Combustion state classification by character signals mining and support vector machines

A new method for diagnosing combustion stability of coal-fired utility boiler is proposed in this paper. Moving window approach was introduced to extract character signals that had high correlation with fuel flow from database. Three features of character signals were extracted including mean, standard deviation, peak-peak value, which were prepared for the input of SVM classifier. Then a nonlinear mapping relation model between character signals and combustion state by SVM was build. Operation plant data was used for modeling and validation. The results of simulation experimentation indicate that our method has 100% accuracy for stable combustion identification, so the proposed method is very effective to diagnose combustion state.