Short-Term Load Forecasting for an individual building using Shape-based Clustering and Random Forest

Recently, ICT-based smart grid-related businesses have been increasing as distributed energy resources are expanding. The load forecasting is one of the key technologies for the efficient operation of the businesses, so many related studies have been published. However, since loads of individual buildings are more volatile than a large-scale load in general, forecasting individual consumers’ loads is much more challenging and only limited studies have been published. In this paper, we propose a hybrid method to forecast electricity loads of individual buildings in a day-ahead manner. Using DTW similarities in load profiles was calculated focusing on their shapes, and clustering is conducted for pattern recognition. We estimate the pattern for the next day by random forest, and combined historical loads with weather information to forecast the hourly load of the day. For performance evaluation, 1,065 days of building load data were tested. The clustering method in our study provided better quality clusters and the classification model outperformed the benchmark model. Also, the hybrid structure recorded high prediction accuracy compared to a single structure.22Nkc