Predictive Control Algorithm And Its Application. Iteration

As to a complicated object with long time delay, slow time variability, strong disturbance and considerable difficulty to model exactly, it is difficult to achieve the desirable control effect by modem control theory which is based on state space method. Faced with such specific character of industry process control, based on the finite time domain optimizing rule of MPC, and enlightened by the step-by-step amendment ability of iterative learning, a kind of iterative learning predictive control algorithm (ILPC) is proposed in this paper.According to the error between the model predictive output and the future expected output, ILPC carries out an iterative learning and amending process on the current and the future control input vector, namely performs "forecast, iterative amendment, forecast again, iterative amendment again" repeatedly in the iterative domain. But in the time domain, only the current control variable of iterative predictive control vector is put on the controlled process, and the "progressive optimization" is carried out in the next time, in order to emendate the uncertainty caused by model mismatch, time variation and disturbance. The concept of "iterative amendment" and the "progressive optimization" achieves favorable condition for itself.In view of the specific character and control request of the controlled process, this paper proposes and studies a series of iterative predictive control algorithm on basis of nonparametric model, structured parameter model and neural network model, so that the sphere of application is expanded from linear system to nonlinear system. Compared with MPC, ILPC needs no matrix inverse operation, so that less calculation is required; moreover, it accords with the character of computer control. On the other hand, the performance function of MPC must contain the control constraints so as to satisfy the condition of inverse matrix. But ILPC doesn't care whether the matrix is reversible, so that the constrain condition of predictive control law is loosen which is essential for more excellent control capability.Another work of this paper is to put forward a kind of amphimictic genetic algorithm (AGA) based on error classification idea, which can effectively enhance the optimization search speed. AGA discards the optimizing and searching process in the negative gradient aspect of performance function. Thus, AGA keeps ILPC from falling into partial optimization and makes it in line with the character and request of nonlinear system control.Based on the iterative predictive control theory, a new temperature control system for high-power arc furnace has been developed and put into industrial practice successfully, which, running well continuously with remarkable benefits, has passed the provincial-level authentication by the Science & Technology Department of Hunan province. Meanwhile, this system has been awarded with the second prize for Advanced Science & Technology of Hunan Province (No.2002350218-2-05) and listed among the National Prior New Product Projects (No.2002ED770005).