INVESTIGATING THE PERFORMANCE OF MACHINE LEARNING ALGORITHMS ON BUILDING ENERGY PREDICTION MODELS
Project no. 21169 · Status: Kapanmış · 2019
Summary
Energy has significant impacts on the economies of countries as it is directly linked to the growth and development of the sectors. However, energy scarcity stands out as an important problem due to the increase in energy consumption day by day. Especially since buildings constitute approximately 40% of total energy consumption, it is of great importance to predict energy consumption correctly.The aim of this study is to examine the performance of machine learning and deep learning techniques in predicting energy consumption in buildings. In this context, heating, cooling and ventilation energy consumption of a building was modeled using Multiple Linear Regression (MLR), Random Forest Regression (RFR), Support Vector Regression (SVR) and Artificial Neural Networks (ANN). Six independent variables including daily average outdoor air temperature, relative humidity, global radiation, sunshine ratio, daily degree - days and days of the week were used to develop the models.Prediction results were evaluated via RMSE, MSE, MAE, CV-RMSE and R2 performance criteria. The results show that the most successful model is ANN with performance values of 0.9028 R2 and 0.0855 RMSE, 0.0056 MSE, 0.0574 MAE, and 0.1865 CV-RMSE.In addition, the independent variables were rearranged via using the Wrapper method and daily average outdoor air temperature, relative humidity, number of degrees-days per day, and days of the week were used to develop energy models. The performance results of these models show that the most successful model is obtained by ANN with 0.848 R2 value. It can be concluded that ANN models are more successful in predicting building energy consumption than other methods.