Yu. I. Nechaev

Use of neural network technologies by development of on-board intelligence systems (IS) of a safety of operation of a ship as complex dynamic object (DO) is connected with the decision of urgent tasks of the control, forecast and management. These tasks characterize of DO functioning at high probability of non-failure operation and given efficiency. A basis of such system is the management of a technical condition by results of the control and forecasting of meanings of dynamic parameters of object and external environment. The multiprocessor computer complex represents integrated IS of the control of seaworthiness, unsincability and strength of a ship. The functioning of a complex is provided on the basis of the given measurements, dynamic base of knowledge and methods of mathematical modeling. The processing of a flow of the information is carried out in a mode of real time depending on features of dynamics of interaction of object with external environment in complex hydrometeorological conditions. The formation of conceptual model of functioning of dynamic of knowledge base IS provides generation of the alternative concepts, analysis of variants and choice of preferable technology. Such formalization determines creation of procedures and development of the various approaches to a choice of preferable variant of the decision from set of alternatives on the basis of various methods of search of the decisions. The analysis has shown, that the most preferable variant of construction of dynamic base of knowledge is use of logic system of knowledge on the basis of нейро-indistinct systems and principle of an adaptive resonance used in the theory artificial neural networks (ANN) The theoretical base of IS is formed on the basis of an effective combination of the saved system of knowledge to the new approaches and paradigms of artificial intelligence (AI). Among them the important role belongs to methods and models constructed and use ANN within the framework of a competitions principles and ensuring: formalization and integration of knowledge (construction of membership functions), the mechanism of a logic conclusion (neuro-fuzzy systems), search of the decisions (neuro-approximation and neuro-forecast); development of the practical recommendations in conditions of uncertainty and incompleteness of the initial information. The perspective approach to synthesis of ANN algorithms is use of cognitive methods of representation and processing of the entrance information on the basis of construction of cognitive structures (cognitive spiral, texture data, cognitive card). The important result of researches connected to increase of quality of functioning of a measuring complex on-board ИС, is the creation of «intelligence» sensors on a basis ANN. The ANN realization within the framework of a principle of a competition by development on-board IS of new generations provides use of multilevel models of organization of multiprocessor computing environment. The top level is submitted to the managing computer which is carrying out communication with on-board subsystems, input of the information from external devices and compilation from languages of a high level. The bottom level consists of set of in parallel working processors ensuring the decision of separate functional tasks according to loading, determined managing computer. The acceleration of calculations is promoted by use of parallel principles and conveyor of researched tasks of an estimation of DO behavior. Therefore is provided re-configuration not only at a level of the divided computing system, but also at a level of information sensors of dynamic measurements with intelligence logic. In work the following sections are submitted: ANN technology at processing the information in multiprocessor computing environment, ANN technology by development of dynamic knowledge base, ANN technology at realization of a principle of non-linear self-organization, ANN models at control of DO and environment, ANN technology in non-formalization tasks, cognitive paradigm at designing of ANN controllers, ANN technology at realization of intelligence sensors. The used literature includes 27 names.