Quantitative Characterization Of Coal Combustion Process

The correct evaluation of the combustion quality in the furnace is of guiding significance for the efficient operation of boiler.Flame visualization and characterization technology is one of the important tools to understand pulverized coal combustion in depth,aiming at providing guidance for combustion process.This paper studies and analyzes the existing methods of the flame monitoring and combustion diagnosis,then we proposed a method that to adopts the deep learning method to extract the feature of the flame images,then recognize and monitor the combustion status.We aim to realize the high accuraccy and effective monitoring of furnace combustion state,ensure the safety and economy of the thermal power unit operation.First of all,a new method based on deep learning is proposed to recognise the combustion state of the furnace——the convolution neural network(CNN).Through the end-to-end network,the feature extraction and classification are integrated into a framework.We carried out the combustion adjustment experiment in the 660 MW combustion boiler,to collect 14400 continuous flame images under different combustion states,of which the training data is 10800 and the testing is 3600.The results show that the method has great potential in practical application of power plant.Secondly,an unsupervised classification framework based on the convolutional auto-encoder(CAE),the principal component analysis(PCA),and the hidden Markov model(HMM)is proposed to monitor the combustion condition with the uniformly spaced flame images,which are collected from the furnace combustion monitoring system.First,CAE is adopted to extract the features from the flame images,which obtain the sparse representations in the images.Then,PCA is applied to project the feature vectors into the orthogonal space for robustness and computation efficiency.Finally,a HMM is built to calculate the corresponding optimal states by learning the temporal behaviors in the compressed representations.A coal combustion adjustment experiment was conducted in a 660 MW opposed-firing boiler,and the sequential 14,400 flame images with three different combustion states were obtained to evaluate the effectiveness of the proposed approach.We tested six different compression dimensions of the latent variable z in the CAE model and ensured that the appropriate compress parameter was 1024.The proposed framework is compared with five other methods:the CAE+Gaussian mixture model(GMM),CAE+Kmean.theCAE+fuzzyc-meanmethod,CAE+HMM,andthetraditionalhandcraft feature extraction method(TH)+HMM.The results show that the proposed framework has the highest classification accuracy(95.25%for the training samples and 97.36%for the testing samples)and has the best performance in recognizing the semi-stable state(85.67%for the training samples and 77.60%for the testing samples),indicating that the proposed framework is capable of identifying the combustion condition,changing when the combustion deteriorates as the coal feed rate falls.Then,we proposed a quantitative characterization method of pulverized coal flame image during comsbution process based on Variational autoencoder(VAE).By maximizing the lower bound of the log-likelihood function,the VAE model could describe the generation characteristics of input data better.On the basis of reconstruction of the input image better,the model obtain the latent variable z,obey to the certain probability distribution(such as gaussian distribution),so that the model enhances the expression ability of the model and has the generation ability.The method was verified on the flame images from a 660 MW thermal power unit.Using the AE model and VAE model to extract the feature of the flame images respectively,the results show that the characteristic quantity extracted from flame images by the VAE model can reflect the combustion condition well.Finally,based on the feature variables extracted by VAE in the previous chapter,a steady stete detection method of the combustion process on sliding window is proposed.Based on the fact that the extracted flame characteristics obey the normal distribution,the detection index is constructed and the student’s t-test is used to construct the steady-state judgment criterion to realize the steady-state detection of combustion process.In order to improve the steady-state detection rate,the EWMA filter method is introduced to improve the detection index,and the width of sliding window is reasonably selected.The experimental results show that the method based on sliding window has a good detection rate when the combustion is unstable,which further validates the effectiveness of the method.