Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dynamic Prediction of Full-Ocean Depth SSP by Hierarchical LSTM: An Experimental Result

Jiajun Lu, Wei Huang, Hao Zhang

TL;DR

This work tackles the challenge of forecasting the full-ocean-depth sound speed profile (SSP) to support underwater PNT. It introduces a hierarchical LSTM (H-LSTM) that layers data by depth and trains depth-specific models to predict the next-step SSP, with LSTM updates such as $C_t = f_t C_{t-1} + i_t \tilde{C}_t$ and $\hat{S}_{t+1} = o_t \tanh(C_t)$. Evaluations on ARGO-derived data and South China Sea ocean experiments show RMSEs of approximately $0.264\,\mathrm{m\,s^{-1}}$ and $0.153\,\mathrm{m\,s^{-1}}$, respectively, outperforming mean-value, polynomial fitting, and BP baselines, and the model effectively captures periodic SSP variations. The approach offers a dynamic, depth-aware SSP prediction capability with potential to enhance oceanic PNT and acoustic forecasting, by adjusting time steps to dataset characteristics and assembling full-depth SSP from layered predictions.

Abstract

SSP distribution is an important parameter for underwater positioning, navigation and timing (PNT) because it affects the propagation mode of underwater acoustic signals. To accurate predict future sound speed distribution, we propose a hierarchical long short--term memory (H--LSTM) neural network for future sound speed prediction, which explore the distribution pattern of sound velocity in the time dimension. To verify the feasibility and effectiveness, we conducted both simulations and real experiments. The ocean experiment was held in the South China Sea in April, 2023. Results show that the accuracy of the proposed method outperforms the state--of--the--art methods.

Dynamic Prediction of Full-Ocean Depth SSP by Hierarchical LSTM: An Experimental Result

TL;DR

This work tackles the challenge of forecasting the full-ocean-depth sound speed profile (SSP) to support underwater PNT. It introduces a hierarchical LSTM (H-LSTM) that layers data by depth and trains depth-specific models to predict the next-step SSP, with LSTM updates such as and . Evaluations on ARGO-derived data and South China Sea ocean experiments show RMSEs of approximately and , respectively, outperforming mean-value, polynomial fitting, and BP baselines, and the model effectively captures periodic SSP variations. The approach offers a dynamic, depth-aware SSP prediction capability with potential to enhance oceanic PNT and acoustic forecasting, by adjusting time steps to dataset characteristics and assembling full-depth SSP from layered predictions.

Abstract

SSP distribution is an important parameter for underwater positioning, navigation and timing (PNT) because it affects the propagation mode of underwater acoustic signals. To accurate predict future sound speed distribution, we propose a hierarchical long short--term memory (H--LSTM) neural network for future sound speed prediction, which explore the distribution pattern of sound velocity in the time dimension. To verify the feasibility and effectiveness, we conducted both simulations and real experiments. The ocean experiment was held in the South China Sea in April, 2023. Results show that the accuracy of the proposed method outperforms the state--of--the--art methods.
Paper Structure (12 sections, 3 equations, 6 figures, 4 tables, 1 algorithm)

This paper contains 12 sections, 3 equations, 6 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Structure of H--LSTM
  4. Workflow of H-LSTM for SSP Prediction
  5. Results and Discussion
  6. Ocean Experiments
  7. Experimental Settings
  8. Data source
  9. Platform and parameters
  10. Accuracy Performance of H-LSTM
  11. H-LSTM’s performance in predicting cyclical changes in SSPs
  12. Conclusion

Figures (6)

  • Figure 1: The structure of H-LSTM.
  • Figure 2: Comparison between predicted hierarchical or full-ocean depth SSP and actual hierarchical or full-ocean depth SSP (ARGO Dataset). (a), (b) or (c), (d) respectively represent February and October 2021. (a), (b) original 58 layers' SSP. (c), (d) resampled SSP.
  • Figure 3: Data collection. (a) Sampling by CTD and XCTD. (b) Data location.
  • Figure 4: Spatial positions of ARGO and Ocean Experiments SSP samples.
  • Figure 5: Comparison between predicted hierarchical SSP and actual hierarchical SSP (Ocean Experiments Dataset). (a)Comparison of layered data. (b) Comparison with different state--of--the--art methods.
  • ...and 1 more figures