Prilog istraživanju analitičkih metoda za proračun tehničko-eksploatacionih pokazatelja brodova dunavske plovne mreže u cilju poboljšanja plovidbe

Abstract

Predmet istraživanja ovog doktorskog rada je snaga na propelerskim vratilima brodova-potiskivaca. Istraživanje se zasniva na izvršenim eksperimentalnim ispitivanjima brodova-potiskivaca koji su bili u potiskivanim sklopovima tokom ispitivanja. Eksperimentalna ispitivanja brodova-potiskivaca su obavljena na dunavskoj plovnoj mreži. Rezultati eksperimentalnih ispitivanja brodova-potiskivaca pokazali su da se dobijaju prihvatljive vrednosti za snage na propelerskim vratilima brodovapotiskivaca. Istraživanjem razlicitih analitickih metoda zakljuceno je da se ne mogu dobiti dovoljno precizne vrednosti snage na propelerskim vratilima brodova-potiskivaca. Zbog toga je razvijen matematicki model za proracun snage na propelerskim vratilima brodova-potiskivaca kao izlazne vrednosti iz modela. Matematickim modelom omoguceno je modeliranje preslikavanja vrednosti ulaznih u vrednosti izlazne promenljive (snage na propelerskim vratilima brodova-potiskivaca). Predložen je model jedne veštacke neuronske mreže za proracun snaga na propelerskim vratilima brodovapotiskivaca. Kako ni ova metodologija nije dala rezultate ocekivane tacnosti, usvojen je model skupa veštackih neuronskih mreža (“Ensemble Neural Networks“). Veštacke neuronske mreže su odabrane zbog tacnosti i jednostavnosti modeliranja preslikavanja vrednosti ulaznih u vrednosti izlazne promenljive. Po odredivanju metodologije za modeliranje preslikavanja vrednosti ulaznih u vrednosti izlazne promenljive za ulazne velicine uzete su brzine plovidbe potiskivanih sklopova u mirnoj vodi, glavne dimenzije ispitivanih potiskivanih sklopova i deplasman potiskivanih sklopova. Odredena je i jedna izlazna promenljiva i ona je snaga na propelerskim vratilima brodova-potiskivaca. Glavni cilj disertacije je da se odredi funkcionalna veza izmedu pomenutih ulaznih i jedne izlazne promenljive cime se zahtevi za eksperimentalnim ispitivanjima brodova-potiskivaca smanjuju...

The subject of this doctoral thesis is the towboat shaft power required to propel pushed convoys. The research is based on the performed full-scale trials obtained at the serbian part of the river Danube. The main aim of the dissertation is to determine the functional relationship between the pushed convoy speeds through water and towboat shaft powers according to main dimensions of pushed convoys. On the basis of functional relationships between pushed convoy speeds through water and towboat shaft powers ship captains and crews will receive the necessary information about the operating characteristics of the certain pushed convoys. Data modelling was performed by using Artificial neural networks. Data modeling is performed by using artificial neural networks. Artificial neural networks have been chosen as being the most suitable data analytic method with abilities to effectively work with highly nonlinear and multi-dimensional data, their modeling flexibility, their generalization ability, their adaptability and good predictive ability. Ensembles of neural networks with their component networks were used to predict towboat shaft power. The entire set of experimental data was divided into a training data set and testing data set. Before the training of component artificial neural networks, Akaike information criterion was adopted to rank several configurations of networks. According to their rankings, several component networks with different architectures were selected to make ensemble. Once that training of comonent networks was finished, the analysis of output results was made and the boundaries of the model of Ensemble of beural network was determined. In this paper the original model of neural network ensemble is proposed to calculate the required towboat shaft power as well as to select appropriate towboat installed power depending on on the pushed convoy speed and main dimensions. The model is applicable to different pushed convoys that were part of full-scale trials...