A Hybrid Model Based on Symbolic Regression and Neural Networks for Electricity Load Forecasting

This paper proposes a hybrid model for electricity load forecasting. Symbolic regression is initially used to automatically create a regression model of the load. Then the explanatory variables and their transformations that have been selected in the model are used as input in an artificial neural network that is trained to predict the electricity load at the output. Therefore symbolic regression operates as a feature selection-creation method and forecasting is done by the artificial neural network. The proposed hybrid model has been successfully used in an electricity load forecasting competition.

I. Introduction

The continuous need for better forecasting models and the rapid development of new tools and fields like e.g. machine learning, have increased the interest in electricity load forecasting. As a result, numerous load forecasting competitions have been organized by both academia [1], [2] and the industry [3], [4] where various forecasting methods have been proposed and tested against each other.