FORECASTING ELECTRICITY CONSUMPTION IN THE STATE OF PIAUÍ: A MACHINE LEARNING-BASED APPROACH

Abstract

The electricity consumption in the state of Piauí shows significant variations over time, influenced by economic, climatic, and behavioral factors. In this context, forecasting methods become essential to support energy planning and guide strategic decisions made by utility companies and governmental agencies. This work applies Machine Learning techniques to predict the monthly electricity consumption in Piauí using real data from the period between 2020 and 2024. The developed model was based on the Random Forest algorithm, following preprocessing steps, creation of lagged variables, and separation of the data into training and testing sets. The forecast for the year 2024 showed satisfactory performance, achieving a Mean Absolute Percentage Error (MAPE) of 3.33%, demonstrating the model’s strong ability to capture both the growth trend and the seasonality of the time series. The variable importance analysis confirmed that the historical trend and the month of the year are the most determining factors for consumption in the state. The results demonstrate that the use of Machine Learning is an effective alternative for electrical load forecasting and can contribute to the state’s energy planning.