Online Optimization for Control Parameters of Underwater Gliders Based on Machine Learning

As an ocean exploration equipment with low energy consumption, underwater gliders have attracted extensive attention in recent years. For long-range exploration missions, this paper mainly studies the online optimization method for control parameters of underwater gliders (UGs). First, the actual data obtained by experimental design are used for training the machine learning model (MLM). The input parameters of MLM are the UG’s control parameters and starting point coordinates for a single profile. The output parameters include the specific energy consumption, average motion error, and average voyage velocity. Then, combining MLM and multi-objective optimization algorithm, the optimal control parameter values can be online determined and used for subsequent exploration missions. Especially, the online optimization will be executed again, when the UG’s performance shows a significant fluctuation. The feasibility of the proposed method is verified by dynamic simulation and radial basis function (RBF) neural network. This research work may provide certain theoretical guidance for UG’s intelligent control.