Large-scale Circulating Fluidized Bed Modeling Based On Deep Belief Network

After more than 20 years of development,Circular Fluidized Bed Boiler(CFBB)technology has made great contributions to China's energy conservation and emission reduction.It is now an indispensable component of thermal power plants.However,its modeling and control problems have not been resolved.The CFBB system has complex properties such as large inertia,strong coupling and strong nonlinearity.The traditional mechanism modeling method contains a large number of empirical formulas and calculation approximations,which makes the established mathematical model inaccurate and directly affects the model operation and application in the field.Therefore,how to establish a circulating fluidized bed model accurately,simply and effectively has been a hot topic.This paper focuses on the large-scale circulating fluidized bed boiler system,and has an in-depth understanding of its process flow,structural equipment,modeling characteristics,and the main relationships between multiple parameters.Using the 330MW CFB on-site real-time operation data,the deep belief network data-driven modeling method was used to establish a coordinated model of the circulating fluidized bed boiler control system.The main contents are as follows:(1)The characteristics of the coordinated control system of the circulating fluidized bed boiler were analyzed.The main parameters were analyzed,and the established model was simplified.The coal volume,valve opening,primary air volume as input variable,output power,main steam pressure and bed temperature as output variable.Established a three-input-three-output model.Using the on site operational data of the power plant,data-driven modeling is performed using traditional neural network methods and deep belief network methods.By contrast,the deep belief network modeling method can overcome the shortcomings of traditional neural network training and achieve a better accuracy.(2)Based on the deep belief network modeling method,sparse regular representation and sparse connection strategy are introduced.The performance of the deep belief network modeling method is enhanced,and the accuracy of the modeling results is ensured,and the network training speed is also accelerated.Finally,the paper analyzes the shortcomings in the research and looks forward to the follow-up work.