Enhanced Evaluation on Electrical Load Forecasting Methodologies in Urban Distribution Networks

Load forecasting is becoming increasingly prominent due to the rising electrification of end-uses, especially in urban distribution systems. An accurate electrical demand forecasting is essential for ensuring efficient and reliable system operations, optimal grid planning, and proper exploitation of local flexibility services. Specifically, this paper presents four 'very short-term' forecasting algorithms, designed to predict the electrical consumption a few hours ahead: three linear models and one ARIMAX model. An extensive analysis of their performance has been conducted using the urban scenario of Milan as a reference, varying both the training and forecasting periods to identify the best-performing algorithm. To this purpose, forecasts generated by short-term algorithms have been assumed as a benchmark. The methods were trained on a dataset of electric demand data from the past three years of a district of the Milan metropolitan area, including exogenous variables, such as auto-regressors and persistence indicators. Among the proposed models, the ARIMAX model trained between 7 and 14 days demonstrated the highest accuracy, significantly reducing the prediction error of short-term forecasts.