Research On Combustion State Recognition Of Exhaust Combustion Chamber In SOFC System

Solid Oxide Fuel Cell(SOFC)power generation system is a new type of power generation device,which has the characteristics of high fuel adaptability,less pollution and low noise.Now,it is a hot research topic in the field of energy and materials.The exhaust gas combustion chamber(hereinafter referred to as the combustion chamber)is responsible for the system heat energy supply and exhaust gas treatment.Unstable combustion state will affect the system pressure balance and heat supply,resulting in damage to the combustion components and even system shutdown.Therefore,accurate identification of combustion chamber combustion state is the guarantee of efficient and stable operation of SOFC system.At present,there are few studies on SOFC combustion chamber state identification.Due to the complex combustion chamber environment,the traditional combustion chamber state recognition has the problems of less state classification,higher model complexity and low recognition accuracy.In this paper,the following studies are conducted to address this problem:(1)The flame image of SOFC combustion chamber has the characteristics of complex color,low contrast and fuzzy details,and the direct recognition of flame image is not effective.In order to improve the image quality and lay a foundation for the subsequent image recognition,this paper proposed an image enhancement method based on wavelet fusion.For the color complexity of the original image,a variational framework based on the human visual perception system was used to separate the color of the image.The method improved the hue preserving model,a detail retention model was proposed to adjust the image hue while maintaining the image structure.Wavelet fusion was used for feature extraction and feature fusion,which solves the problem of fuzzy edge information and obtains clear detail image.(2)In view of the problems of high complexity and low accuracy of traditional combustion chamber state recognition methods,this paper proposed a SOFC combustion chamber state recognition network based on attention mechanism and image feature pyramid.To reduce the complexity of the network,deep separable convolution was used to construct the network.The compression-excitation structure was improved by using two full connections with additional 1×1 convolutions,and combined with the attention mechanism,a hybrid attention structure was proposed to improve the feature extraction capability of the network.To enhance the multi-scale information exchange capability of the network,a multi-scale bidirectional fusion pyramid was proposed using bidirectional computation and multi scale fusion.Experiments show that the proposed SOFC combustion state recognition method achieved higher recognition accuracy with smaller model complexity.The proposed method can effectively identify the combustion state of SOFC combustion chamber.