Dynamic Positioning Control of Unmanned Surface Vehicles Based on an Improved Long Short-Term Memory Network

This article addresses dynamic positioning control for an unmanned surface vehicle subject to state time delays and ocean disturbances, and develops a long short-term memory (LSTM) learning-based approach. A Lyapunov matrix-based Lyapunov–Krasovskii functional is constructed to incorporate delay information, on which a delay-compensation strategy and a dynamic positioning controller are designed. External disturbances are estimated by an enhanced LSTM model with a selective state-update mechanism and an adaptive mixed-gradient rule, improving the learning of rapidly varying disturbance components while reducing computational burden. The resulting scheme ensures closed-loop stability and satisfactory positioning performance, and the proposed LSTM-based estimator achieves more accurate reconstruction of nonlinear time-varying ocean disturbances than existing methods. Finally, simulations demonstrate the effectiveness and advantages of the control strategy.