REGRESSION MODELS FOR SMART HOME DATA ANALYSIS

Abstract

Listen

Translate article

The development of smart home systems using modern approaches based on statistical and machine learning involves processing data and establishing quantitative relationships between multiple measurements. Since most of the tasks in the context of smart home relate specifically to the problems of ensuring energy efficiency, optimizing energy consumption for heating and maintaining comfortable temperature conditions, a necessary stage of design is the development of mathematical or statistical models of thermal behavior that connect data on heating energy consumption, temperatures, humidity, etc. The paper presents the regression-correlation modeling of the smart home data with the aim of developing either predictive models and researching the relationships between corresponding data measurements. Two separate data groups were selected to conduct the research: sets of temperature and energy consumption measurements obtained within the REFIT Smart Homes project and temperature distributions measured in the laboratory of intelligent autonomous systems of the Faculty of Electronics and Computer Technologies of Ivan Franko Lviv National University the period from February 1, 2021 to September 1, 2021. The approach to the pre-processing of climatic parameters of the smart home, which involves the use of the STL-decomposition method, is considered and implemented. Development and research of the regression-correlation models were performed for several combinations of data: a) internal and external temperatures; b) gas consumption and temperatures on heating elements and c) internal and external temperatures and gas consumption used for heating. The developed predictive regression models can be used both for the implicit assessment of the heat-saving characteristics of the smart home and for the statistical analysis of the thermal behavior. For example, the values of the coefficients of determination, the angles of inclination of the regression lines implicitly determine the efficiency of heating process or the properties of the building to maintain temperatures, and can complement more complex models to optimize the functioning of the smart home. Key words: regression modeling, smart home, XGBoost, data mining.