Time series decomposition approach for simulating electricity demand profile

Abstract

Electricity demand profiles of dwellings are mainly composed of various known (deterministic) and unknown (stochastic) processes. Effective data processing approaches (such as time series decomposition) are mainly used to simplify underlying patterns in the complex stochastic processes by fragmenting the different layers of hidden processes (referred as components of time series). This paper will demonstrate the performance of state-of-the-art STL (a Seasonal-Trend decomposition procedure based on Loess) techniques (Cleveland, Cleveland, McRae, & Terpenning, 1990), embedded within the framework of the HMM-GP model, in simulating dynamics of high-resolution electricity demand data. The method is applied to the case studies located in the Findhorn community.