Functional Principal Component Analysis as a Versatile Technique to Understand and Predict the Electric Consumption Patterns
Understanding and predicting the electric consumption patterns in the short-, mid- and long-term, at the distribution and transmission level, is a fundamental asset for smart grids infrastructure planning, dynamic network reconfiguration, dynamic energy pricing and savings, and thus energy efficiency. This work introduces the Functional Principal Component Analysis (FPCA) as a versatile method to both investigate and predict, at different level of spatial aggregation, the consumption patterns. The method was applied to a unique and sensitive dataset that includes electric consumption and contractual information of Milan metropolitan area. The decomposition of the load patterns into principal functions was found to be a powerful method to identify the physical and behavioral causes underlying the daily consumptions, given knowledge of exogenous variables such as calendar and meteorological data. The effectiveness of long-term predictions based on principal functions was proved on Milan's metropolitan area data and assessed on a publicly-available dataset.