Modeling and Predicting PM2.5 Concentration at a Construction Site Using the Artificial Intelligence

Abstract

Introduction. High concentration of PM2.5 has the adverse effect on people's health. According to the evaluations made by the researchers, the impact of the particulate matter from the construction dust emissions resulted in 18% of deaths from respiratory diseases. Due to the growth of construction production volume and consequent increase of dust emission volumes, there arises the need to expand the scope of using the end-to-end technologies, namely the artificial intelligence technologies, for predicting the fine-dispersed dust particles PM2.5 concentration in dust emissions at the construction site.Materials and methods. To achieve this goal, the measurements of PM2.5 concentration at the construction site were carried out using the Handheld 3016 IAQ particle counter in the period from July 1 to July 6, 2023 taking into account the meteorological characteristics of the territory, which then became the input data for modelling the forecast of dust pollution concentration using such algorithms as ARIMA, EMA, XGBoost, etc., and the ensemble models that included the above machine learning algorithms. The efficiency of using these technologies for predicting was determined by comparing the results of the forecast and the field measurements data.Results. A correlation analysis was performed using the Modeltime program, which determined the relationship between PM2.5 concentration and meteorological variables. Autocorrelation was performed using Pearson correlation. At the first stage, four one-dimensional models based on the artificial intelligence were evaluated to determine the accuracy of mean concentration forecast. The next step was to evaluate the capacity of predicting the mean PM2.5 concentration using the multidimensional models that took into account the relationships between the independent and dependent variables. At the final stage of the research, three most efficient predictive models were included to test the ensemble model.Discussion and conclusion. The reliable predictive models can be the useful tools for understanding the concentration impact factors. In the present research, seven machine learning algorithms were used to predict the concentration of PM2.5. The research, as a whole, presents the evidences of the integrated modeling method efficiency for predicting the air pollution.