Medium-term electricity load forecasting

Abstract

In the context of competitive power markets and liberalization, electricity can be bought and sold at market prices like any other commodity. However, unlike most other commodities, electricity cannot be stored for future use in massive quantities. The process of power generation, transmission, distribution and consumption usually happens at the same time. This unique characteristic makes electricity a complex commodity to handle. Additionally, seasonality has to be taken into account, at the daily, weekly and annual time scales. Moreover, there are many exogenous variables to be considered as key drivers of electricity demand patterns such as weather conditions, economic activity, regional market characteristics and working patterns. Thus, traditional risk-averse electric utilities have to deal with a substantially increasing amount of risk. Managing a company in an efficient manner involves yet more and more statistical analysis, as well as careful forecasting both electricity demand and prices. Deregulation has made forecasting a key necessity for all market agents to hedge their corresponding risk exposures. Traditionally, vertically-integrated utilities used shortterm load forecasts to ensure security of supply, and long-term load forecasts for future capacity investments. However, since competition was introduced, this is no longer the case, and the minimization of volumetric risk has never been of such importance as it is today. For this reason, load forecasting is gradually becoming the most important stage for utilities, system operators, retailers and other market participants both at the planning and operation&maintenance levels. Demand forecasting deals with hourly, daily, weekly and monthly values of the system load and peak system load. The forecasting of different horizons is important for different activities within a company. This distinction has typically leaded to a forecasting classification within time horizons. Although the thresholds that detach them differ within publications and authors, the common clusters are: short-term, mediumterm and long-term. Other publications may also include very-short-term (real time) and very-long-term load forecasting as well. This Master´s thesis addresses the problem in a comprehensive way. It first analyses the state of the art in energy forecasting, making a clear distinction between time horizons and main areas including price, load and renewable energy sources forecasting among others. Then, a main classification of different methods is done, in which both qualitative and quantitative approaches are reviewed. Within the latter group, both explanatory and time-series models are introduced. The section finishes by stating the main methods and methodologies used in electricity load forecasting. Then, the models are presented following the same structure. Fist, the main variables are analysed, then the model is developed and the mathematical equation computed. Finally, results are obtained and the model is assess taking into account different statistical measures such as the MAPE (accounting for the model’s error) and the R squared (which gives insight on how much of the data is being explained by the model). After comparing both models developed, a final section is devoted to the conclusions and future developments so improvement insights are given for further research on the topic.