1、12 2 Fn Vol. 12 No. 21999 M6 JOU RNAL OF PETROCHEM ICAL U NIVERSIT IES Jun. 1999RBF * “dK e史国栋 王其红 薛国新 徐 燕( F, 213016)K 1 为了实现系统的预测最优化控制, 问题的关键是如何准确而迅速地对未来一段时间内的系统状态进行预测,然后利用此预测结果及优化指标来控制有关系统变量, 人们对此进行了许多研究,但尚有其不足之处b为此,提出压缩预测变量集规模等办法来增加 RBF 神经网络有效预测时间长度, 在此基础上利用稳态最优解和优化指标来控制有关量b此方法被用于某化工过程, 结果能使系统运行
2、更为平稳,并使有关量达到了预测的优化指标b1oM RBF 神经网络; 预测; 稳态; 最优控制ms | TP273 M ,K eTB5 e m , s jbV,YVK e 3 , V -4/ P1 qr V ?,V7 |A6rmb LCK e,1oB b1.d ZE, MYE =ME, P * ZE pz a 7 s 1 1b * PBP * , HW , OJ lb7RBF* 7 s j 1, 4, 5byN|RBF * Vb H, “ 1 1 , y “,t4 * y ZE 2bNZE 1 ab y, “d, VCv8b - T M“d1o Y1BHWV ?8C, HY| HWb 6, t
3、 Z4M “d # z9 NVT“d1o 3YbyN,收稿日期: 1999- 03- 30第一作者:男, 43岁,教授b 5 ,“dsZF, t T l M 1 vSM,7 6Bt 51VM HW T M 3 AYb P eM CC,9Y * r HWbN,4 P 9 T!RBF “,i “d s ,YV K e r “b1 RBF 预测神经网络 m yj, 1 j m f,y= (y1, , , ym)T V Um bW, !X “( x(k),y(k) | 1 k Kbx(k) I Rnbx I Rn, VRBF TbB f/, | H n H K, HRBF Tyj = EHh= 1wj
4、h exp - (x - xh)2R2h 1 j m(1) TMRBF * , x s ;H ; Rh, x h ; wjh(1 h H , 1 j m) W bI n“dK e5,| s : u1, , uRM1, , MSbB R 1 p ,= S 5 (11 p b |B HW3, 2QW HWt1= t- (Q- 1)$t, t2 = t-(Q- 2)$t, , , tQ = t, tQ+ 1 = t+ $t, , ,t2Q= t+ Q$tF,u(q)rM(q)SsYV UurMS HW tqb 7x = (u(1)1 , , u( 1)R ; M( 1)1 , , M(1)S ; ,
5、; u( Q)1 , ,u(Q)R ; M(Q)1 , , M(Q)S )T ( 2)y = (M( Q+ 1)1 , , M( 2Q)1 ; ,; M( Q+ 1)S , , M( 2Q)S )T( 3) n= (R+ S)Q, m= QSb , (2) Tx(1) T (3) Tyb 1 V“ b RBF * r,A | M b2 预测最优控制方法 L=“dK e5,4/n bRBF * M “,c“d M , O *t ?8C ? o pM b“, Vvvhl5? bi O V|5 e5s 7b5, eM T1,QV T “d M , M 1 eM N1, PT eM yT,AY9 rb
6、 M , | ,7 | BHW F s,“, V9FTr HWb| tF f |9f ,“, L - Y v, L=5? pMBb| 1 p |1M , 1 p“d M $T % 1Hqby s5, p KbiNK “S, e TM b 5 c dLZFK5b BT /: T T Hqf i(x 1, , xn, A1, , AS) = 0 1 i nUj(x 1, , xn, A1, , AS) = 0 1 j m(4), A1, , AS Q THqBFMb Hq/ “Sf m inx1, , xnJ(x 1, , xn) (5)YZE p K5x(0)1 , x( 0)n ; A(0)1
7、 , , A(0)Sb P TM A1, ASrA(0)1 , , A(0)S HW$T(0),1PID eb3 应用实例 K e, ZFB ,/ T:Mij (xij, Vj, Tj) = Axij - Dj = 01 i c 1 j N (6)Sj(xij, Vj , Tj) = Eci= 1Kijxij - 1 = 01 i N (7)Hj(xij , Vj, Tj ) = (hVj+ 1- hLj ) Vj+ 1- (hVj - hLj )( Vj + Gj ) - (hLj - hLj- 1)Lj- 1+(hFj - hLj )Fj - Qj = 0 2 j N - 1(8), c8
8、“Fs , iFs ; jV U ) , N 9) ; Fj) ; Vj ) M ; Lj )AM ; Tj ); xij )AMF; K ij ; QjWF b(6) TcNZT,(7) T(8) TNZ T,yN, ) (r H Z9 (c+ 2)Nb M 5 82 Fn 12 )FsAM ixij, cN, )Tj , N,# ) M Vj ,9Nb# M 9 ( c+ 2) NbyNM Z Mb ZF VbI nt F 7#Q F 7 K ebN f/,1PID e 1 # / T:m F , 1, D 7,t D7, F 7 D 7, F 7 D 7b V HWM? p /:Fin
9、= Fin(t) ( 9)T in = T in(t) (10)| sYZ 7 ) b B sY:t(1)Ft(1)Tb |$3( 0) = 14 m in(t( 1)F , t(1)T ) (11)N HWT P TM K HW e_ PID e b_ | PK eZE q5 qT 1 bnV1bV1 Vn, PZE e,t a F 7 q991545% 10%b 6 V A, e“db y“d KA1HqBb4 结 论K e B5,4 %ZEb|RBF *“d M “,i4| M HWV F sT,4 r,41 P“d sK e TM ZE, ?4 l q,7 O9 ? P“dbV1 q1
10、 Table 1 Comparasion between the product ratest/ min航煤产率, %优控前值 优控后值轻石脑产率, %优控前值 优控后值重石脑产率, %优控前值 优控后值15 25. 29 26. 75 11. 62 12. 74 10. 05 10. 7130 25. 00 26. 85 11. 80 12. 81 10. 04 10. 7545 24. 15 26. 65 12. 08 12. 87 9. 90 10. 7160 25. 50 26. 63 11. 66 12. 90 9. 74 10. 7975 24. 00 26. 65 12. 15
11、12. 92 10. 08 10. 8290 25. 20 26. 68 11. 89 12. 86 9. 95 10. 90105 24. 75 27. 05 11. 74 12. 81 10. 22 10. 86120 24. 63 26. 95 11. 60 12. 80 9. 91 10. 88 ID1 Foster W R, Collopy F, Ungar L H. Neural network forecasting of short noisy time series. Computers Chem.Enging. , 1992, 16( 4) : 293 2972 M ich
12、ael L T, M ark A K. M odeling chemical processes using prior knowledge and neural netw orks. AICHE Journal ofprocess System Engineering, 1994, 40( 8) : 496 5023 符曦. 系统最优化及控制.北京: 机械工业出版社, 1995. 459 4854 Ruan R R, Almer S, Zhang J. Prediction of dough rheological properties using neural networks. Cere
13、al Chemistry,83 2 NS. RBF * “dK e1995, 72( 3) : 7 135 Daviod D Z. Neural networks system design methodology. PeKing: Tsinghua University Press. , 1996. 1 7Application of RBF- Neural Network Combined w ith the Optimal StableSolution for the Predicative Optimum Control of SystemShi Guodong Wang Qihong
14、 Xue Guoxing Xu Yan(Jiangsu Institute of Petrochemical Technology, Jiangsu Changzhou 213016)Abstract To realize the predicative optimum control of a system, the key of t he problem is how to predicate the states oft he system in the coming time space accurately and quickly and then use the predicate
15、d results and the opt imal indexes tocontrol the related system variables. A lots of study w orks in this aspect have been done. But there are still many short-comings. This paper proposed a method by which the scale of the predicted variables set could be reduced. By combiningt his method w ith oth
16、er methods, the valid predication time length of t he RBF- neural network was increased. Based ont his, the optimum stable state of t he system and the optimum indexes were used to control related variables. The met hodspresented in this paper have been applied to the simulation of a chemical proces
17、s. The results show ed t hat t he system couldreach t he indexes as well as run in a stabler state.Keywords RBF- neural network; Predication; Stable state; Opt imum( Ed. : ZW. W. )(上接第80页)Self- T urning PID Controller Based on theRealization of Neural NetworksLi Chunxiang Zhong Biliang(Guangdong Col
18、lege of Petrochemical Technology, Guangdong Maoming 525000)Mao Zhongyuan(Department Automation, South China University of Technology, Guangdong Guangzhou 510641)Abstract A self - adaptive PID controller based on neural network is introduced in this paper. It has not only thelearning ability and the
19、adaptability, but also the self- adjusting factor function. Digital simulation and experiment resultsshow that this new controller can improve the dynamic performance of temperature regulation system and robustness.Keywords Neural network; Self- adaptive; PID control; Stabilizat ion( Ed. : ZW. W. )84 Fn 12
