Abstract:The accurate prediction of drift trajectories holds paramount significance for disaster response and navigational safety. The future positions of underwater drifters in the ocean are closely related to their historical drift patterns. Additionally, leveraging the complex dependencies between drift trajectories and ocean currents can enhance the accuracy of predictions. Building upon this foundation, we propose a Transformer model based on double-branch adaptive span attention (DBASformer), aimed at capturing the multivariate time-series relationships within drift history data and predicting drift trajectories in future periods. DBASformer can predict drift trajectories more accurately. The proposed adaptive span attention mechanism exhibits enhanced flexibility in the computation of attention weights, and the double-branch attention structure can capture the cross-time and cross-dimension dependencies in the sequences. Finally, our method was evaluated using datasets containing buoy data with ocean current velocities and Autonomous Underwater Vehicle (AUV) data. The raw data underwent cleaning and alignment processes. Comparative results with five alternative methods demonstrate that DBASformer improves prediction accuracy.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of accurately predicting drift trajectories in the ocean. Specifically, the author focuses on how to use historical drift data and complex ocean current dependencies to improve the prediction accuracy of the future positions of underwater drifting objects (such as buoys, Autonomous Underwater Vehicles (AUVs), etc.). #### Main challenges: 1. **Time - series features**: The positions of underwater drifters change continuously over time, and this change is affected by the speed of ocean currents. Therefore, it is necessary to effectively capture the temporal characteristics of these time - series data. 2. **Multivariable dependence**: Drift trajectories are not only affected by time, but also related to multiple other variables (such as ocean current speed, direction, etc.). Traditional models have difficulty effectively handling the complex dependencies among these multivariables. 3. **Long - and short - term dependencies**: In long - time - series prediction, traditional models may encounter the problems of vanishing or exploding gradients, leading to a decline in prediction accuracy. #### Solutions: To solve the above problems, the author proposes a Transformer model based on the Double - Branch Adaptive Span Attention mechanism (DBASformer). The main features of DBASformer include: 1. **Adaptive - span attention mechanism**: By introducing a learnable attention span, the flexibility of the attention mechanism is improved, and the attention weights can be more flexibly allocated. 2. **Two - branch attention structure**: It captures the dependencies between the time dimension and multi - dimensional variables respectively, thereby better extracting the features in the sequence. 3. **Data preprocessing and experimental verification**: Experiments are carried out using a real - world dataset containing buoy and AUV data, and compared with five other methods to verify the effectiveness and superiority of DBASformer. Through these improvements, DBASformer can exhibit higher prediction accuracy and computational efficiency when processing multivariable time - series data. ### Summary The core problem of this paper is to improve the prediction accuracy of drift trajectories in the ocean by improving the attention mechanism in deep - learning models, especially the Transformer model. This not only helps in disaster response and navigation safety, but also provides support for marine ecology and resource management.

Prediction of Drift Trajectory in the Ocean Using Double-Branch Adaptive Span Attention

Developing an Artificial Intelligence-Based Method for Predicting the Trajectory of Surface Drifting Buoys Using a Hybrid Multi-Layer Neural Network Model

Enhanced Multimodal Trajectory Prediction for Autonomous Vehicles Using Advanced Diffusion Model Techniques

Short-Term Drift Prediction of Multi-Functional Buoys in Inland Rivers Based on Deep Learning

Influence of Ocean Current Features on the Performance of Machine Learning and Dynamic Tracking Methods in Predicting Marine Drifter Trajectories

Multi-Scale Ocean Driving Factors and Time–Space Attention Mechanism

TripleConvTransformer: A deep learning vessel trajectory prediction method fusing discretized meteorological data

MSTFormer: Motion Inspired Spatial-temporal Transformer with Dynamic-aware Attention for long-term Vessel Trajectory Prediction

Enhanced Transformer Framework for Multivariate Mesoscale Eddy Trajectory Prediction

DiffuTraj: A Stochastic Vessel Trajectory Prediction Approach via Guided Diffusion Process

Ship Trajectory Prediction Using ACoAtt–LSTM and AIS Data

Deep bi-directional information-empowered ship trajectory prediction for maritime autonomous surface ships

Flow Field Analysis and Development of a Prediction Model Based on Deep Learning

STIA-DJANet: Spatial–Temporal Intention-Aware vessel trajectory prediction based on Dual-Joint Attention Network for e-navigation

Dual-Path Spatial–Temporal Attention Network with Multi-Attribute Information

A Trajectory Prediction Method of Drogue in Aerial Refueling Based on Transfer Learning and Attention Mechanism

Using LSTM with Trajectory Point Correlation and Temporal Pattern Attention for Ship Trajectory Prediction

High-fidelity data supported ship trajectory prediction via an ensemble machine learning framework

Ship Trajectory Prediction in Complex Waterways Based on Transformer and Social Variational Autoencoder (SocialVAE)

Vessel Navigation Behavior Analysis and Multiple-Trajectory Prediction Model Based on AIS Data

Hybrid-driven vessel trajectory prediction based on uncertainty fusion