Short-Term Deterministic Forecasting of Individual Household Electricity Consumption Using the Hungarian Algorithm

This work proposes a new approach for improving one day ahead point forecasting of stochastic individual household electricity consumption. The focus is tackling the double peak penalty effect and improving peak predictions. Each prediction is generated by comparing a household's energy usage of the seven days leading up to the target day with all seven day periods from all households in the dataset. The households with the closest consumption patterns are then used to create the forecast. The proposed method selects nearest neighbors in a similar manner as in the kNN algorithm. However, it utilizes the Hungarian algorithm to extend this approach to allow for comparisons between consumption values that occur at different times. A case study using an open dataset composed of electric consumption data from 100 Irish households demonstrates that this method improves performance of RMSE over kNN by up to 4.5% and 10.6% for persistence forecasting.