The Application Of Neural Dynamic Programming Based On BP Network In Boiler Combustion Control

Boiler combustion plant is a complex process system with the characteristics—multivariables, strong jamming and large lagging, meanwhile its internal and external disturbance are very frequent, so using conventional control scheme is hard to control the boiler combustion system.On the other hand, Neural Dynamic Programming is a method used to approximate optimal control over time in nonlinear systems. It is for working out the uncertainty problem created by non-linearity of plant or system modeling. Also, it is suitable for dealing with time-varying complex system and dynamic varying complex task.This paper expatiates the research and application of boiler combustion control and Neural Dynamic Programming, focuses on studying a class of Neural Dynamic Programming : Dual Heuristic Programming (DHP) based on BP network and Action-Dependent Heuristic Dynamic Programming (ADHDP) based on BP network. Neural Dynamic Programming is a method of interacting with the system (environment) and improving control effect. Dual heuristic programming (DHP) mainly consisting of three modules: model, critic and action, and action-dependent heuristic dynamic programming (ADHDP) mainly consisting of two modules: critic and action. This paper introduces the differences of these Neural Dynamic Programming methods in structure, evaluation and critic's output, meanwhile analyzes the principle of operation and control behavior of boiler combustion process, qualitatively considers the coupling relation of each variables in combustion process, takes boiler combustion basic assignment as object, then uses dual heuristic dynamic programming and action-dependent heuristic dynamic programming (ADHDP) in Neural Dynamic Programming to consider the solution of boiler combustion multivariable control system, realizes the emulational control of boiler combustion process, analyze learning capacity, controlling effect and adaptive capability of the approach. All of three states reach the expected control objectives. Finally, this paper summarizes the difficulty of this technical application. The design and realization about control algorithm in this paper have great significance for practical application in future.