Prediction of PM10 using Support Vector Regression

This paper describes the development of a support vector regression (SVR) model for the PM10 forecasting in Bangkok. Particulate matter (PM10) with aerodynamic diameter up to 10 mu m (PM10) is targeted because these small particles effects to people health and it constitutes the major concern for air quality of Bangkok. The models developed are used to establish the relationships of PM10 with meteorological variables including globe radiation, net radiation, air pressure, rainfall, relative humidity, temperature, wind direction and wind speed as well as the air quality concentrations of Carbon monoxide, Ozone, Nitrogen dioxide and Sulfur dioxide. The data sets examined in the current study were collected by monitoring station operated by Pollution Control Department of Thailand corresponding to PK concentrations for the years 2007-2009. In order to provide with an operational air quality forecasting module for PM10, Support Vector Regression.method is investigated and applied. The results of this research show the model that set with value C = 5,000, epsilon = 0.001 and sigma = 0.1 work out most precise in forecasting over other tested models. Based on the test forecasting data, the mean squared error (MSE) was 1.0588x10(-10), which means this model was very satisfactory. The model reports that support vector regression can be used in forecasting PM10 successfully.