20.500.12556/DKUM-57593 Primerjava pristopov k napovedovanju porabe električne energije Comparison of approaches to forecasting electricity consumption Magistrsko delo obravnava primerjavo pristopov k napovedovanju porabe električne energije. Delo je razdeljeno na štiri dele. V prvem poglavju so predstavljeni osnovni statistični koncepti, ki so potrebni za razumevanje opravljenih analiz in primerjav, to so: slučajna spremenljivka, statistična populacija in vzorec, statistični test, srednje vrednosti, mere variabilnosti in časovne vrste. Pregledu matematičnega področja sledi poglavje, kjer so predstavljene tehnike modeliranja porabe električne energije. Predstavljene so mere kakovosti modelov ter pet metod modeliranja: večstopenjska linearna regresija, umetne nevronske mreže, metoda podpornih vektorjev, eksponentno glajenje in metoda glavnih komponent. V tretjem poglavju je povzet pregled uporabe taksonomij, izdelana je taksonomija napovedovanja v elektrogospodarstvu, predstavljen je pregled domenske literature in prikazana je uvrstitev pristopov iz člankov v izdelano taksonomijo. Glavni rezultat poglavja je graf primerjav tehnik modeliranja, ki smo jih zasledili v literaturi. V zadnjem poglavju so predstavljeni primeri modeliranja porabe električne energije. Najprej je predstavljena korelacijska analiza osnovnih in izpeljanih atributov s časovno vrsto porabe električne energije, nato so prikazani rezultati implementiranih modelov napovedovanja, njihove natančnosti in finančne učinkovitosti. This master thesis deals with the comparison of approaches to forecasting electricity consumption. The thesis is divided into four parts. In the first chapter, we present the basic statistical concepts, which are necessary for understanding the analyses and comparisons, these are: random variable, statistical population and statistical sample, statistical test, the mean values, measures of statistical dispersion, and time series. Review of mathematical fields is followed by a section, which presents the modelling techniques of electricity consumption. It presents the measures of quality of models and five modelling techniques: multiple linear regression, artificial neural networks, support vector machines, exponential smoothing and principal components analysis. In the third chapter, we review the use of taxonomies, design a taxonomy to predict the electricity consumption, present an overview of the domain literature and show the classification of the approaches set out of literature into our taxonomy. The main result of the chapter is a graph of comparisons of modelling techniques. In the last chapter, we present examples of modelling electricity consumption. First we make a correlation analysis of basic and derived attributes of a time series of electricity consumption, then we present five different implemented forecasting models, their qualities and financial efficiencies. napovedovanje korelacijska analiza linearna regresija podporni vektorji umetne nevronske mreže eksponentno izravnavanje analiza glavnih komponent taksonomija. Keywords: forecasting correlation analysis multiple linear regression support vector machine artificial neural networks exponential smoothing principal component analysis taxonomy. true true false [S. Šmigoc] Slovenski jezik Angleški jezik Magistrsko delo/naloga 2016-03-01 12:48:42 2016-03-29 16:42:37 2022-06-14 03:06:40 2021-05-16 03:23:06 2016 0 Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 519.233(043.2) 22050824 URN:SI:UM:DK:KXZTZUM1 MAG_Smigoc_Sabina_2016.pdf MAG_Smigoc_Sabina_2016.pdf 1 AB0A6B212EC5EBD8D3000EC05BDB80B2 7dae63a6f1a237e41c5c457f501ea84254d634769bb5a8137ce73a36fa11dc78 f06a758e-7c0b-11eb-bb7a-00155d0001ca https://dk.um.si/Dokument.php?lang=slv&id=86724 Fakulteta za naravoslovje in matematiko 0 0 0