Load forecasting through functional clustering and ensemble learning

摘要

In this paper a load forecasting methodology for 2 days-ahead based on functional clustering and on ensemble learning is presented. Due to the longitudinal nature of the load diagrams, these are segmented using a functional clustering procedure to group together similar daily load curves concerning its phase and amplitude. Next, ensemble learning of extreme learning machine models, developed for several load curves groups, is made to fully integrate the advantages of all models and improve the accuracy of the final load forecasting. The quality of this methodology is illustrated with a real case study concerning load consumption patterns of clients with different economic activities from a Portuguese energy trading company. The forecasting results for 2 days-ahead are good for practical use, yielding a \(R^{2} = 0.967\).