Abstract :
In load forecasting applications special focus is placed on the demand patterns of the anomalous days. These days refer to public holidays, some weekdays between holidays, days with social events and generally, days characterized by atypical demand behavior. Anomalous days can be either excluded from the training and test sets or included by adding an identification indicator in the input of the forecasting model. In this work, through the load profiling methodology, the public holidays of the training set are clustered based on the similarities of the load curves shapes. The cluster label of the anomalous days is entered in the forecaster. Experimental results highlight the robustness of this approach applied in Short-Term Load Forecasting (STFL) of a sub-urban area bus of the Greek electric distribution system.