Main content area

PM10 emission forecasting using artificial neural networks and genetic algorithm input variable optimization

Antanasijević, Davor Z., Pocajt, Viktor V., Povrenović, Dragan S., Ristić, Mirjana Đ., Perić-Grujić, Aleksandra A.
The Science of the total environment 2013 v.443 pp. 511-519
European Union, air pollution, algorithms, aluminum, copper, emissions, energy, environmental indicators, gross domestic product, iron, neural networks, paper, paperboard, prediction, regression analysis, steel, swine, wood
This paper describes the development of an artificial neural network (ANN) model for the forecasting of annual PM10 emissions at the national level, using widely available sustainability and economical/industrial parameters as inputs.The inputs for the model were selected and optimized using a genetic algorithm and the ANN was trained using the following variables: gross domestic product, gross inland energy consumption, incineration of wood, motorization rate, production of paper and paperboard, sawn wood production, production of refined copper, production of aluminum, production of pig iron and production of crude steel. The wide availability of the input parameters used in this model can overcome a lack of data and basic environmental indicators in many countries, which can prevent or seriously impede PM emission forecasting. The model was trained and validated with the data for 26 EU countries for the period from 1999 to 2006. PM10 emission data, collected through the Convention on Long-range Transboundary Air Pollution — CLRTAP and the EMEP Programme or as emission estimations by the Regional Air Pollution Information and Simulation (RAINS) model, were obtained from Eurostat.The ANN model has shown very good performance and demonstrated that the forecast of PM10 emission up to two years can be made successfully and accurately. The mean absolute error for two-year PM10 emission prediction was only 10%, which is more than three times better than the predictions obtained from the conventional multi-linear regression and principal component regression models that were trained and tested using the same datasets and input variables.