Primerjava SVM, MLR in PCA metod pri napovedovanju fotovoltaične proizvodnje v Sloveniji

magistrsko delo

Anja Goričan (Author), Drago Bokal (Mentor), Matjaž Miklavčič (Co-mentor)

Abstract

Zaradi teženj po trajnostni in obnovljivi energiji se v elektroenergetski sistem priključuje vedno večji delež fotovoltaičnih virov elektrike. Stabilnost elektroenergetskega sistema je ena ključnih nalog, ki jo mora zagotavljati operater prenosnega omrežja. Ob vedno večjem deležu fotovoltaike v sistemu je to vedno težje zagotavljati, saj se vrednosti proizvodnje fotovoltaičnih virov elektrike spreminjajo nepredvidljivo. Deležniki elektroenergetskega sistema pa morajo vnaprej zagotoviti usklajenost porabe in proizvodnje električne energije. V ta namen pregledamo vpliv različnih metod na napovedovanje proizvodnje fotovoltaike na območju Slovenije. Ob vplivu napovedovalnih metod preverjamo tudi vpliv napovedanega in realnega vremena na proces modeliranja in napovedovanja. V prvem delu magistrskega dela pregledamo osnovne matematične pojme, ki jih potrebujemo za nadaljnjo teorijo o napovedovalnih metodah. Nato predstavimo matematične koncepte metod napovedovanja. V drugem delu se osredotočimo na prikaz rezultatov napovedovanja po napovedovalnih metodah in različnih vhodnih podatkih. Po pregledu rezultatov ugotovimo, da se na predstavljenem kontekstu napovedovanja najbolje obnese metoda podpornih vektorjev z radialnim jedrom. Upoštevati moramo tudi pred procesiranje podatkov, saj je pred napovedovanjem potrebno podatke preslikati z metodo glavnih komponent. Pomemben delež k izboljšanju napake pri napovedovanju prinese tudi uporaba dejanskih podatkov o vremenu, ki se uporabijo v procesu modeliranja.

Keywords

magistrska dela;fotovoltaika;napovedovanje fotovoltaične proizvodnje;metoda podpornih vektorjev;metoda glavnih komponent;multipla linearna regresija;

Data

Language:	Slovenian
Year of publishing:	2019
Typology:	2.09 - Master's Thesis
Organization:	UM FNM - Faculty of Natural Sciences and Mathematics
Publisher:	[A. Goričan]
UDC:	621.3:51(043.2)
COBISS:	24917512
Views:	829
Downloads:	91
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Comparfison of SVM, MLR and PCA methods in predicting photovoltaic
Secondary abstract:	Due to the desire for sustainable and renewable energy, an increasing number of solar power plants are installed in the electricity system. Stability of the electricity system is one of the key tasks that transmission system operators need to mantain. With increasing share of photovoltaics in the system, this is becoming increasingly difficult to provide, as solar plant production changes unpredictably. However, the stakeholders of the electricity system must ensure that electricity consumption and production are coordinated. To this end, we review the impact of different methods on the prediction of photovoltaic production in Slovenia. In addition to the influence of predictive methods, we also check the impact of forecasted and real weather data on the modeling and forecasting process. In the first part of the master's thesis, we review the basic mathematical concepts we need to further explain the theory of forecasting methods. Then we present their mathematical theory. In the second part, we focus on the presentation of results by forecasting methods on different input data. After reviewing the results, we find that in the presented forecasting context, the radial kernel with support vector machine yields best results. However, we also have to consider data preprocessing, since it is necessary to map the data using the principal component analysis before forecasting. The use of actual historical weather data in the modeling process also plays a significant role in improving the prediction error.
Secondary keywords:	master theses;photovoltaics;photovoltaic production forecasting;support vector machine;principal component analysis;multiple linear regression;
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za naravoslovje in matematiko, Oddelek za matematiko in računalništvo
Pages:	XIII, 86 str.
ID:	11206020