Journal of Science & Technology 101 (20i4) u A Disturbance Identification Method Based on Neural Networlc for a Class Predictive Control System with Delay Le ThiHuyen Linh', Lai Khac Lai^, Cao Tien Huynh^* 'Thai Nguyen University of Technology 3-2 Street Thai Nguyen City, Vietnam 'Thai Nguyen University ^Institute of Automation - Militaiy Technology, 89B, Ly Nam De Sir., Hoan Kiem District Hanoi Received March 04, 2014; accepted: April 22, 2014 Abstract Industrial objects with delay are normally influenced by various kind of disturbance, especially the disturbance can not be mesured The impacts of disturbance lower the quality of system control significantly, even in some cases, they make system unstable The identification of unmesuared disturbance is a challenging problem, attracting the interests of many scientists Therefore this paper proposes a disturbance identification method for the class of industnal objects inluding delay in order to determine disturbance and set a foundation to offset its impacts based on the using of Radial Basic Functions (RBFj Neural Networks and parallel model The weights updated rule of RBF Neural Netwoik ensures the convergence of disturbance indentification process and the system stabilization Keywords: Model Predictive Control, delay, disturbance identificahon RBF Neural Nehwork I Introduction The objects containing delay are common in oil fiiation industry, petrochemiscal industry, chemiscal industry, food industry and paper industry These objects are often affected by different types of disturbance, especially unmeasureable disturbance The existence of delay effects and other noises limit the quality of system, even lead to unstable in some situations There are many proposed methods to establish control systems for objects contaming delay [1-7], One of the noticeable method is control system with predictive model (MPC - Model Predictive Control) impacts will be solved and the online optimizafion for MPC system will be more feasible This paper suggests the identifymg disturbance method based on RBF Neural Network for object class regarding to disturbance Disturbance identification problem for object class with delay in control channel Suppose the dynamic of objects with delay is descnbed by the equation, y'"'W f E « ,?/'"(*) = Ku(t - r) + fiy,y^'\ -,t) (1) Where: Model Predicfive Control showed its own preement for objects containing delay, slow dynamic plants and in case of constrain with control signal and state vector [1,5,6], However, one of the mam difficulties of MPC is lookmg for online optimal solufion It can be more difflcuh under the impacts of disturbance, especially unmeasured disturbance, [1,5], To reduce the difficulty metioned above, a requirement of identifying and offsetting its effects is essential This problem has not been resolved adequately When impacts of disriirbance on system are identified, the compensated problem of their Corresponding Author: Tel: (+84) 982 731 666 Email caoihh(S;gmail.com y(t) - output of control object u{t) - manipulated variable, \u{t)\ < U^^^^,^ T- delayed time a,4 = 0,1,2.•••n-l:K of objects' dynamics, - characteristic parameters /{) = f{y>'!i'\ -,i) - unmeasured disturbance smooth nonlinear fijncfion (to be able to perform mulfiple differenfial), state depend on disturbance and change slowly f{)^0 (This process occurs much slower than dynamic of objects If it changes speediy, it is imposible to deal with) This type of noise is common in mdustrial fields [8] Journal of Science & Technologj' 101 (2014) OZO-025 Where A =A State variables: a,,^ = B r^ = r, F(Y) = (0 /(Y)j' Due to nonlinear function /(Y) meets the conditions of Stone - Weierstrass [9] theorem, therefore, by using RBF Neural network we can approximate with any accuracy »,(') = »(*) s,(i) = v"-"m (3) Due to the slow change of /{•) along with state variables above, the unstable nonlinear function describing disturbance can be simplified as / ( Y ) Where: w',? = l,2, -,m-are idea! weights; E approximate error, which satisfy e < £,,, while £,^10 In state variables space, equation of objects" dynamic (1) will be formed as the smallest random given number, 11 (Y).f -= l,2, -,m- Basic fiinctions are chosen as Y(t)=AY{t)-i-Ba(t-z) + F{Y) (la) [14]: Where: 0 ( l|Y-c|f 0 ; B- exp — 0 K - „ — I ^^ F(Y) = o(Y) = Y^cxp — J(Y)' The given problem is to identify disturbance for to set a foundation to offset its impacts In case of using Model Predictive Control (MPC), if disturbance is identified, the online optimizafion can be easily implemented [5,6] (4) |Y-CJ| 2X-), Substituting (15) and (16) into (14) gives : V = E'(i) - Q -AQj"a'(^)rf^ + APu'(() E(i) + 2AF'()PE(()/^'(Orf^ + E » A Substituting (II) into (17) regarding to (7) and (8) gives' V = E'(i)Lq - AQJôã(?)