Energy Consumption Forecasting in an Office Building using Machine Learning Techniques

The electrical energy demand in building sector has been continuosly increasing. Machine learning techniques have become a great interest of society in predicting the energy demand of office buildings. The techniques are able to support building energy management by analyzing energy trends from simulation results. The research focused on studying various types of machine learning technique for predicting energy consumption in buildings. Data were collected from Thailand Post Company Limited, consisting five large buildings to predict energy demand and to analyze the efficiency of the techniques using MATLAB program. Results showed that the nonlinear autoregressive with external input with BR algorithm is the most accurate for predicting with R-square and root mean square error of 0.97 and 15.44 kWh, respectively. Moreover, simulated results of the electrical consumption before and after implementing energy-saving measures from 2024 to 2030 could reduce the units of electricity up to 27.11 percent and not lower than 16.55 prercent.