A1309
Title: On periodic and Markov switching models for electric load time series of households
Authors: Leia Betting - TU Dortmund University (Germany)
Marco Jeschke - TU Dortmund University (Germany)
Roland Fried - TU Dortmund University (Germany) [presenting]
Abstract: Flexible Markov switching and periodic autoregressive models are investigated for modeling load time series of individual households. Such data exhibit different kinds of seasonalities but not with a fixed pattern, and there are rather stable periods interrupted by occasional abrupt shifts. Hidden Markov models describe such changes through unobservable switches in customer behaviour. The benefits obtained from fitting flexible Markov switching models and periodic autoregressive models are compared in this context. The latter model class offers simplicity and an explicit formulation in terms of innovations but raises questions of choosing a proper, possibly time-varying model order and robustness to abrupt changes. The modeling approaches are compared based on their ability to generate accurate forecasts and the detection of extraordinary events.
