ?i=1 [αi(? yi(t + j) ? yr(t + j))]2,hence[p ?i=1 αiΔui(t + j ? 1)]2 ? pp ?i=1 [αiΔui(t + j ? 1)]2.(10)Equation (10) implies that the sum of t

26、he weighted squarederrors can be an upper bound of the cost function J. Rewrit-ing (9) givesEN ?j=dp ?i=1 qj[αi(? y(t + j) ? yr(t + j))]2+M ?j=1p ?i=1 λj[αiΔui(t + j ? 1)]2= Ep ?i=1 (αi)2 N ?j=d qj[? yi(t + j) ? yr(t + j

27、)]2+p ?i=1 (αi)2 M ?j=1 λj[Δui(t + j ? 1)]2=p ?i=1 (αi)2Ji, (11)whereJi = EN ?j=d qj[? yi(t + j) ? yr(t + j)]2+M ?j=1 λj[Δui(t + j ? 1)]2. (12)Equation (11) shows that minimizing Ji is essentially thesame as that of mini

28、mizing J. From (12), a set of local gen-eralized predictive controllers is obtained, which forms partof the NFGPC. The local GPC [10] is given by,ΔUi(t) = (GT i QiGi + λi)?1GT i Qi[Yr(t + 1)?FiΔUi(t ? 1) ? Si(z?1)yi(t)],

29、 (13)whereYr(t + 1) = [ ? Yr(t + 1), ? Yr(t + 2), · · · , ? Yr(t + N)]T,ΔUi(t) = [Δui(t), Δui(t + 1), · · · , Δui(t + M ? 1)]T,ΔUi(t ? 1) = [Δui(t ? nb), Δui(t ? nb + 1),· · ·

30、 , Δui(t ? 1)]T,Si(z?1) = [Si1(z?1), Si2(z?1), · · · , SiN(z?1)]T.Si(z?1) and Ri(z?1) satisfy the Diophantine equation:1 = ¯ Ai(z?1)Rij(z?1) + (z?j)Sij(z?1), (14)andGij(z?1)= Bi(z?1)Rij(z?1)= gi j,0 +

31、 gi j,1z?1 + · · · + gi j,nb+j?1z?(nb+j?1), (15a)Qi = diag(qi1, qi2, · · · , qiN), (15b)λi = diag(λi1, λi2, · · · , λiM), (15c)GT i =?? ? ? ? ? ?gi 1,0 gi 2,1 · · &#

32、183; gi N,N?1gi 1,0 · · · gi N?1,N?2 ... . . . 0 gi N?M+1,N?M?? ? ? ? ? ? , (15d)Fi =?? ? ? ? ? ?gi 1,nb gi 1,nb?1 . . . gi 1,2 gi 1,1gi 2,nb+1 gi 2,nb . . . gi 2,3 gi 2,2 . . . . . . . . . . . .gi N,nb+N?

33、1 gi N,nb+N?2 . . . gi N,N+1 gi N,N?? ? ? ? ? ? . (15e)The optimized M steps ahead control is computed, andonly the first step ahead control is implemented, using a re-ceding horizon principle [10], givingΔui(t) = dT i1[