Title
Kratkoročna prognoza potrošnje električne energije zasnovana na metodama veštačke inteligencije
Creator
Božić, Miloš M. 1982-
Copyright date
2014
Object Links
Select license
Autorstvo-Nekomercijalno-Deliti pod istim uslovima 3.0 Srbija (CC BY-NC-SA 3.0)
License description
Dozvoljavate umnožavanje, distribuciju i javno saopštavanje dela, i prerade, ako se navede ime autora na način odredjen od strane autora ili davaoca licence i ako se prerada distribuira pod istom ili sličnom licencom. Ova licenca ne dozvoljava komercijalnu upotrebu dela i prerada. Osnovni opis Licence: http://creativecommons.org/licenses/by-nc-sa/3.0/rs/deed.sr_LATN Sadržaj ugovora u celini: http://creativecommons.org/licenses/by-nc-sa/3.0/rs/legalcode.sr-Latn
Language
Serbian
Cobiss-ID
Theses Type
Doktorska disertacija
description
Datum odbrane: 30.03.2015.
Other responsibilities
mentor
Stajić, Zoran
član komisije
Stanković, Milena 1953-
član komisije
Rajaković, Nikola 1942-
član komisije
Korunović, Lidija 1971-
član komisije
Janić, Aleksandar
Academic Expertise
Prirodno-matematičke nauke
Academic Title
-
University
Univerzitet u Nišu
Faculty
Elektronski fakultet
Group
Katedra za računarstvo
Alternative title
Short-term load forecasting based on artificial intelligence methods
Publisher
[M. M. Božić]
Format
XIII, 166 listova
description
Biobibliografija: listovi 163-164
Abstract (en)
The topic of this dissertation is a short-term load forecasting using artificial intelligence methods. Three new models with least squares support vector machines for nonlinear regression are proposed.
First proposed model is a model with forecasting in two stages. This model use additioal feature, maximum daily load which is not known for day ahead. Forecating of maximum daily load is obtained in the first stage. This forecasted value is used in second stage, where forecasting of hourly load is done. Model with feature selection, using mutual information for selection criteria, is a second proposed model. This model tries to find an optimal feature set for a given problem. Forecasting model based on an incremental update scheme is a third proposed model. This model is based on the incremental update of the initial training set by adding new instances into it as soon as they become available and throwing out the old ones. Then the model is trained with new training set. By this approach the evolving nature of the load pattern is followed and the model performance is preserved and improved.
For models evaluation, the forecasting of hourly loads for one year is done. Electrical consumption data for the City of Niš, which have about 260000 habitans and average daily demand of 182 MW, is used for testing. Double sesonal ARIMA and Holt-Winters as representatives of clasical models and artificial neural networks, least squares support vector machines and relevance vector machines as representatives of artificial models, are used for models evaluation. For a measure of accuracy, mean absolute percentage error, symetrical mean absolute percentage error, square root mean error and absolute percentage error are used.
Obtained results show that the best model is model with incremental update scheme, followed by double sesonal ARIMA and artificial neural networks models. The worst results are obtained by relevance vector machines and double sesonal Holt-Winters models. It has been shown that the best model could be successfully used with the short-term load forecasting problem.
Authors Key words
kratkoroĉna prognoza potrošnje elektriĉne energije, veštaĉka inteligencija, mašinsko uĉenje, elektriĉno opterećenje, vremenske serije
Authors Key words
short-term electrical load forecasting, artificial intelligence, machine learning, electrical load, time series
Classification
621.317.38:005.52]:004.8(043.3)
Type
Elektronska teza
Abstract (en)
The topic of this dissertation is a short-term load forecasting using artificial intelligence methods. Three new models with least squares support vector machines for nonlinear regression are proposed.
First proposed model is a model with forecasting in two stages. This model use additioal feature, maximum daily load which is not known for day ahead. Forecating of maximum daily load is obtained in the first stage. This forecasted value is used in second stage, where forecasting of hourly load is done. Model with feature selection, using mutual information for selection criteria, is a second proposed model. This model tries to find an optimal feature set for a given problem. Forecasting model based on an incremental update scheme is a third proposed model. This model is based on the incremental update of the initial training set by adding new instances into it as soon as they become available and throwing out the old ones. Then the model is trained with new training set. By this approach the evolving nature of the load pattern is followed and the model performance is preserved and improved.
For models evaluation, the forecasting of hourly loads for one year is done. Electrical consumption data for the City of Niš, which have about 260000 habitans and average daily demand of 182 MW, is used for testing. Double sesonal ARIMA and Holt-Winters as representatives of clasical models and artificial neural networks, least squares support vector machines and relevance vector machines as representatives of artificial models, are used for models evaluation. For a measure of accuracy, mean absolute percentage error, symetrical mean absolute percentage error, square root mean error and absolute percentage error are used.
Obtained results show that the best model is model with incremental update scheme, followed by double sesonal ARIMA and artificial neural networks models. The worst results are obtained by relevance vector machines and double sesonal Holt-Winters models. It has been shown that the best model could be successfully used with the short-term load forecasting problem.
“Data exchange” service offers individual users metadata transfer in several different formats. Citation formats are offered for transfers in texts as for the transfer into internet pages. Citation formats include permanent links that guarantee access to cited sources. For use are commonly structured metadata schemes : Dublin Core xml and ETUB-MS xml, local adaptation of international ETD-MS scheme intended for use in academic documents.