Description: For automated forecasting of adverse accidents in flight, the article proposed an approach based on the use of models and methods of artificial intelligence. The article in the framework of information technology of automated forecasting of adverse accidents in flight using deep neural networks shows the results of modeling the learning process of a hybrid neural network based on convolutional and recurrent neural networks using the Keras and TensorFlow frameworks to determine the optimal value of the learning rate for a certain number of epochs learning. The purpose of the article is to assess the effectiveness of the use of information technology for automated forecasting of adverse accidents in flight using deep neural networks. Using TensorBoard capabilities, graphs are provided for the dependence of correct prediction of adverse accidents in flight on the number of epochs of learning of a hybrid neural network to identify and classify adverse accidents from a particular alphabet of classes. The accuracy of forecasting adverse accidents in flight during the operation of the hybrid neural network using the proposed information technology was evaluated. A comparison was made of the results of evaluating the accuracy of forecasting unfavorable aviation accidents in flight using existing approaches based on neural network models of the classic RNN, LSTM, CNN and the proposed approach based on a modified neural network classifier using the CNN and RNN. The comparison results allow us to conclude that the application of the developed information technology, which implements this hybrid neural network model, allows us to obtain a gain in accuracy and completeness of the classification of adverse accidents in flight. Prospects for further research in this direction may be the development of proposals for software implementation of the developed hybrid neural network model as part of an intelligent decision support system for automated forecasting of adverse aviation incidents in flight.
Keywords: information technology, deep neural network, forecasting, unfavorable aviation incident, framework, efficiency
1. Kuklev, E.A. (2003), “Otsenivanie urovnya bezopasnosti poletov v grazhdanskoy aviatsii v riskovyih situatsiyah na osnove tsepey sluchaynyih sobyitiy” [Safety assessment in civil aviation in risk situations based on chains of random events], Science and technology of transport, No. 2, pp. 4-14.
2. Sharov, V.D. and Makarov, V.P. (2011), “Metodologiya primeneniya kombinirovannogo metoda FMEA-FTA dlya analiza riska aviatsionnogo sobyitiya” [Methodology of using the combined FMEA-FTA method for analyzing the risk of an aviation event], Scientific Bulletin of MSTU GA. Series “Operation of air transport. Flight safety”, No. 174, pp. 18-24.
3. Zubkov, B.V. and Sharov, V.D. (2010), “Teoriya i praktika opredeleniya riskov v aviapredpriyatiyah pri razrabotke sistemyi upravleniya bezopasnostyu poletov” [Theory and practice of identifying risks in airlines in developing a safety management system], MGTU GA, Moscow, 196 p.
4. Sharov, V.D. (2007), “Metodika otsenki veroyatnosti vyikatyivaniya vozdushnyih sudov za predelyi VPP pri posadke” [Methodology for estimating the likelihood of aircraft rolling out of the runway during landing], Scientific Bulletin of MSTU GA. Series “Operation of air transport and repair of aircraft. Flight safety”, No. 122, pp. 61-66.
5. Butov, A.A. (2012), “Avtomatizirovannaya sistema prognozirovaniya i predotvrascheniya aviatsionnyih proisshestviy pri organizatsii i proizvodstve vozdushnyih perevozok. Etap 4. Adaptatsiya razrabotannyih algoritmov i programmnyih sredstv AS: otchet o NIR” [Automated system for predicting and preventing accidents in the organization and production of air traffic. Phase 4 Adaptation of developed algorithms and software AU: report on research], UlGU, Ulyanovsk, 317 p.
6. Karpathy, A. (2015), The unreasonable effectiveness of recurrent neural networks, available at: www.karpathy.github.io/2015/05/21/rnn-effectiveness/.
7. Britz, D. (2015), Understanding convolutional neural networks for NLP, available at: www.wildml.com/2015/11/understanding-convolutional-neural-networks-for-nlp/.
8. Aleksieiev, O.M. (2009), “Avtomatyzatsiia protsesiv upravlinnia bezpekoiu polotiv v aeronavihatsiinii systemi: dysertacija” [Automation of flight safety management processes in the aeronautical system: dissertation], Kyiv, 268 p.
9. Vasyliev, V.M. (2006), “Metody monitorynhu ta rozviazannia konfliktnykh sytuatsii v systemi kooperatyvnoho upravlinnia povitrianym rukhom: dysertacija” [Methods of monitoring and solving conflict situations in the system of cooperative air traffic control: dissertation], Kyiv, 358 p.
10. Nedelko, V.N. and Nedelko, S.N. (2006), “Mnogotselevoy modeliruyuschiy kompleks obsluzhivaniya vozdushnogo dvizheniya” [Multipurpose modeling air traffic service complex], Collection of sciences. Works of the Institute of Modeling Problems in Cybernetics named after G.E. Puhov “Modern training equipment and systems”, pp. 90-93.
11. Harchenko, V.P. (1994), “Problemyi razvitiya i metodyi upravleniya effektivnostyu sistem aeronavigatsionnogo obsluzhivaniya: dysertacija” [Development problems and methods for managing the efficiency of air navigation systems: dissertation], Kyiv, 448 p.
12. Khrashchevskyi, R.V. (2011), “Metodolohichni osnovy syntezu adaptyvnoi systemy planuvannia rozpodilu povitrianoho prostoru: dysertacija” [Methodological bases of synthesis of adaptive airspace planning planning system: dissertation], Kyiv, 390 p.
13. Google Research Team, (2016), Tensorflow: Large-scale machine learning on heterogeneous distributed systems, available at: arXiv:1603.04467 [cs.DC].
14. Nguen, T.K. (2015), “Informatsionnaya tehnologiya lokalizatsii tekstovyih oblastey na izobrazheniyah so slozhnyim fonom: dysertacija” [Information technology of localization of text areas on images with a complex background: dissertation], Odessa, 202 p.