A Comparison of Feature Selection Techniques for Neural Network Based Load Forecasting

Abstract

The performance of neural networks in load forecasting tasks is highly influenced by the selection of the inputs. This selection is either problem specific or is relied on the literature. The scope of the present study is to compare two features (i.e. inputs) selection methods based on metaheuristics. Thus, the inputs selection process is treated as an optimization process. The test case involves the day-ahead load forecasting of the aggregated load covered by a distribution substation in a large urban area. © 2019 IEEE.