MATLAB Implementation of Electricity Demand Forecasting Using Double and Triple Seasonal Exponential Smoothing

Abstract

Control and Scheduling of the electricity demand in power supply systems using time series forecasting is nowadays a powerful methodology used worldwide in all power distribution systems. The main reason why it is so important is very simple: The electricity cannot be saved in big quantities, therefore the production and the consumption must match precisely, in order not to waste energy and save costs. Time series forecasting is a very powerful tool for power supply systems, and it is used worldwide by most of the system regulators or network distributors to predict precisely the electricity demand. These series use to show more than one seasonal pattern, hence double seasonal exponential smoothing has become the best solution for making forecasts for such kind of time series. Despite of this importance, there isn't nowadays any software that deploys this seasonal model. The regulators are demanding better tools in forecasting that capture this multiple seasonal pattern, and the later works on double and triple seasonal exponential smoothing seems to be a feasible solution. This project concentrates on a MATLAB implementation of Taylor's double seasonal exponential smoothing model, and explains its fundamentals and how it works. Later, it uses an hourly recorded time series of electricity demand in Spain to draw conclusions about the advantages of using this newer model for predictions.