Estimation of Concentration of Air Pollutants in Shazand Thermal Power Plant with Support Vector Machine Model Based on Selection of Effective Input Variables with Partial Mutual Information (PMI) Algorithm of Distribution of Air Pollutants

Document Type : ORIGINAL RESEARCH PAPER

Authors

1 Department of Environment, Islamic Azad University, North Tehran Branch, Iran

2 Department of Environment, Ahvaz Branch, Islamic Azad University, Ahvaz, Iran

3 Department of Water Resources Engineering, Ahvaz Branch, Islamic Azad University, Ahvaz, Iran

10.22034/ap.2021.1923264.1091

Abstract

Due to the difficulty of estimating the pollutant gas concentration in power plants, this study aimed to estimate the concentration of the air pollutants in a thermal power plant using the support vector machine model (SVM).
The concentration of environmental pollutants in the thermal power plant, Shazand, Iran, at different distances from the chimney was estimated using SVM. The effective input variables in the SVM model were selected using the Partial Mutual Information (PMI) algorithm. The modeling period was weekly from December 2018 to December 2019.
The PMI algorithm showed that the effective input variables for estimating the concentration of carbon monoxide (CO), carbon dioxide (CO2), sulfur dioxide (SO2), and nitrogen dioxide (NOX) pollutants at different distances are the same gas’ concentration at the power plant chimney. Among air pollutants, the maximum concentration is related to Co2 (2811.63 µg/m3), occurring at a distance of 5 km from the power plant chimney and the lowest concentration is related to Co (5.5 µg/m3, occurring at a distance of 20 km from the power plant chimney.
The polynomial kernel function is the best kernel function of the SVM model for estimating SO2 and NOX concentrations at different distances and the best kernel function in the SVM model for estimating CO2 and CO concentrations.
The SVM model has good accuracy and performance in estimating the pollutant concentrations, and selecting effective input variables in the SVM model with the PMI algorithm increases the model accuracy.

Keywords

Main Subjects