Displacement Prediction Method for Surfacing Position of Underwater Gliders

Accurate prediction of the surfacing position of underwater gliders is crucial for enhancing mission execution efficiency and ensuring the safe recovery of the glider. Existing prediction methods are limited in accuracy and often involve high computational costs. To address these challenges, this paper proposes a Surfacing Position Prediction (SPP) displacement prediction method, which projects both the entry and surfacing positions of the glider onto the Universal Transverse Mercator (UTM) coordinate system and defines the prediction target as the glider's displacement. The predicted displacement is then added to the entry position to obtain the estimated surfacing position. To validate the proposed method, this study employs a Multilayer Perceptron (MLP) and eXtreme Gradient Boosting (XGBoost) as predictive models. Both models are trained on real-world sea trial data of gliders. The experimental results demonstrate that both models achieve satisfactory prediction performance, with MLP exhibiting higher accuracy compared to XGBoost. Specifically, the average prediction error, measured as the Haversine distance between the predicted and actual surfacing positions on the test set, is 0.44 kilometer for MLP. The proposed method is highly practical and can be applied to underwater robotic path planning and related tasks.