Research On Modeling Method Of Combustion Stability Based On Flame Image

Thermal power generation is the main form of electricity production in China. For large or ultra-large coal-fired boilers(supercritical units), the instable combustion in the furnace not only can lead to reducing of thermal efficiency of the boiler and produce large quantities of pollutants, but also can cause flame out or serious furnace safety accidents in extreme cases. Combustion flame is the direct reflection of the combustion stability. Therefore, to ensure safe and economic operation of the boiler, it is necessary to monitor the combustion stability of the furnace in real-time.In the issue of combustion stability determining, the flame image-based combustion stability determining method is a mainstream research direction at present. In this thesis, combustion parameters are extracted from furnace flame images based on image processing technology, and then be used for modeling method research of combustion stability. A method based on interval number theory and Multiple Attribute Decision Making and a model based on rough sets and adaptive fuzzy neural network for combustion stability determining issue are proposed.The first step of the proposed combustion stability determination method in this thesis is employing interval number theory to express combustion parameters. It is due to the existence of flame pulsation phenomenon. Because soft computing methods rely on samples and staffs working experience, which are difficult to obtain, the method based on interval number theory and Multiple Attribute Decision Making for combustion stability determining issue is proposed. In the method, a concept of Deviation Stability Degree is defined. In the procedure of extracting weight, a method for extracting samples is proposed, which can overcome the shortcoming of only dealing with small samples of Multiple Attribute Decision Making. Experiments show that the method based on interval number theory and Multiple Attribute Decision Making is feasible and has good results, and provides a new idea for the issue of determining furnace flame combustion stability.Combining the logical inference of fuzzy control and the advantages of neural network control, such as good learning ability, parallel computing, etc., the proposed combustion stability determination model makes use of the learning ability of adaptive neural network to train the initial FIS model to obtain the best model parameters. And then, the trained FIS model can be used to determinate combustion stability. Given that the initial FIS model has a great impact on the model training, and it is difficult to obtain its subordinating degree function and fuzzy rules of fuzzy inference, rough set theory and fuzzy clustering method are employed to extract fuzzy rules and obtain initial subordinating degree function, respectively. How to select the relevant parameters of the model is described in detail After comparing with adaptive fuzzy neural network and BP neural network, the model shows its superiority.