Data-driven pattern moving and generalized predictive control

In this paper, a generalized predictive control method based on pattern recognition technique is proposed for a class of uncertain complex systems with statistical characteristics. The system dynamics is described by working condition pattern moving rather than output equation or state space model. First, working condition data are collected around the nominal operating condition to construct “pattern moving space”. The space can be updated on-line. Then “pattern class variable” is defined in this space and the prediction model based on this new variable is also constructed. At last, the generalized predictive control method is also given. This method can identify the optimal controller´s parameters directly by the prediction equations and input/output data. And it can avoid many intermediate operations for online solving Diophantine equations. Some simulations based on the actual run status data collected from sintering process of Anyang iron and steel plant are given to verify the effectiveness.