Machine Learning Algorithms Based for Demand-Side Energy Forecasting of an Office Building Loads

Nowadays, buildings are increasingly turning to photovoltaic energy to improve their use of natural resources. This creates remarkable energy consumption, which can manifest itself in many forms: electrical loads, lighting loads, air-conditioning loads, and many others. In our article, we have studied a database containing the energy consumption of an office built in 2015 in Berkeley, California. As this database has already been prepared for prediction purposes, we have worked on predicting load types as a function of the time parameter. In this paper we studied the performance of some selected artificial intelligence models to predict the different types of loads in the last quarter-hour of the day as a function of other times periods recorded. Several models were tested, and the aim was to distinguish the best model in terms of performance and accuracy. We treated each type of load separately in the prediction for both sides of the building, namely: various electrical loads north and south sides, lighting in the south wing of the building and the Heating Ventilation and Air Conditioning loads for both South and North Wings of the building. Taking results in account, the Random Forest (RF) could be considered as a relevant model for the presented dataset for all load types, with an R2_score of around 0.9859 and no less than 0.89. Thus, due to its notable ability to track intensive fluctuations.