Vol.2, No 5, 2005 pp. 485 - 492
UDC 341.226:62-73
THE MODELING OF AIR POLLUTION CONTROL DEVICES
USING NEURAL NETWORKS
Miomir Raos, Ljiljana Živković, Nenad Živković, Branimir Todorović
Faculty of Occupational Safety, University of Niš, Serbia & Montenegro
Abstract. The majority of methods for pollutant elimination assume the flow of the polluted gas through the pollution control system. The system is made of various devices which have to be chosen based on the characteristics of the pollutant: aerosol, solid particles, droplets or gaseous. The chosen framework and facilities depend on the type of the pollutant: aerosol, solid particles, droplets or gaseous. There are a number of basic parameters which have to be considered in order to define air pollution control devices. This study represents a modeling of the named parameters which are related to the framework and facilities of air pollution control. In order to set the optimal parameters of a purification device, a deterministic model of the process of purification should be determined. Such a model is often difficult to construct, since physical and chemical characteristics of the source of pollution are not completely known. In this paper we propose a black-box modeling tool based on the application of an artificial neural network.
Key Words: Air pollution control, modeling, neural network
MODELIRANJE PREČISTAČA GASOVA
PRIMENOM NEURONSKE MREŽE
Većina metoda za prečišćavanje gasova podrazumeva prolazak gasne struje kroz neki sistem za prečišćavanje. Takav sistem se uglavnom sastoji od ralzicitih uređaja za prečišćvanje. Izabrani sistem  i njegove karakteristike zavise o kakvom se zagađenju radi, dali su  to aerosoli, čestice, raspršene kapi  ili gasovi. Svakako da noseći gas,proces emisije i promene u izvoru zagađenja utiču na izbor sistema za prečišćavanje. Postoji veliki broj parametara koje treba razmotriti u procesu izbora sredstava i sistema kontrole, a ova studija predstavlja njihovo modeliranje. Osnovni zadatak je dobijanje modela nepoznate, vremenski promenljive nelinearne zavisnosti. Predložen je algoritam za sekvencijalnu adaptaciju mreže  radijalnih bazisnih funkcija (RBF). Adaptacija parametara i strukture je inkorporirana u sistem proširenog Kalmanovog filtra. Za vreme adaptacije strukture RBF mreže kombinovana su dva prilaza: izgradnja (rast) i uprošćenje.
Ključne reči: prečišćavanje vazduha, modeliranje, neuronska mereža.