Analysis of Flue Gas Emission Data from Fluidized Bed Combustion Using Self-Organizing Maps

Efficient combustion of f uels w i th lower emissions le vels h as become a demanding task in moder n power p lants, and n ew to olsare n eeded to diag n ose t heir energ y production. T he goals of t he study were to find dependencies between pro cess var i ables andthe concent r a t i ons of g aseous emission comp onents and to c reate mult ivar i ate n onlinear mo dels descr i bing their for mat i on in theprocess. First, a gener ic process m odel was created by using a s elf-organizing map, which was clustere d w ith t he k-means algor ithmto create subsets representing the d ifferent s tates of t he process. Char acter i stically, t hese process s tates m ay inclu de hig h- and l ow-load situations and t r ansition states w here the load is increased or decreased. T hen emission mo dels were const r uc te d for b o th theentire process and for t he process s tate of hig h b oiler load. The m ain conclusion i s t hat t he methodolog y used is able to re veal suchphenomena t hat o ccur w i thin the pro ce ss states and t hat could other w ise b e d ifficult to obser ve.