Relative Entropy-Based Density Selection in Particle Filtering for Load Demand Forecast

Abstract

Particle filtering (PF) is an eminent simulation-based state estimation technique, which is capable of handling massive data sets and diverse external factors. Here, an effective selection of the importance density plays a pivotal role in performances of PFs by not only preventing degeneracy problems at early stages of process, but also taking both the transition prior and likelihood into account when the likelihood appears in the tail of the prior. To this end, we propose a novel importance density selection scheme for PF based on the minimum relative entropy principle. Theoretical derivation of the proposed scheme is presented, and its effectiveness is evaluated against various sampling schemes that exist in the literature using synthetic experiments. Finally, the performance of the proposed minimum relative entropy-based density selection scheme is successfully demonstrated for short-term electricity demand forecasting of a company located in Miami, FL, USA.