Application Study Of Predictive Control To Power Plant

In this paper, a predictive controller based on identification model has been studied.Since accurate math model is hard to be built for power plant process, predictive controlis difficult to apply to the industry. In order to solve this problem, firstly a predictivecontroller based on neural network model is designed. Using the nonlinear mappingfaculty of neural network, model of the controlled plant is built and predictive controllerbased on the model gets optimal control rules by way of minimizing assessment function.Simulation for coordinated control system shows a good control performance anddisturbance attenuation. Secondly a predictive controller based on Model-on-Demand(MoD) model is designed. MoD builds local linear model at every sampling time and getoptimal model through looking for most suitable number of neighborhood data andweight. This method is hard to be trapped in local minimum point and improves theaccurate of modeling. Then predictive controller based on this model gets optimalcontrol regulation. Simulation on superheated temperature and coordinated controlsystem shows a good performance and robust.