Comparative Study of Forecasting Methods to Predict the Energy Demand for the Market of Colombia

Abstract

An important challenge for the energy sector worldwide is the selection of the best method for forecasting the energy demand of a country. This research addresses the Colombian electricity market, a developing country where power generation is predominantly hydroelectric. The goal is to compare commonly used and state-of-the-art methods to predict the general short-term demand. The selected methods are the long short-term memory (LSTM) network, the autoregressive integrated moving average (ARIMA) and Facebook Prophet. These methods were used to predict the hourly total electricity demand for Colombia for the next day (24 h). The cross-industry standard process for data mining methodology was partially used, excluding the deployment phase. The hourly time series considers data from 2022 to 2023. Models are evaluated using the root mean square error (RMSE) and the mean absolute percentage error (MAPE). Prophet had the best RMSE with 428.5 MW, followed by ARIMA (441.7 MW) and LSTM (541.3 MW). Prophet and ARIMA achieved the best MAPE (4.1%), outperforming LSTM (4.9%). Results shown that the hourly intraday energy demand of the Colombian market can be predicted accurately for both work and non-work days.