Vinsamlegast notið þetta auðkenni þegar þið vitnið til verksins eða tengið í það: http://hdl.handle.net/1946/31402
Accurate modelling and forecasting of electricity load is important for many aspects in managing and maintaining a power system. Methods for short-term, medium-term and long-term forecasting include conventional time series approaches as well as machine learning techniques and powerful hybrid models. Next to projections of load aggregates, different stakeholders in the power market are interested in user type specific data and forecasts. Detailed consumer related data sets can be created using bottom-up or top-down methods, which require the input of behavioural and socio-demographic data or measured user type specific loads respectively. Data of this type is not available in Iceland, however, user type specific loads can roughly be approximated using a Monte Carlo simulation approach which samples values from seasonal ARIMA models. Those are based on representative user type curves extracted from a data set of hourly total load observations at 44 substations in the Icelandic system and their corresponding end user divisions (average load for the year 2015). This approach has several advantages over the benchmark method, user type specific curves based on the average end user divisions. It however fails to model higher resolution patterns so that the potential for better output quality lies in improvements of the sampling method.