City scale energy demand forecasting using machine learning based models

Abstract

The rapid change during the last decades in the activities of people and the global energy trends in the usage of daily equipment, led to massive shortage not just in the energy sector, but almost in all of the related sectors, especially in the developing countries. Consequently, predicting energy demand is a vital issue in order to fulfill the future strategy of any country. In this study, four machine learning algorithms have been deployed in order to forecast the short-term demand load for the city of Kirkuk based on hourly recorded real data. This data-driven study has taken in count the metrological weather data as a function of electrical energy demand. The stages of developing the models are discussed, and the forecasting models validated using 6 months of out-of-sample data. Among the four (Machine Learning (ML) models, two predictive algorithms (i.e. Artificial Neural Network (ANN) and Decision Trees (DTs)) showed better predictive capabilities achieving lower error rates.