Tecnologico de Monterrey, Monterrey 64849, Mexico

---

Abstract: Strategic and operative decisions in the electricity sector are mostly driven by forecasts. Electricity dispatch is one of the most important operative process in electricity sector. In electricity dispatch, short-term load forecasting (STLF) is a necessary tool to achieve an efficient decision-making process. The lack of accuracy of the STLF method causes an excessive or insufficient electric supply to the power grid. An excessive supply of electricity may cause damage to the electricity generation equipment, whereas insufficient electric supply may cause load cuts that affects directly to the electricity users. STLF methods can be trained to construct models in different levels of aggregation. In this paper, we study the effects of STLF training method in two levels of aggregation. This study is conducted in the ERCOT dataset, which is composed of 8 load zones, and the objective is to predict the load of ERCOT in a hourly manner. We train two different artificial intelligence methods in every zone: Artificial Neural Networks and Support Vector Machines. Using these methods, we compare two approaches: Total and Integrated. The integrated approach consists in training both artificial intelligence methods in every one of the load zones, so the final prediction is de addition of the prediction of every model. The total approach consists in training both artificial intelligence methods directly to the total demand of ERCOT. The results show that the integrated approach is in general more accurate than the Total approach.

---

Full Paper.pdf

Copyright © 2020 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.