magistrsko delo
Taja Debeljak (Author), Jaka Smrekar (Mentor), Blaž Krese (Co-mentor)

Abstract

Dereguliran trg električne energije in vedno večje povpraševanje po električni energiji povečujeta potrebo po modelih za napovedovanje odjema. Natančne napovedi udeležencem na trgu pomagajo pri načrtovanju, upravljanju in nadzoru sistemov električne energije. V magistrskem delu se ukvarjamo s problemom napovedovanja preostalega odjema električne energije. Med preostali odjem sodi odjem pri vseh odjemalcih, katerih priključna moč je manjša od 43 kW. Za napovedovanje se pogosto uporabljajo regresijske metode, ki večinoma dajejo zadovoljive rezultate. V našem primeru pa smo se problema napovedovanja lotili z metodo Gaussovih procesov, ki temelji na Bayesovi statistiki. Model Gaussovih procesov je v magistrskem delu predstavljen z dvema pristopoma, to sta pristop v prostoru funkcij in pristop v uteženem prostoru. Rezultat Gaussovega procesa pri novi vhodni točki je normalna verjetnostna porazdelitev določena s povprečjem in varianco. Povprečje predstavlja najverjetnejšo izhodno vrednost, varianca pa predstavlja informacijo o zaupanju v napovedano vrednost izhoda. Na količino odjema pomembno vplivajo različne vplivne spremenljivke, ki jih v delu analiziramo in preverimo katere je smiselno vključiti v model. Za implementacijo modela smo uporabili Pythonovo knjižnico za strojno učenje Scikit-Learn. Model smo preizkusili na podatkih preostalega odjema na območju distribucijskega omrežja Elektro Ljubljana, d.d.

Keywords

napoved preostalega odjema električne energije;Gaussovi procesi;Bayesovo sklepanje;strojno učenje;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FMF - Faculty of Mathematics and Physics
Publisher: [T. Debeljak]
UDC: 519.21
COBISS: 57711107 Link will open in a new window
Views: 1380
Downloads: 135
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Electricity load forecasting using Gaussian processes
Secondary abstract: The deregulated electricity market and growing electricity demand are increasing the need for electrical load forecasting models. Accurate forecasts help market participants to plan, manage and control electricity systems. In the master's thesis we deal with the problem of forecasting the electricity consumption of all customers with a connection power of less than 43 kW. This includes households and small businesses. Electricity consumption will be predicted using Gaussian process model which is based on Bayesian statistics. We introduce two views to interpret Gaussian process regression models; function space view and weight space view. The result of Gaussian process is the normal probability distribution determined by mean and variance. The mean represents the most probable output value and the variance gives us information about confidence in the predicted output value. An energy demand is driven by many variables. A study of input variables was made to check which variables should be included in our model. We implemented this model using the Python's machine learning library, i.e. Scikit-Learn. Model was tested on the electrical load from a distribution network Elektro Ljubljana, d.d.
Secondary keywords: load forecasting;Gaussian processes;Bayesian inference;machine learning;
Type (COBISS): Master's thesis/paper
Study programme: 0
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za matematiko in fiziko, Oddelek za matematiko, Finančna matematika - 2. stopnja
Pages: V, 59 str.
ID: 12712803
Recommended works: