Знання-орієтовані технології прийняття рішень під час руху суден у стиснених водах

The article determines that the most common cause of accidents, up to 80% of cases, is the problem of the «human element», which requires the construction of navigation decision support systems that are capable of using knowledge-based technologies that correspond to the order of human thinking, to provide the navigator with predicted data. Neural network algorithms allow taking into account these features, being specially oriented for solving forecasting problems. Neural networks act as a universal approximator of functions of several variables, which makes them a promising tool for solving ship maneuvering problems in compressed waters. A neural network that models ship dynamics is trained on the basis of samples obtained using ship motion equations (kinematic or differential dynamic). It is determined that neural network algorithms allow taking into account these features, being specially oriented for solving such problems. Calculation of the ship's path can be performed using a different set of navigation sensors. The output signal of the calculation system can be the coordinates of the vessel, their increments, which constitute the relative speed and other quantities. Of course, the most desirable option would be the case when the calculation system determines the coordinates of the vessel or at least their increment. The advantage of using coordinates as the output of the network has two aspects. The choice of a neural network model precedes the next stage, at which its training is performed. In this case, the problem of developing a method for obtaining the required number and quality of samples arises. The choice of a particular method for training the neural network also plays an important role. The choice of a neural network model precedes the next stage, at which its training is performed. In this case, the problem of developing a method for obtaining the required number and quality of samples arises. The choice of a particular method for training the neural network also plays an important role