Analysis of the neural network application effectiveness in predicting collision avoidance maneuvers for two vessels

The effectiveness of using neural networks to determine collision avoidance maneuvers between two vessels is analyzed in the paper. A brief description of the algorithm and MATLAB script that facilitates finding course alterations to prevent collisions between pairs of vessels is provided. The process of creating training data using a previously developed script, including the preliminary data processing to eliminate unrealistic scenarios of vessel approach, as well as situations where there is no risk of collision, is described. The neural networks are trained using Levenberg-Marquardt and Adam algorithms. Throughout the study, 11 neural networks with various parameters are trained. The one that allows predicting course changes for safe distance avoidance for pairs of vessels with an accuracy of 94.8 % is selected. The accuracy of the neural networks predictions in this study is defined as the number of initially dangerously approaching vessel pairs whose closest point of approach after being processed by the neural network is within 0.8 to 1.2 miles, divided by the total number of vessel pairs. The time spent on calculating avoidance maneuvers using the algorithm and the neural network is compared. It is shown that as the number of dangerously approaching vessels increases to four or more, the neural network takes five times less time to predict an avoidance maneuver than the algorithm. With an increasing number of dangerously approaching vessels, the gap in data processing time between the neural network and the algorithm widens, which confirms the appropriateness of using neural networks in processing large data sets with pairs of dangerously approaching vessels. Future research is aimed at developing an algorithm to address the challenge of calculating safe collision avoidance maneuvers for groups of vessels through pairwise analysis of collision risks.

1. Huang, Yamin, Linying Chen, Pengfei Chen, Rudy R. Negenborn, and P.H.A.J.M. van Gelder. “Ship collision avoidance methods: State-of-the-art.” Safety science 121 (2020): 451–473. DOI: 10.1016/j.ssci.2019.09.018.

2. Tripolets, Oleg Y. “Overview of existing methods of autonomous vessels collision avoidance.” Vestnik Gosudarstvennogo universiteta morskogo i rechnogo flota imeni admirala S. O. Makarova 13.4 (2021): 480–495. DOI: 10.21821/2309-5180-2021-13-4-480-495.

3. Golovinov, A., and E. Klimova. “Advantages of neural networks before traditional algorithms.” Eksperimental’nye i teoreticheskie issledovaniya v sovremennoi nauke. Novosibirsk: Assotsiatsiya nauchnykh sotrudnikov «Sibirskaya akademicheskaya kniga», 2017. 11–15.

4. Kononova, N. V., T. I. Avdeeva, P. E. Oblasov, and I. V. Grigorieva. “Аdvantages of neural networks.” Aktual’nye problemy informatizatsii obrazovaniya: opyt, problemy, perspektivy razvitiya. Cherkessk: Bibliotechnoizdatel’skii tsentr Severo-Kavkazskoi gosudarstvennoi akademii, 2020. 91–93.

5. Ahn, Jin-Hyeong, Key-Pyo Rhee, and Young-Jun You. “A study on the collision avoidance of a ship using neural networks and fuzzy logic.” Applied Ocean Research 37 (2012): 162–173. DOI: 10.1016/j.apor.2012.05.008

6. Burmaka, I. A., and A. Yu. Bulgakov. “Management of the group of vessel in the situation of dangerous approach.” Vestnik Gosudarstvennogo universiteta morskogo i rechnogo flota imeni admirala S. O. Makarova 6(28) (2014): 1–13.

7. Burmaka, I. A., G. E. Kalinichenko, and M. A. Kulakov. “Management by pair of vessels in situation of dangerous rapprochement.” Vestnik Gosudarstvennogo universiteta morskogo i rechnogo flota imeni admirala S. O. Makarova 3(37) (2016): 64–70. DOI: 10.21821/2309-5180-2016-7-3-64-70.

8. Tripolets, Oleg Y. “Calculating a collision avoidance maneuver for two unmanned ships by minimizing a cost function in MATLAB.” Vestnik Gosudarstvennogo universiteta morskogo i rechnogo flota imeni admirala S. O. Makarova 15.5 (2023): 876–884. DOI: 10.21821/2309-5180-2023-15-5-876-884.

9. Mhaskar, Hrushikesh, Qianli Liao, and Tomaso Poggio. “When and why are deep networks better than shallow ones?” Proceedings of the AAAI conference on artificial intelligence. Vol. 31. No. 1. 2017. DOI: 10.1609/aaai.v31i1.10913.

10. Liang, Shiyu, and Rayadurgam Srikant. “Why deep neural networks for function approximation?” 5th International Conference on Learning Representations (ICLR). 2016. DOI: 10.48550/arXiv.1610.04161.