Assessment of calculation performance of a neural network-based algorithm for bathymetric prediction of vessel latitude

This study presents a performance comparison of two algorithms for predicting vessel latitude based on seabed depth data. The first algorithm is classical, relying on the search for a string in a depth reference matrix that is closest to current measurements according to the mean absolute error. The second algorithm is based on a neural network, which predicts vessel latitude using a sequence of measured depth values as input. The neural network model is constructed using algorithms for dataset creation, training, and testing. The network consists of ten hidden layers, each containing 200 neurons with hyperbolic tangent activation functions. A validation set is employed during training to calculate the maximum absolute error, which serves as a criterion for optimal network state. Training, validation, and test datasets are generated via pseudo-random variations of the reference depth matrix to account for sea level fluctuations and systematic measurement errors. Depth data are prepared for five different spatial steps in latitude and longitude based on a layer of spot soundings from an electronic navigational chart. For each dataset variant, the neural network is trained and its computational performance is compared to that of the classical search algorithm. Results demonstrate a significant computational advantage of the neural network for maximal reference data volume. However, as the dataset volume decreases due to increased spatial steps, this advantage diminishes, and the neural network no longer outperforms the classical method.

1. Klyueva, S. F., and V. V. Zav’yalov. Sintez algoritmov batimetricheskikh sistem navigatsii. Vladivostok: Mor. gos. un-t imeni G. I. Nevel’skogo, 2013.

2. Stepanov, O. A. Metody otsenki potentsial’noy tochnosti v korrelyatsionno- ekstremal’nykh navigatsionnykh sistemakh: Analiticheskiy obzor. Sankt- Peterburg: “Kontsern “Tsentral’nyy nauchno- issledovatel’skiy institut “Elektropribor”, 1993.

3. Ma, D., Y. Ben et al. “A robust fusion terrain- aided navigation method with a single beam echo sounder.” Ocean Engineering 286 (2023): 115610. DOI: 10.1016/j.oceaneng.2023.115610.

4. Golden, J. P. “Terrain Contour Matching (TERCOM): A Cruise Missile Guidance Aid.” Image Processing for Missile Guidance — 0238SPIE, 1980: 10–18. DOI: 10.1117/12.959127.

5. Savit, Carl H. U.S. 3 212 189, IPC G 01 C 21/20. Marine navigation procedure. Publ. 19 Oct. 1965.

6. Longenbaker, W. E. “Terrain- aided navigation of an unpowered tactical missile using autopilot-grade sensors.” Journal of Guidance, Control, and Dynamics 7.2 (1984): 175–182. DOI: 10.2514/3.8564.

7. Deryabin, V. V. and A. E. Sazonov. “Depth-aided prediction of vessel latitude based on a neural network.” Vestnik gosudarstvennogo universiteta morskogo i rechnogo flota imeni admirala S. O. Makarova 17.1 (2025): 94– 104. DOI: 10.21821/2309-5180-2025-17-1-94-104.

8. Deryabin, V. V. “Depth-aided prediction of a vessel’s longitude using a neural network.” Vestnik gosudarstvennogo universiteta morskogo i rechnogo flota imeni admirala S. O. Makarova 17.4 (2025): 481–492. DOI: 10.21821/2309-5180-2025-17-4-481-492.

9. Deryabin, V. V. “Neural net-based bathymetric solution of vessel position fixing.” Transport Business in Russia 2 (2024): 257–260.

10. Wang, M.-X. and Y. Qu. “Approximation capabilities of neural networks on unbounded domains.” Neural Networks 145 (2022): 56–67. DOI: 10.1016/j.neunet.2021.10.001.

11. Gorban’, A. N. “Obobschennaya approksimatsionnaya teorema i vychislitel’nye vozmozhnosti neyronnykh setey.” Sibirskiy zhurnal vychislitel’noy matematiki 1.1 (1998): 11–24.

12. Kingma, D. P. and J. Ba. “Adam: A Method for Stochastic Optimization.” 3rd International Conference on Learning Representations, 2017. DOI: 10.48550/arXiv.1412.6980.