Otimização de previsões de demanda de energia elétrica: aplicação de índices meteorológicos compostos em métodos de previsão de carga

Abstract

This study aims to develop a method for optimizing electricity demand forecasts by exploring different approaches to representing meteorological variables in various load forecasting methods and their respective configurations. To achieve this goal, composite meteorological indices, weighted by demographic factors, are developed based on numerical weather simulations conducted using the WRF model. These indices are integrated into two distinct load forecasting methodologies: one based on traditional regression techniques and the other utilizing neural network methods. Through the integration of these indices along with multiple calibration and training approaches for the forecasting models, the aim is to explore optimization strategies for load estimates. To validate the proposed method, a case study was conducted in the Northeast subsystem of the National Interconnected System (SIN). The results indicate that it is feasible to optimize load forecasts in this region, with a substantial influence of wind on the predictions. Both models demonstrated satisfactory results. However, it was observed that using extensive load histories for model calibration leads to a tendency to underestimate the forecasts. In summary, the developed method enables load forecast optimization throughout all stages, from data acquisition to the incorporation of environmental variables and the configuration of different load forecasting models.