Neural Network Based Nonlinear Model Predictive Control

Model predictive control has achieved much improvement both inwidth and depth since it was first brought up in 1970s. However, with theindustrial processes being more complicated, traditional lineal modelpredictive control methods cannot meet nowadays' industrial controlrequirements much more. Nonlinear prediction model has become one ofthe focuses of control theory research. Nonlinear model predictive controlnot only inherits the advantages of traditional predictive control, but alsogets rid of the bind of linear mathematical model, which emphasizes themodel's prediction capability.Neural network, firstly came up in 1940s, has been widely applied invarious control strategies since it can converge to nonlinear relationsclosely, learn and adapt to dynamic system characteristics as well. As aresult, neural network will bring fresh blood to predictive control becauseof its nonlinear mapping characteristics. In this paper, a sort of neural network based nonlinear model predictive control strategy is put forward.The process is modeled by neural network, and the manipulated variable isalso calculated through neural network based controller.The proposed nonlinear model prediction control strategy is executedin a DCS to control a pilot tank level, with satisfying control resultsobtained. Meanwhile, this method is also applied to the Tennessee Eastrrianproblem and is proved to be successful.