Research On The Stability Determination Method Of Flame Combustion Based On Images And Improved BP Algorithm

Keeping stable coal combustion is the most fundamental requirement for the electric power station safe operation.The combustion flame is the most direct reflection of the state of combustion stability.For the purpose of automatic monitoring of the boiler combustion stability and quantifying the degree of combustion stability,this thesis extracted combustion parameters from furnace flame images according to the digital image processing technology and built combustion parameter database,which provide testing and training samples for established models.This thesis proposed two neural network models.As for shortcomings of classical BP algorithm such as bad anti-jamming ability,low learning rate and easy plunging into local minimum,a new kind of improved BP algorithm was proposed with varying slope of activation function and dynamically adjusting different learning rates.Considering the deficiency of network training and insufficient learning rate,this thesis improved a transfer function of the hidden layer,invented a new composite error function and adopted a new method of dynamic adjustment of different learning rate to accelerate the convergence of classical BP algorithm,and to avoid plunging into the local minimum point.The sample group training method was adopted that the samples are divided into training samples and test samples for training and testing the established model.Experience has shown that the improved model not only has better fault-tolerance and mapping ability but also improves recognition rates and computing speed,which can meet the real-time requirement of stability determination.In addition,this thesis built interval number sample decision database based on multi-attribute decision method;for obtaining the membership function parameters and fuzzy rules in fuzzy inference,simplified sample decision database based on rough set,discretized conditional attributes of decision database,realized the reduction of attributes and their values,increased the reliability of sample parameters.Combining the logical inference of fuzzy control and the advantages of neural network control,such as good learning ability,parallel computing,etc.,this thesis built a T-S fuzzy neural network model for combustion diagnosis,and selected appropriate fuzzy partition number to define “5-4 Model” then build the determination model of flame combustion stability based on “5-4 Model”.Contrasting the simulation diagrams after training with the before,the results show that the model is feasible and has good results.At last,this thesis contrasted performance parameters of two neural network models,due to the accuracy performance of BP neural network model is better than “5-4 Model”,come to conclusion that the improved BP neural network model is more suitable for the research on the stability determination method of flame combustion.