Predicting and Modeling of Daily Concentration of Particulate Matter (PM2.5 & PM10) in Hamadan Winter with Multilayer Perceptron Neural Network

In recent years, the number of days with high concentrations of particulate matter (PM) has been increased in Hamadan city. Since this phenomenon is more prevalent in winter, in this research, an optimal artificial neural network model has been proposed to predict the concentration of PM in winter. To investigate the concentration of winter PM in Hamedan, the winter data of air pollutants and meteorological parameters were analyzed with Pearson correlation. Then, according to the results, Multilayer Perceptron Artificial Neural Network (MLP-ANN) with an optimized structure based on the training and testing was used to predict the daily concentration of PM2.5 & PM10. Among meteorological and air quality factors, air quality variables were more correlated with winter PM concentration. MLP-ANN predicted PM2.5 in a model with 3 input layers, 1 hidden layer and 4 middle layer processors with R: 0.93 and also with 1 input layer, 1 hidden layer and 5 middle layer processors with R: 0.92. Predicted PM10. Alvand Mountains are located between the prevailing wind in the western parts of the Iran and Hamedan, therefore the effect of meteorological factors on PM concentration is low. Also, the increase of winter PM in the city can be due to increased fuel consumption in winter. The MLP-ANN tool with the lowest and most accessible data has the ability to predict PM early and can be used to control PM effects.