Power Plant Boiler Air Preheater Hot Spots Detection System Based on Least Square Support Vector Machines

Abstract

Air preheater is the important heat exchanger in power plant units. Recombustion accident can be caused by inadequacy combustion of fuel or badly heat-dispersed condition aroused by low air or gas velocity after boiler outage. In the paper, discriminant models of 3 pairs of fire status have been built based on Least Square Support Vector Machines (LS-SVMs) for two kinds of kernel functions. Utilizing polynomial and RBF kernel, the hyperparameters of classifiers were tuned with Leave-one-out(LOO) cross-validation. Receiver Operating Characteristic(ROC) curve comparison shows that LS-SVMs classifiers are able to learn quite well from the raw data samples. Experiment results show that SVMs has good classification and generalization ability and RBF kernel function has more accurate than polynomial kernel function for this problem from the area under the ROC curve (AUC) values of two kernel functions.KeywordsSupport Vector MachineGeneralization AbilityFire DetectionStructure Risk MinimizationEmpirical Risk MinimizationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.