A METHODOLOGY FOR DEVELOPING A CASCADE MODEL FOR CALCULATING SHIP COORDINATES BASED ON DEEP LEARNING

Abstract

The aim of the article is to study the methodology for developing a cascade model for calculating ship coordinates based on deep learning, capable of predicting the relative speed of a ship in a fixed coordinate system under the influence of wind and waves. This goal is achieved by building the topology of the cascade model for calculating the ship's coordinates, determining the procedure for forming the training data set, and testing the training results. To determine the ship's location coordinates using autonomous navigation systems, it is necessary to ensure the stability of the ship along a given trajectory. However, satellite navigation systems have their drawbacks, such as limited use in certain environments, such as in environments where radio waves do not pass through, as well as problems with noise and system integrity. In cases where force majeure natural events are encountered that may damage satellite systems, ship tracking may depend on algorithms for processing navigation information in autonomous systems to improve the accuracy and integrity of integrated navigation systems that use satellite navigation data. Often, the determination of some external factors, such as disturbance parameters, is accompanied by significant errors, which leads to significant uncertainty in predicting the calculated coordinates. Compared to existing algorithms, the use of neural networks can be more effective under conditions of uncertainty, as they have nonlinear properties of converting input signals into output signals. Thus, the prediction of the calculated coordinates under the influence of external factors fully meets these conditions. The most significant result is a cascade model for calculating the ship's coordinates based on long short-term memory (LSTM). As a further line of work, it can be noted that even if this predictive model is more accurate than traditional methods, its efficiency should be improved to achieve optimal results. Also, it is important to note that the optimisation hyperparameter is a key element for deep learning models. Hyperparameters determine both the training performance of the model and the quality of its testing.