Experimental Machine Learning Study on CO2 Gas Dispersion

Machine learning (ML) is expending its application in many practical areas such as image recognition, natural language processing, games, etc. Simulated modeling of gas diffusion can be one of the applications. This experimental research was designed to know the potential of machine learning methods in modeling CO2 gas dispersion. Dispersion data of gases can be collected with sensing devices so that ML-based techniques can be applied to simulate the diffusion. In this study, three methods were explored and compared; linear interpolation, Multi-Layer Perceptron (MLP) and Deep Multi-Layer Perceptron (DLP). A set of experiments was conducted to collect dispersion data of CO2 gas. The experiments were executed in a wide room with two doors and eight windows that are enough to refresh the room air. Three sets of data were collected for learning and one set for testing. The Root Mean Square Deviation (RMSD) was applied to compare the three methods. The DLP method showed the lowest RMSD comparing with real test data, the linear interpolation the next and the MLP the last.CCS Concepts•Computing methodologies ~Machine learning•Applied computing~Environmental sciences