Abstract: Three modifications are introduced into the Generalized Predictive Control (GPC) algorithm proposed by Clarke et al.(1984):
1) The recursive solution to the Diophantine equations within the procedures for multi-step ahead predictions of the controlled plant output is simply replaced by the successive one-step ahead predictions directly based on the plant model equation.This reduces the computational effort greatly;
2) The approximate values and mumerical ranges of the parameters and the tending-to-one theorem for parameter sum of the sampled model with relatively small sampling period are applied in setting the initial values of parameter estimation.Thus, a satisfactory response to the step setpoint change can then be gained even during start-up of the control;
3) Employing an optimal moving average filter, the effect of the stepwise or sinusoidal deterministic disturbance may be suppressed without excessive amplification of the random noise.
The modified GPC algorithm based on these improvements is derived, and it still keeps the all robust properties of the original GPC algorithm.