Journal of Korean Institute of Architectural Sustainable Environment and Building Systems ISSN:1976-6483(Print) 2586-0666(Online) 한국건축친환경설비학회논문집

The optimal machine learning model depends on building types was selected by comparing and analyzing short term load forecasting (STLF) performance of primary school and commercial reference building based on 4 machine learning models such as ANN, SVM, CHAID, and, RF. The research consists of data collection-storage, data analysis, meteorological variables extraction, energy consumption forecasting and analysis on typical primary school and commercial building energy model. TMY (Typical Meteorological Year) of Incheon, Korea was applied and based on weather forecasting data provided by the KMA (Korea Meteorological Agency). In case of building energy consumption data, primary school and medium commercial reference building energy consumption data by on EIA's Commercial Buildings Energy Consumption Survey (CBECS) were used. Key weather variables were extracted for each machine learning model between the input variables and the output which is building energy consumption in 15 minutes interval. Finally, forecasting of energy consumption on different building types conducted a comparative analysis of the forecasting performance of building energy consumption based on 4 machine learning models using optimal input variables. The results shows ANN model outperforms other models with 5.44% of CV (RMSE) for 7 days school building energy forecasting trained 8 weeks prior data. Whereas, RF model performs better than the others with 10.96% of CV (RMSE). It may be concluded that the priority of variables which have impacts on energy consumption is important and the most suitable model for energy forecasting is different by the building types.