In this paper, we conducted a numerical experiment on office buildings as a study of the application of machine learning to the prediction of power demand. We evaluated the usefulness of machine learning in power demand prediction by comparing the prediction results of two types of machine learning methods, neural network and random forest regression, with the prediction method using linear approximation so far. As a result, we clarified the practical use of machine learning from the evaluation of prediction errors by time and month, and we propose a new prediction method that hybridizes linear approximation and machine learning.．

View full abstract