Abstract:
Air pollution monitoring and analysis, particularly for particulate matter with
a diameter of 2.5 microns or less (PM2.5), is crucial due to its significant impact
on public health worldwide. Existing air pollution data primarily focus on large-scale
areas such as cities, provinces, or entire countries, which may not provide sufficient
information on pollution levels in smaller areas or locations without air quality
monitoring stations. In this issue, the thesis was to develop a method for monitoring
air pollution at a microclimate level by establishing seven PM2.5 monitoring stations
within the research area. These stations are strategically placed within a 1-kilometer
radius of each other to capture pollution variations in microclimatic conditions.
The stations continuously collect PM2.5 concentration data for 24 hours over
two years, from December 2021 to December 2023. Additionally, meteorological
factors such as air temperature, humidity, wind speed, and wind direction were
recorded to study and analyze their influence on air pollution variations
in microclimate environments. The research findings indicate that air humidity
and wind speed significantly influence PM2.5 concentration variations in the study area.
Based on these findings, an air pollution monitoring approach was developed using
a heatmap visualization technique generated through the Inverse Distance Weighting (IDW) method. This method estimates PM2.5 levels in areas without monitoring stations and represents pollution levels using different color shades, illustrating the movement and variation of PM2.5 concentrations. Furthermore, PM2.5 data and meteorological
factors were used to develop an air pollution forecasting model tailored to different
weather conditions. A clustering approach was employed to categorize the meteorological conditions into 4 clusters based on temperature, humidity, and wind speed. The results demonstrate that forecasting models based on Long Short-Term Memory (LSTM), Artificial Neural Network (ANN), and Wavelet Neural Network (WNN) achieved high accuracy, exceeding 90% in predicting PM2.5 concentrations for each weather cluster. These models can effectively forecast and monitor PM2.5 variations influenced by microclimatic meteorological factors in the study area.