Uso de algoritmos de machine learning para predicción de consumo de energía eléctrica.

Abstract

Abstract: In order to improve energy planning in the country, this paper analyzes the use of machine learning algorithms to predict electric energy consumption in Ecuador. As numerical prediction methods, experiments were carried out with simple and multiple linear regression, decision tree forests or Random Forest, and LSTM neural networks, and KDD was used as a methodological framework. This document shows the process of collecting, analyzing and processing electricity consumption data in the Ecuadorian market between 2014 and 2023. The prediction methods used are evaluated using metrics such as MSE, RMSE and R2, to determine the model that best fits the available data, demonstrating the importance of using machine learning algorithms in predicting energy consumption, optimizing the management of the electrical resource through timely planning of plans, projects and resources that avoid problems such as electrical rationing. The data set was obtained from the SISDAT portal of the Electricity Regulation and Control Agency (ARCONEL). The experimentation carried out concludes that the Random Forest model presents the best evaluation metrics for predicting electricity consumption. On a smaller scale, the results of the LSTM model present good results for predicting time series data, while the simple and multiple linear regression models, for the specific data set, present limited accuracy.