We present a system for forecasting the changes in PM 10 and PM 2.5 particulate matter air pollution concentration. The system is based on immissions data from automatic measurement stations of the Voivodship (Regional) Inspectorate for Environmental Protection in Warsaw (Poland) and a numerical forecast of meteorological parameters from the Interdisciplinary Centre for Mathematical and Computational Modelling, Warsaw University. The concept of the program is based on various models based on artificial neural networks as well as a support vector machine working in regression mode. The approach uses wavelet decomposition and Blind Source Separation for better, more accurate forecasting. This universal system provides a tool for early warning of exceedance of daily maximum levels of PM10 and PM2.5and is dedicated to local authorities to evaluate the ecological efficiency of environmental recovery programs.
Keywords: air pollution, suspended dust, forecasting, particulate matter, artificial neural networks, ANNs, support vector machines, SVM, wavelet decomposition, blind signal separation, BSS, air quality, PM forecasting, early warning systems, local authorities, ecological efficiency, environmental recovery