SHIPBUILDING

DOI.org/10.5281/zenodo.897021

Pyatakovich V., Vasilenko A., Mironenko M.

VALERIY PYATAKOVICH, Candidate of Engineering Sciences, Associate Professor, Chief of a Research Laboratory, e-mail: pva.877com@mail.ru
ANNA VASILENKO, Candidate of Engineering Sciences, Researcher of a Research Laboratory, e-mail: kahunya@gmail.com
The Pacific Higher Naval College named after S.О. Makarov
6 Kamskiy pereulok, Vladivostok, Russia, 690062
MIKHAIL MIRONENKO, Doctor of Engineering Sciences, Professor, Chief Researcher, Hydrophysics Laboratory
Special Research Bureau for Automation of Marine Researches FEB RAS
25 Gorky St., Yuzhno-Sakhalinsk, Russia, 693023

Neural network training as a stage in the expert system development to classify the sources of physical fields when monitoring water areas

Abstract: The article presents the results of the next stage in the authors' investigation aimed to create a monitoring system of marine areas based on the development of non-linear rear-illuminated tracing hydroacoustics that uses neural network technology. Numerous experiments on training the neural networks have demonstrated that the cumulative local optimisation algorithm, «knocking out» the network of local minimum and procedures for increasing the number of neurons lead to successful learning of neural networks. It has been demonstrated that the solution of a number of tasks: simulating instant location of recognised object and its movement parameters; the simulation of climatic variability distribution of the vertical speed of sound in water and sonar settings fields using a database of typical vertical distribution curves of the speed of sound; the simulation of information flows that occur in situations of “receiver-source”; the development of a module managing the simulation model; the assessment of the adequacy of the model; the optimization of the simulation model, enables one to arrive at a sample a sample of the sufficient amount of data for training neural networks.

Key words: сlassification of sea objects, fuzzy logic, neural networks, neuro-fuzzy models, automatic control systems, training algorithms of neural networks.