Neural Network Generalized Predictive Control For Boiler Combustion System

Industrial coal boiler is a kind of important dynamic device, and has significant state in the development of national economy. Combustion system for industrial boiler is a complicated time-varying and dynamic process, with obvious non-linearity, tight coupling, long time lags, and strong interference. So it's hard to achieve ideal control performance and higher rate of auto-control in operation with conventional control algorithms. Based on the analysis of the present situation and development trend of boiler control domestic, advanced control for industrial boiler was researched in this paper and it is proved that this method can get good regulating quality, which not only has value on academic study but also has economic benefits and social benefits.Two improved algorithm were proposed: neural network generalized predictive control based on LM optimizer and multi parallel network generalized predictive control based on jump predictive. The former built one step predictive model based on BP neural network and deduced multi-step predictive model through recursive calculation method. They provided jacobian matrix information of system to controller. LM algorithm was instead of traditional gradient descent tuning method to optimize the controller. LM improved algorithm made full use of jacobian information, speeded up the optimizing of controller, improved control performance. The latter built multi-step predictive model based on multi parallel networks. It's performance function was composed of jump predictive information, it got controller through optimizer. The improved algorithm reduced the quantity of networks, so it reduced the calculation. Simulation prove those the improved algorithms is valid and feasible.Improved algorithms were employed in boiler combustion control. Coal control system adopted feedforward-cascade structure , which used heatquantity demand as feed-forward input and temperature of combustion chamber as middle controlled variable; input wind control system adopted feedforward-feedback structure which used coal as feed-forward input; output wind control system adopted feedforward-feedback structure which used input wind as feed-forward input. at last, Simulation prove that the combustion control system is satisfactory.