Forecast of electricity consumption using the support vectors machines

V magistrskem delu obravnavamo problem kratkoročnega napovedovanja združenega odjema (porabe) električne energije posamezne države. Združen ali agregiran odjem predstavlja eno od vrst napovedi odjema električne energije in se uporablja kot vhod pri napovedi cene električne energije na trgu na debelo. Na evropskih trgih je cena električne energije navadno enotna, zato nanjo vpliva združen odjem države. Napovedi se lotevamo z metodo podpornih vektorjev, ki je ena od metod strojnega učenja. Metoda podpornih vektorjev je v delu podrobneje opisana. Pri napovedi porabe električne energije imajo pomembno vlogo različni vplivni dejavniki, zato je del tega dela namenjen analizi njihovega vpliva. Analizirane vplivne dejavnike uporabimo pri napovedih porabe električne energije dveh držav. Za vsako državo naredimo dve napovedi z uporabo dveh različnih modelov oziroma dveh programskih knjižnic metode podpornih vektorjev. Ovrednotenje in primerjava napovedi obeh modelov ter obeh držav sta narejena z nekaterimi pogostokrat uporabljenimi statističnimi kazalniki. V magistrskem delu pokažemo, da z uporabo metode podpornih vektorjev lahko uspešno napovemo porabo električne energije in dobimo napoved, ki je glede točnosti primerljiva z ostalimi modeli.The Master’s thesis addresses a problem of short-term aggregated electric load forecasting for a whole country. Aggregated electric load or aggregated electricity consumption presents one type of electric load forecast and it can be used as an input for electricity price forecast on a wholesale market. Usually on European electricity markets electricity price is uniform, which means that it is affected by the aggregated load of the whole county. In this work a support vector machine (SVM) method is used to predict electricity consumption. Support vector machines are one of the machine learning tools and they are comprehensively described in this work. When forecasting electric load we have to consider a large number of influential variables, which have a major impact on electricity consumption. For this reason a part of this work is focused on the analysis of the influence that different variables have on electricity consumption. Analysed variables were used as inputs for the forecast of electricity consumption for two different countries. For each country two forecasts were made, using two different models or two different libraries for support vector machines. The evaluation and comparison of both models are made with some frequently used performance (error) metrics. The Master’s thesis shows that the us of the support vector machine method for electric load forecasting can lead to successful results, which is comparable to other models from the literature