Low-voltage power demand forecasting using K-nearest neighbors approach

Demand response in the low-voltage domain has been ofter proposed to mitigate the volatility of the renewable energy supply. Therefore, an accurate demand forecast in this domain is indispensable to effectively manage balancing power. At the same time, load profile based forecasting techniques, such as standardized load profiles commonly used in the distribution grid, are inadequate for this purpose. In this article, we introduce a novel short-term forecasting model based on a K-nearest neighbors approach. Using historic smart meter data as the only input, it forecasts the load for the next day without any explicit knowledge of the consumer. Therefore, our model requires no manual setup while being parametrized automatically. Its accuracy is shown to be superior to individual load profile technique for various samples of low voltage end-consumers, and their aggregation of any group size. This makes the proposed model viable for wide-area application in the low voltage domain.