Determination of features for air pollution forecasting models

The Sostanj coal fired Thermal Power Plant (STPP) is a major regional source of SO/sub 2/. To monitor air pollution around the STPP, a modern environmental information system was built, which measures ambient concentrations of SO/sub 2/ and meteorological parameters. All data are collected at half hour intervals. Harmful air pollution and environmental legislation are the major reasons why the STPP needs to have a reliable short term air pollution forecasting model. Such a short term air pollution model should forecast half hour average concentrations of SO/sub 2/ at the measuring points for up to 2 hours in advance. One of the most important tasks in building such a model is determining the features. This work presents several algorithms for feature reduction, all of them based on the case of SO/sub 2/ forecasting at particular measuring stations of the Environmental Information System of the STPP.