Generalized Predictive Control To Simplify The Algorithm,

The generalized predictive control (GPC) was proposed originally by Clarke and his collaborators in 1984, which used the traditional parameter models (such as CARIMA model). The number of model parameters is small and it is easy to estimate parameters online for slow time-varying systems. The unequal forecast level and control level are introduced in GPC. Forecast model, rolling optimization and feedback correction are taken as three basic characteristics in GPC which shows excellent control performance and is considered as one of the representative predictive control algorithms.It is paid great attention by the academic and engineering. However, heavy on-line computation burden which is caused by computing inverse matrix in GPC algorithm is not suitable for real-time control system. Base on a large number of domestic and foreign literatures, generalized predictive control algorithm is simplified in this paper. Lots of simulations and experimental studies are done to typical models of industrial process.(1)Fast Generalized Predictive Control Algorithm with Non-overshoot is proposed for a class of system requirement based on integrating stair-like GPC and varying trend of control increments. The heavy on-line computation burden for computing inverse matrix is avoided by single-valued GPC and current input increment is compensated by next step input to overcome potential overshoot. The steps of the algorithm is described in detail .Simulation results show the algorithm has good control performance in response speed, anti-interference, restricting the overshoot and robustness.(2)A new output increment feedback GPC with input constraints is proposed for real control system with constraints. The operation of inverse matrix is avoided by approximately computing the future control increment sequence off line, so that the on-line computation burden is strongly reduced. Moreover, with inputs and their increment constraints, the algorithm can obtain the optimal control input by adjusting the set-value of max-output increment and guarantees that output close to the set-value of the plant. Simulation results show that the proposed algorithm is effective.(3) An output increment feedback predictive control is simplified by approximately computing future control increment sequence off line and the computation of inverse matrix is avoided. The maximal output increment is adjusted online by neural network which guarantees that output close to the set-value of the plant. Simulation results show that the algorithm proposed is effective.(4)According to a large time-delay and slow time-variation object, an improved GPC (Generalized Predictive Control Algorithm) possessing rectifier filter function is proposed, which revise the model predictive output with actual error, integrate stair-like control law and utilize the varying trend of control increments based on the general GPC. A water tank fluid position control system is designed by using Kingview and the algorithm proposed is tested. The experimental results indicate that, compared to general GPC, the algorithm proposed has better performance in overshoot restraint, anti-interference and robustness.The proposed algorithm in this paper was derived not only in theory, but also carried out in a large number of simulation experiments and experimental researchs which show the effectiveness of the proposed algorithm.