Designing K-nearest neighbors model for low voltage load forecasting

Until recently, low-voltage end-consumers were assumed to be uncontrollable, and load forecasting in the distribution systems was of limited interest. With the increasing share of decentralized supply connected to the distribution grid, balancing of supply and demand will have to be done locally, at the low-voltage level, for which short term load forecast becomes indispensable. At the same time, forecasting techniques based on standardized or individual load profiles, commonly used in the distribution grid, are inadequate for disaggregated loads. In this publication, we investigate K-Nearest Neighbors approach that we have proposed earlier for the low voltage load forecasting. We focus on various model design decisions such as parametrization scheme, the choice of a distance notion and combination function aiming to further improve the model. We use it to forecast the load of 600 consumers of different types showing how different model designs can substantially improve the forecast accuracy.