Machine Learning Models for the Prediction of Electricity Generation From Crude Oil and Diesel in Ecuador

This study presents a predictive model for electricity generation at a petroleum station in Ecuador, using crude oil and diesel as primary energy sources. The objective is to ensure the energy self-sufficiency of the station, which is located in a remote area. To achieve this, tools based on time series and machine learning were employed. Historical data on fuel consumption and electricity generation from 2019 to 2023 were used to train prediction models aimed at anticipating energy demands and optimizing the management of resources involved in this process. To forecast the input variables of the predictive model, ARIMA time series were applied. The predictive models implemented were based on the Decision Tree algorithm and proved successful, as the prediction obtained was compared with actual electric power generation measurements and yielded very low errors. For example, a Mean Square Error of 0.09 MW, a Mean Absolute Error of 0.24 MW, and a coefficient of determination of 0.99 were obtained. As a conclusion, the model developed in this research, which is a combination of ARIMA with decision trees, demonstrates high feasibility for operational implementation at the plant.