Extended Weighted Nearest Neighbor For Electricity Load Forecasting

We present EWNN, a new approach for forecasting the hourly electricity load profile for the next day, from a time series of previous electricity loads. EWNN extends the well-known and successful weighted nearest neighbor method WNN by operating at an hourly level and by incorporating feature selection. We evaluate EWNN using two years of electricity load data for Australia, Spain and Portugal. The results show that EWNN provides accurate predictions outperforming WNN on all datasets, and also outperforming two other advanced methods (pattern sequence similarity and iterative neural network) and three baselines used for comparison.