Forecasting Electricity Demand in Greece: A Functional Data Approach in High Dimensional Hourly Time Series

This paper presents a novel approach to forecasting electricity demand in Greece by employing functional data analysis to analyze power consumption. Unlike traditional methods that predict a single future point, this approach forecasts a curve representing daily power consumption. This functional object enables electricity providers to better tailor their pricing and supply strategies in the upcoming day's megawatt hours (MWh) market. To achieve this, the ARIMA algorithm supplemented by functional principal components is utilized for one-step-ahead forecasting. The analysis also incorporates a functional regression using a functional linear model for functional responses, identifying patterns that link one day's consumption to the preceding day. The results demonstrate that the FDA-based method enhances the accuracy of short-term demand forecasts, outperforming classical time series algorithms and neural networks. By providing a more dynamic and comprehensive model of electricity demand, this approach offers significant implications for energy policy and management practices.