Research On Combustion State Detection Of Gas Boiler Based On Image Processing

As an environmentally friendly,clean and low-cost fuel,natural gas is widely used in central heating equipment in various regions.In the process of heating equipment operation,it is necessary to monitor the combustion state inside the gas boiler,so as to dynamically adjust the oxygen intake and natural gas consumption,so as to ensure the normal operation of equipment,improve the efficiency of production and the stability of the system.In order to effectively monitor the gas boiler,this paper focuses on the combustion state of the boiler.At present,the general flow of boiler flame detection algorithm is to extract flame features first and then conduct algorithm modeling,which is an effective combination of image processing,machine learning and deep learning technology.However,the current detection algorithm of boiler flame combustion state still has some aspects that can be improved,such as image segmentation is not accurate enough,the algorithm is traditional and the recognition accuracy is low.Based on the flow of flame detection algorithm,the research in this paper is mainly divided into the following three parts: flame image pretreatment,flame feature extraction,flame combustion state detection algorithm.The specific research contents are as follows:（1）Image preprocessing based on the gas boiler flame shot by CCD camera,including image denoising,image segmentation and image cavity filling steps.Aiming at the poor segmentation effect of traditional segmentation algorithm caused by uneven flame illumination in boiler,the threshold segmentation method of S-component based on HSV space was proposed to improve the image segmentation effect,so as to accurately extract the real parameters of flame.（2）In the stage of feature extraction of flame images,traditional feature extraction methods usually only extract flame parameters from a single frame image.Since the continuous flame combustion process includes the transition stage from ignition to stable combustion and from stable combustion to extinction,the real-time flame features of these two stages have many similar characteristics,so the accuracy of traditional feature extraction method for flame status identification in the transition stage is low.In this paper,flame parameters are extracted based on real-time features and historical features.Real-time features reflect the state information of the flame at the current moment,and historical features can reflect the state change information of the flame at the previous moment,which can effectively improve the accuracy of flame burning state identification algorithm.（3）Considering that there are few flame parameters extracted based on manual experience,this paper adopts tree model architecture to carry out high-order feature combination for flame parameters and complete feature expansion.The previous research work is mainly based on the combined framework of gradient lifting decision tree and logistic regression（GBDT+LR）for algorithm modeling to enhance the ability of feature expression.Considering that boiler flame combustion is a continuous state evolution process,this paper based on one-dimensional convolutional neural network and long and short-term memory network（1DCNN-LSTM）deep learning model framework to extract higher-order features of the flame,and at the same time to effectively learn the boiler flame timing information.Through experimental verification,the accuracy of GBDT+LR and 1DCNN-LSTM algorithms adopted in this paper reaches 93% and 94% respectively.Compared with support vector Machine（SVM）and Deep neural network（DNN）,the two algorithms adopted in this paper can achieve better results in various indicators and are more stable.