Uporaba nevronskih mrež za napoved pretovora v Luki Koper, d.d.

Žumer, Tom

| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

First page > Show document

Show document

Title:	Uporaba nevronskih mrež za napoved pretovora v Luki Koper, d.d.
Authors:	ID Žumer, Tom (Author) ID Dragan, Dejan (Mentor) More about this mentor...
Files:	MAG_Zumer_Tom_2017.pdf (4,14 MB) MD5: 83A9637B1EE2AA55B76E5AA362E79652
Language:	Slovenian
Work type:	Master's thesis/paper
Organization:	FL - Faculty of Logistic
Abstract:	Uspešne odločitve podjetij med drugim temeljijo tudi na napovedih. Le-te morajo biti dobre in natančne, da lahko podjetja ohranjajo svojo konkurenčno prednost. Napovedi se danes izvajajo z naprednejšimi metodami, med katere spadajo tudi nevronske mreže. V magistrskem delu smo želeli ugotoviti, ali so umetne nevronske mreže primerne za napovedovanje pretovora v Luki Koper, d. d. Podatki pretovora so bili sestavljeni iz generalnega in tekočega tovora, zaradi česar smo razvili dva modela umetne nevronske mreže, in sicer model mreže časovne vrste generalnega tovora in model mreže časovne vrste tekočega tovora. Modela vsebujeta t. i. NARX (ang. nonlinear autoregressive network with exogenous inputs) arhitekturo nevronske mreže. Izdelavo modela smo razdelili v dva koraka. V prvem koraku smo naredili redukcijo makroekonomskih kazalnikov, ki so nam predstavljali eksogene vhode modela. Izvedli smo jo z metodo analize glavnih komponent v kombinaciji z Monte Carlo simulacijo ter multiplo linearno regresijo. Modelu umetne nevronske mreže generalnega tovora smo namenili deset spremenljivk, modelu za tekoči tovor pa smo namenili štiri spremenljivke. V drugem koraku smo razvili umetno nevronsko mrežo generalnega in tekočega tovora. Rezultati obeh modelov so bili zadovoljivi. Poleg solidnega prileganja ocenjenih in dejanskih podatkov pretovora sta modela izpolnila tudi vse kriterije za kakovost modela. Glede na dobljene rezultate obeh modelov menimo, da so umetne nevronske mreže primerne za napovedovanje pretovora v Luki Koper, d. d.
Keywords:	umetna nevronska mreža, analiza glavnih komponent, Monte Carlo simulacija, makroekonomski kazalniki, pretovor, napovedovanje, analiza časovnih vrst
Place of publishing:	Celje
Publisher:	[T. Žumer]
Year of publishing:	2017
PID:	20.500.12556/DKUM-65556
UDC:	004.42
COBISS.SI-ID:	512842813
NUK URN:	URN:SI:UM:DK:KK52YOTG
Publication date in DKUM:	05.06.2017
Views:	2142
Downloads:	193
Metadata:
Categories:	FL
:	Copy citation

Average score:	(0 votes)
Your score:	Voting is allowed only for logged in users.
Share:

Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Secondary language

Language:	English
Title:	Using neural networks to forcast throughput of port Koper
Abstract:	Successful business decisions are, among others, based on forecasts. They have to be good and precise to enable the company to maintain its competitive advantage. Forecasts are now conducted by advanced methods, which also include neural networks. In this thesis, we wanted to determine whether the artificial neural networks are suitable for forecasting the throughput of Luka Koper, d. d. Data throughput consisted of general and liquid cargo, which is why we developed two models of artificial neural networks, namely, time series neural network model of general cargo and time series neural network model of liquid cargo. Models contain so called NARX (Nonlinear Autoregressive Network with Exogenous Inputs) architecture of neural network. We divided the model elaboration in two steps. In the first step, we have made the data reduction of macroeconomic indicators, which we accounted them as exogenous inputs for our model. Data reduction was carried out by the principal component analysis method in combination with Monte Carlo simulation and multiple linear regression. For our neural network model of general cargo we allocated ten exogenous variables, while for neural network model of liquid cargo we allocated four variables. In the second step, we have developed an artificial neural network of general and liquid cargo. The results of both models were satisfactory. In addition to a solid fit of the estimated and actual data throughput, the models also meet all the criteria for the quality of the model. With respect to results obtained by our neural network models, we believe that the artificial neural networks are suitable for throughput forecasting of Luka Koper, d. d.
Keywords:	artificial neural network, principal component analysis, Monte Carlo simulation, macroeconomic indicators, throughput, forecasting, time series analysis

Comments

Leave comment

You must log in to leave a comment.

Comments (0)

0 - 0 / 0

There are no comments!

Back