ANN-based data mining for the detection of the most influential variables causing mismatch of super-heater outlet temperatures in a thermo-plant

Abstract

ANN-based data-mining techniques are introduced to detect the most influential variables causing the mismatch of super-heater outlet steam temperatures. The strategies are: (1) Rough set selection: the rough variables set most likely to have possible effects on the mismatch of the outlet steam temperatures is deduced from about 3000 variables available in the power system data-base, by correlation analysis.(2) Relation capturing with ANN: the variables in the rough set are used as the input of an ANN, with the samples being appropriately chosen to train the ANN. (3) Sensitivity calculation of each ANN input for each training sample. (4) Influential variables set extraction: the criterion is to derive the sub-set characterized by the outstanding variables with the largest average of absolute sensitivity values. The influential variable set thus obtained, not intuitively known prior to the investigation, is found to be consistent with the general understanding of the power-plant engineers.