Study On Machine Vision Based Combustion Diagnosis For Larger Scale MSW Incineration

With the annual cleaning and transportation volume of the municipal solid waste(MSW)is increasing in China,the number of waste incineration power generation projects and the incineration disposal rate have been increasing year by year.And the furnace type is also become large scale.But China's waste composition is complex and changeable,and with the development of society,the heat value of the waste is gradually higher than the original design heat value of the incinerator,leading to many combustion problems.For example,deterioration of combustion lead to pollutants exceeding standards,partial burning resulting in uncompleted garbage burning,and higher heat burn off rate.At present,the combustion monitoring and diagnosis in the furnace is mainly carried out by thermocouples.The monitoring system has a narrow coverage and the measurement of parameters is lagging.Therefore,it is necessary to carry out corresponding research on the diagnosis method of the combustion in the furnace.In this paper,based on the 750 t/d waste incinerator flame image,the combustion diagnosis methods are studied.The characteristic parameters extracted from the image are used to quickly characterize the combustion state and evaluate it.Prediction of the main steam temperature in the future by artificial neural network.And realize the quick and intelligent diagnosis of the partial burning problem.Firstly,the key parameters of combustion image,such as pixel mean value,flame area rate,flame high temperature rate and flame front,are obtained by image processing technology,and the combustion deterioration is comprehensively evaluated based on the above parameters and factor analysis method.Then developed the K-neighbor algorithm model and convolutional neural network(CNN,Convolutional Neural Networks)model for partial burning state recognition,and compared their respective advantages and disadvantages.For the CNN model,changing the activation function of the convolution layer through K-mean clustering.It simplifies the model structure and improves the efficiency of feature extraction.Finally,the combustion image and the distributed control system operating parameters(temperature,air distribution,main steam parameters,etc.)are combined to develop a neural network for predicting the main steam temperature in the future,and find the image can improve the prediction accuracy.