Pronóstico de la Demanda de Energía Basado en Heurística con Datos Escasos en el Departamento del Huila, Colombia

Within the framework of the energy transition, electrical distribution grid operators require effective tools to predict the demand of individual users. These tools are necessary for an adequate planning of future generation resources and infrastructure modernization. However, understanding future electricity needs poses a significant challenge, especially in emerging economies, where historical data are manually collected on a monthly or bi-monthly basis and exhibit a significant amount of missing information. In response to the above, this work proposes a novel heuristics-based method for medium-term energy demand forecasting with scarce data. Qualitative and quantitative information was abstracted into a mathematical model representing the trend and noise components of historical energy consumption observations. In addition, external factors were considered as an additional layer for the mathematical model, in order to account for events that could not be foreseen by merely using the dataset. A train-test data split was proposed to iteratively search for the best parameters to predict electricity demand with respect to different categorical divisions of users (social stratum, rural or urban location, and municipality). For testing and validation, real historical data were used, as provided by the regional distribution system operator (DSO) of the department of Huila, Colomba. The results suggest a trade-off between accuracy and computational intensity, as well as the fact that a non-categorical approach leads to the algorithm with the best performance (average absolute error of 1.61%) at a low computational cost.