Building power demand forecasting using K-nearest neighbors model - initial approach

Abstract

Buildings, acting as flexible loads have been often proposed to mitigate the volatility of the renewable energy sources. However, an accurate building power demand forecast is indispensable to effectively manage the load flexibility. In this publication, we make an initial proposition for a universal short term load forecasting model for buildings, based on K-nearest neighbors approach. The proposed model is parametrized automatically, and provides a forecast using only historic building load measurements as an input. Therefore, it does not require any manual setup and we apply it on a large sample of simulated mixed-usage buildings of different size. Thereby, model accuracy is shown to be superior to the forecast obtained using individual load profiles created for each building.