prikaz prve stranice dokumenta LSTM model neuronske mreže
Access restricted to authenticated users
master's thesis
LSTM model neuronske mreže
Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2022. urn:nbn:hr:179:566601
Smoljanović, Josip
University of Split
Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture
Department of Electronics and Computing

Institutional repository: Repository of the Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture

Cite this document

Smoljanović, J. (2022). LSTM model neuronske mreže (Master's thesis). Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture. Retrieved from https://urn.nsk.hr/urn:nbn:hr:179:566601

Smoljanović, Josip. "LSTM model neuronske mreže." Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2022. https://urn.nsk.hr/urn:nbn:hr:179:566601

Smoljanović, Josip. "LSTM model neuronske mreže." Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, 2022. https://urn.nsk.hr/urn:nbn:hr:179:566601

Smoljanović, J. (2022). 'LSTM model neuronske mreže', Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, accessed 17 July 2024, https://urn.nsk.hr/urn:nbn:hr:179:566601

Smoljanović J. LSTM model neuronske mreže [Master's thesis]. Split: University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture; 2022 [cited 2024 July 17] Available at: https://urn.nsk.hr/urn:nbn:hr:179:566601

J. Smoljanović, "LSTM model neuronske mreže", Master's thesis, University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Split, 2022. Available at: https://urn.nsk.hr/urn:nbn:hr:179:566601

Metadata
TitleLSTM model neuronske mreže
Title (english)LSTM neural network model
AuthorJosip Smoljanović
MentorDamir Krstinić (mentor)
GranterUniversity of Split
Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture
(Department of Electronics and Computing)
Split
Defense date and country2022-09-01, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Basic Technical Sciences
Automatics
Abstract
U ovom radu istražena je mogućnost predviđanja broja novih slučajeva zaraze virusom SARS-CoV-2 na razini države korištenjem LSTM modela umjetne neuronske mreže. Zbog nesistematičnosti dostupnih podataka proveden je postupak filtriranja. U tu svrhu korišten je niskopropusni Butterworth filtar. S obzirom da se radi o problemu predviđanja vremenskog slijeda podataka filtrirani podaci podijeljeni su na sekvence sukcesivnih vrijednosti. Korištena su dva modela umjetne neuronske mreže. Prvi model upotrjebljen je za dnevna predviđanja (jedan dan unaprijed), a drugi za tjedna (jedan tjedan unaprijed). U oba slučaja radi se o sekvencijalnoj strukturi sastavljenoj od ulaznog, izlaznog i jendog skrivenog neuronskog sloja. Modeli neuronskih mreža razvijeni su pomoću Keras aplikacijskog sučelja. Hiperparametri modela podešeni su metodom „pokušaja i pogreške“. Za osvježavanje težinskih faktora nerurona mreže korišten je algoritam unatražne propagaije pogreške (engl. backpropagation). Pretreniranje modela onemogućeno je korištenjem tehnike ranog zaustavljanja (engl. early stopping). Kao funkcija cilja korištena je srednja kvadratna pogrešla (engl. MSE), a za njezinu optimizaciju korišten je Adam optimizacijski algoritam. Pomoću srednje apsolutne postotne pogreške (engl. MAPE) izračunato je odstupanje prediktivnih vrijednosti od stvarnih vrijednosti.
Abstract (english)
This thesis investigates the possibility of predicting the number of new cases of infection with the SARS-CoV-2 virus at the state level. For that purpose, the LSTM model of an artificial neural network was used. Due to the unsystematic nature of the available data, a filtering procedure was carried out. For this purpose, a low-pass Butterworth filter was used. Filtered data was divided into sequences of successive values because of the time series forecasting nature of the observed problem. Two artificial neural network models were used. The first model was used for daily forecasts (one day ahead), and the second for weekly forecasts (one week ahead). In both cases, neural networks had a sequential structure composed of only one hidden layer. Neural network models were developed using the Keras application interface. The hyperparameters of the model were adjusted by the "trial and error" method. The weights of the neural networks were updated by a backpropagation algorithm. Overtraining of the model was disabled using the early stopping technique. The mean square error (MSE) was chosen as the objective function, and the Adam optimization algorithm was used for its optimization. The deviation of the predictive values from the actual values was calculated by the mean absolute percentage error (MAPE) formula.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:179:566601
Study programmeTitle: Systems and Control
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka automatike i sustava (magistar/magistra inženjer/inženjerka automatike i sustava)
Type of resourceText
File originBorn digital
Access conditionsAccess restricted to authenticated users
Terms of useLicence In copyright icon
Created on2023-06-30 11:48:39