As modern people spend most of their time indoors, it is important to manage the indoor environment and air quality from the perspective of environmental health. Various pollutants are present in the indoor environment, as observed in the outdoor envi...
As modern people spend most of their time indoors, it is important to manage the indoor environment and air quality from the perspective of environmental health. Various pollutants are present in the indoor environment, as observed in the outdoor environment. Among them, fine dust can cause respiratory diseases, cardiovascular diseases, and even death. Methods and technologies for monitoring the concentration of fine dust indoors have been developed for controlling indoor dust. However, it remains difficult to apply these methods to multiple and diverse indoor environments. Numerous models that can predict indoor dust concentrations using artificial intelligence were recently developed.
In this study, we developed an outdoor concentration-based indoor fine dust model for predicting the fine dust concentration in indoor environments using outdoor observation data and deep learning techniques. The concentration of fine dust in outdoor environments has been continuously observed and predicted by national and local administrative organizations.
Linear regression models were used to develop the indoor fine dust concentration prediction models based on outdoor measurements. Three types of models (simple linear regression model, ordinary least squares (OLS)-based multiple linear regression model, and stochastic gradient descent (SGD)-based multiple linear regression model) were developed and learned with differences according to the number of variables and weight estimation method using the linear regression models. The PM10, PM2.5, temperature, and relative humidity measured simultaneously indoors and outdoors in a house for one year were used for model learning. We also attempted to overcome the limitations of existing linear regression models in reflecting time series data and improving model performance by removing outliers from the collected data over time.
The OLS-based simple linear regression model and OLS- and SGD-based multiple linear regression model were developed based on the learning results of the indoor fine dust prediction model. For the learning results between models, the predictive performance of the multiple linear regression model was superior to that of the simple linear regression model. In contrast, based on the predictive values of the model according to the weight estimation method, the model to which the deep learning method was applied showed high predictive power in the low-concentration region.