DEVELOPMENT OF INFORMATION TECHNOLOGY OF AUTOMATED FORECASTING OF NON-RELATED AVIATION EVENTS IN FIELD

Abstract

In the work the information technology of automated forecasting of unfavorable aviation events in flight with the use of deep neural networks for solving problems of recognition and prevention of adverse aviation events is considered little and requires further research. The implementation of the developed information technology (IT) is proposed in four stages: preparatory, main, supplementary and final. The first (preparatory) stage of IT functioning is performed before the forecast of unfavorable aviation events in the flight. The input data is set in the form of archived text messages about the situation in the flight based on data from external sources. Outputs include the topology and weights of the convolutional neural network (NNN) that are transmitted to perform the main stage of the functioning of the IT. Basic, additional and final stages of IT functioning are performed in the process of forecasting unfavorable aviation events in flight. The inputs for the relevant steps are text messages about the situation in the flight based on data from external sources, as well as the topology and ZNM weights obtained at the preparatory stage using the recurrent neural network module (LSTM). The initial data of the main stage of IT functioning is the documented information on the results of forecasting unfavorable aviation events in the flight. The developed IT technology allows to realize the process of automated forecasting of unfavorable aviation events in flight using deep neural networks, with a clearly defined and interrelated set of stages, with the possibility of further parallelizing the implementation of their procedures, taking into account the peculiarities of presentation (formalization) of knowledge about the recognition of unfavorable aviation events in flight on textual information obtained and summarized from various sources on the results of airspace control.