Title: Evolving the ensemble of predictors model for forecasting the daily average PM10
Authors: Krzysztof Siwek; Stanislaw Osowski; Mieczyslaw Sowinski
Addresses: Warsaw University of Technology, Institute of Theory of Electrical Engineering, Measurement and Information Systems Warsaw, Koszykowa 75, Poland. ' Warsaw University of Technology, Institute of Theory of Electrical Engineering, Measurement and Information Systems Warsaw, Koszykowa 75, Poland; Military University of Technology, Institute of Electronic Systems Warsaw, Kaliskiego 2, Poland ' National Centre for Nuclear Research, Andrzeja Soltana 7, Otwock-Swierk, Poland
Abstract: The paper develops the methods of accurately forecasting the daily average concentration of PM10. We apply the Support Vector Machine in the regression mode (SVR) as the main workhorse of prediction. Different approaches to the prediction are tried: the direct application of SVR, the combination of SVR and wavelet decomposition, and the Blind Source Separation (BSS) method for improving the final accuracy of prediction. The main novelty of the proposed approach is the application of the ensemble of predictors integrated using the BSS method. The numerical experiments of predicting the daily concentration of the PM10 pollution in Warsaw have shown good overall accuracy of prediction in terms of RMSE, MAE and MAPE errors, as well as correlation and index of agreement measures.
Keywords: PM10 forecasting; support vector machines; SVM; ensemble; predictor modelling; wavelet decomposition; blind source separation; BSS; environmental pollution; air pollution; air quality.
International Journal of Environment and Pollution, 2011 Vol.46 No.3/4, pp.199 - 215
Accepted: 19 Jan 2011
Published online: 30 Apr 2015 *