Long And Short Term Energy Demand Forecasting Using Xgboost Models

Published in: Prospective and trends in technology and skills for sustainable social development. Leveraging emerging technologies to construct the future: Proceedings of the 19th LACCEI International Multi-Conference for Engineering, Education and Technology
Date of Conference: July 19-23, 2021
Location of Conference: Virtual
Authors: Jose Robles (Universidad Nacional de Ingeniería, PE)
Freedy Sotelo-Valer (Universidad Nacional de Ingeniería, PE)
Johnny Nahui-Ortiz (Universidad Nacional de Ingeniería, PE)
Jorge Lopez-Cordova (Universidad Ricardo Palma, PE)
Full Paper: #219


As part of the technical studies in energy demand required by regulatory entities in Peru, this paper proposes the use of XGBoost Linear and Decision Trees models based on econometric long and short term variables for energy demand forecast. Considering that data of energy demand per year is only available since 1980, resulting in a small dataset, Leave-One-Out Cross Validation method was used in order to measure the performance of the models with unseen data. After training all models, in terms of econometrics, models based on long term variables shows to be more robust than models with short term ones. In addition, Decision Trees shows a better performance than Linear Models with a noticeable difference in the coefficient of determination for both training and test data.