A novel MDPSO-SVR hybrid model for feature selection in electricity consumption forecasting

TL;DR

This paper introduces a hybrid model combining modified discrete particle swarm optimization (MDPSO) with support vector regression (SVR) to enhance feature selection and improve electricity consumption forecasting accuracy.

Contribution

The study presents a novel MDPSO-SVR hybrid model that effectively selects features and outperforms existing models in electricity consumption prediction.

Findings

MDPSO-SVR outperforms other models on real-world datasets.

Feature selection with MDPSO improves prediction accuracy.

The hybrid model is a promising alternative for energy forecasting.

Abstract

Electricity consumption forecasting has vital importance for the energy planning of a country. Of the enabling machine learning models, support vector regression (SVR) has been widely used to set up forecasting models due to its superior generalization for unseen data. However, one key procedure for the predictive modeling is feature selection, which might hurt the prediction accuracy if improper features were selected. In this regard, a modified discrete particle swarm optimization (MDPSO) was employed for feature selection in this study, and then MDPSO-SVR hybrid mode was built to predict future electricity consumption. Compared with other well-established counterparts, MDPSO-SVR model consistently performs best in two real-world electricity consumption datasets, which indicates that MDPSO for feature selection can improve the prediction accuracy and the SVR equipped with the MDPSO can be a promised alternative for electricity consumption forecasting.