Study On Synthesis Approach Of Model Predictive Control And Its Applications For Rotary Kiln Calcinations Process

Since proposed in the1970s, Model Predictive Control (MPC) has been widely used in industry, and MPC theory study has also caused attention of industrial and academic field.This is mainly because of its strong adaptability on complex industrial process. Synthesis approach of model predictive control(S-MPC) with guaranteed stability has become an important branch of the current predictive control. For S-MPC fully uses as reference the optimal control, lyapunov analysis, invariant set and other mature theories, it has promoted new leaps for predictive control theory study and analysis, and has obtained plenteous research results. But unlike the classical predictive control methods are widely used, a big gap is still in existence between these results and the actual industrial applications. There are two main reasons. On the one hand, for most industrial process complex mechanism, it usually exists parameter time varying, uncertainty, multivariable, strong coupling, large inertia and time-delay, etc; on the other hand, for synthesis approach of model predictive control uses state-space model and LMI techniques, there will exist problems such as modeling difficult, model order over high, large amount of online calculation, states non-measurable and initial feasible region limited, etc. The study starting point of this paper is taking actual control demand of typical complex industrial process rotary kiln as background, and researches above problems. Furthermore, after analysis and improvement of S-MPC method, we design an effective strategy and thinking which apply complex S-MPC method in actual industrial process. The primal research contents of the thesis and its contributions are as followes:To solve the problem that large amount of S-MPC online calculation, we take off-line predictive on-line synthetic control method. At first, gives a sequence of N explicit control laws and parameters corresponding to a sequence of N asymptotically stable invariant ellipsoids constructed off-line. Then, on-line compute corresponding state feedback matrix convex combination of two invariant ellipsoids adjacent to the current state. To overcome the optimality loss of the off-line algorithm, an idea of variable terminal constraints set on-line is introduced. By adding optimal parameter9, on-line optimization freedom is increased and the system optimality is improved. At the same time, predictive control algorithm with variable constraints(VC-MPC) is introduced for reducing on-line computational demand. Determine the corresponding control constraints according to different stability ellipse domains where the current state lies in. The scope of the newly created constraints does not affect stability and optimality of solution, and the scope often less than the original constraint value. As search range is reduced, the on-line computational demand of VC-MPC decreases significantly.Aiming at actual industrial process often meets the problem of unmeasured states and dynamic state target tracking, a novel formulation of output-tracking MPC is presented. The matrix that transforms the target output into steady state target is given explicitly. This is convenient for on-line application. Moreover an improving Luenberger robust state observer with LMI forms quadratic stabilization is designed in output feedback, and the stability test criterion of augmented closed-loop system with observer is also given by using off-line method characters. At last a whole output-tracking model predictive control method is introduced, and carries out stability analysis.Realize the adaptive control of output-tracking model predictive control method by recursive subspace. To speed up the convergence rate, model matching error-based time-varying forgetting factor is adopted when reconstructing Hankel matrix. To avoid QR decomposition at each step and save computing time, utilizing the Givens rotation strategy. Simultaneously, to ensure signal fully incentive and not affect the control effect when model identification, the random excitation signal is applied based on model errors.To solve the problem of controlled object with time-delay and large operating regions, we propose the time-delay gain scheduled model predictive control method. At first, compute the compact set which the future state lies in by Schuurmans method, and give the performance index, stability constraints, input and state constraints in terms of LMIs for optimization solution. Based on this, off-line and on-line S-MPC control law for time-delay system is designed. Furthermore, in order to solve the problem of large range target tracking, time-delay gain scheduled model predictive control method is proposed. In each region of stability design the time-delay MPC, and implement them as a single scheduled MPC with on-line switching between the set of local controllers. Finaly, drive states achieving the desired target.The calcining belt state-space model of rotary kiln is built by using PO-Moesp subspace method. After analyzing the main components technology and calcinations reaction mechanism in detail, the input and output variables of modeling are selected according to industrial field operational procedures. A novel order-selecting method is put forward. It takes a new double performance parameters error criterion includes order and delay replace the original single performance parameters error criterion includes only order. This improvement effectively solves the problem of model order too high when modeling object with time-delay, and it also increases the precision of model.Take rotary kiln as the background, calcining belt temperature predictive control system is designed. The control system consists of set value module, calcining belt temperature soft sensor, synthesizing model predictive control module, etc. Aiming at the problems that calcining temperature is difficult to measure directly and is lack of accurate data for soft sensor modeling; soft sensor model of calcining belt is constituted through kiln head model slope and bias correction use model migration algorithm. At the same time, combining time-delay gain scheduled, VC-MPC, output-tracking, recursive subspace adaptive and other methods, and then the off-line/on-line predictive controller of rotary kiln is formed. At last, MATLAB is applied for simulation, experiment runs in set point tracking and mobile tracking situation. Simulation results show its effectiveness and feasibility.