Research On Biomass Flame Combustion Characteristics Based On Flame Spectroscopy

In response to global climate change,protecting the ecological environment and alleviating the contradiction between energy supply and demand,biomass fuel plays an important role because of its low carbon,clean and renewable characteristics.Owing to its variety,wide sources and complex composition,it has a direct impact on the stability,economic benefits and safe operation.The flame is an intuitive reflection of the combustion state.The distribution of the flame intermediate--the flame radical is closely related to the type of fuel,combustion conditions,combustion efficiency and combustion emissions.In this paper,the combustion characteristics are analyzed and features are extracted by collecting the flame spectra and flame radical spectra of different biomass fuels,and then biomass fuel identification model has been established.The main content is as follows:Firstly,the flame spectra and radical spectra of four types of biomass fuel have been obtained by fiber-optic spectrometer based on the combustion experimental rig.Combined with industrial and elemental analysis of biomass fuels,qualitative analysis of flame spectrum,frequency and stability of biomass fuels are finished.Secondly,through spectral analysis of biomass fuel combustion flame,the spectra of the 200-1000nm band of flame and the combustion intermediate products-OH*(310.85nm),CN*(390.00nm),CH*(430.57nm)and C2*(515.23nm,545.59nm)are extracted as features.By using the tree model,including decision tree,random forest,extra tree and GBDT(Gradient Boosting Decision Tree,GBDT),the biomass fuel identification model has been established based on spectral analysis.Finally,combined with the characteristic engineering,this paper proposes a biomass fuel identification technology based on improved support vector machine.Based on the radiation intensity of biomass flame and flame radicals,through Feature Engineering and digging deeply of source data,a feature learning method,which consist of feature extraction,filter feature selection and dictionary learning,has been established to get premium characteristics reflecting biomass fuel.The radial basis kernel parameters and error penalty factors of support vector machines are optimized by using the improved grid search algorithm,and the biomass fuel identification model is constructed.Experimental results from the laboratory-scale combustion rig show the effectiveness of the two biomass fuel identification models.The characteristics of two models have been compared and analyzed.