Identification of anomalous load profile for short term load forecasting

Short term load forecasting methods involve estimation of the model parameters. The estimation is done by using the historical data of load profiles. Therefore quality of the data is very crucial for the better estimation of the model parameters. In practice, many events occur which degrade the quality of data. These events include natural and man-made calamities, network outages, trade strikes, general elections, important sporting events etc. These events impact the load profile in an irregular manner. Inclusion of these events´ data may contaminate the forecast. Anomaly of data could be seen due to significant shift or due to some spikes in the load profile. These anomalous load profiles should be detected and their use should be avoided in estimation process. In this paper, three approaches to identify the anomalous load profiles are proposed. The approaches are based on i) the notion of vector norm ii) probability distribution function and iii) hybrid of the two approaches. The approaches are tested with actual load data of an urban electrical distribution utility.