Machine learning-based semi-online performance optimisation for long-range underwater glider missions

Kunpeng Feng,Lijie Tan,Jinhu Cai,Hongyu Wu,Yuxing Hao,Yunqiang Yang,Shaoze Yan
DOI: https://doi.org/10.1080/17445302.2024.2419514
2024-10-30
Ships and Offshore Structures
Abstract:This paper proposes a semi-online optimisation method for enhancing the performance of underwater gliders during long-range missions. The design variables include control parameters and target depth. The optimisation objective is to maximise the energy efficiency, motion accuracy and voyage velocity for a single profile. Initially, environmental data and glider position are collected by the monitoring equipment and communication satellite in real-time and then fed into the dynamic model. Subsequently, a training dataset is generated through dynamic simulation to establish a surrogate model by machine learning technology. Then, iterative calculations are executed based on the multi-objective optimisation algorithm and surrogate model to determine the optimal solution of design variables. Finally, an optimal solution is transmitted to the glider via a communication satellite, and a single profile gliding motion is executed. This process repeats until the glider reaches the target exploration point. The numerical case validates the effectiveness of this method.
engineering, marine
What problem does this paper attempt to address?