Trajectory Tracking of Unmanned Marine Vehicles Under DoS Attacks: A Hybrid Neural Network Predictive Control Method

This article addresses the trajectory tracking problem of Unmanned Marine Vehicles (UMVs) under Denial of Service (DoS) attacks and proposes a resilient control scheme. The core is the use of a hybrid neural network predictor and the establishment of a generalization error bound, providing new insights for cybersecurity control of marine unmanned systems. Keywords: Unmanned Marine Vehicle, DoS attack, trajectory tracking, neural network, predictive control, cybersecurity, marine robot. Original author: hyecho-1101, Source: GitHub, Release date: 2026-06-13.

With the growing demand for marine resource development, UMVs are widely used in mapping, monitoring, and other scenarios, but their reliance on wireless communication makes them vulnerable to DoS attacks. DoS attacks block communication links, leading to loss of control commands, interruption of state information, timing disorders, and even system instability. Traditional control methods assume communication availability and perform poorly against attacks, requiring resilient control strategies.

The core of the solution is a predictive compensation mechanism: when communication is interrupted, historical data is used to predict future control inputs. A hybrid neural network architecture is adopted: RNN/LSTM captures temporal dependencies, fully connected layers handle nonlinear transformations, and an optional attention mechanism improves accuracy. Additionally, a generalization error bound is established to quantify the impact of attacks on stability and guide parameter design.

1. System Modeling and Data Collection: Establish a UMV dynamic model, collect state-control data under normal communication, and preprocess (including outlier removal, normalization, etc.). 2. Training Strategy: Use MSE/Huber loss, Adam optimizer, combined with regularization and early stopping to improve generalization. 3. Real-Time Integration: In normal mode, receive commands and record data; switch to prediction mode after attack detection; smooth transition when communication is restored.

1. Marine Security and National Defense: Enhance the survivability of military UMVs in adversarial environments. 2. Civil Marine Operations: Ensure task execution when communication is limited (e.g., bad weather, interference). 3. Cyber-Physical Systems (CPS) Research: The method can be extended to drone swarms, smart grids, and other CPS to solve common resilient control problems under DoS attacks.

Summary: The hybrid neural network predictive control scheme enables autonomous control of UMVs under DoS attacks, with theoretical error bound guarantees. Future Directions: Multi-agent collaboration, research on complex attack models, edge computing optimization, adaptive online learning. This scheme provides useful insights for the security of marine robot systems.