A Method to Identify Seawater Pycnocline Boundaries Based on LSTM Neural Network and its Application in Underwater Glider Missions

Underwater glider is a kind of buoyancy driven underwater vehicle; it has the features of low operating noise, long endurance and flexible maneuverability. During the glider long-time navigation, there is a high probability to encounter the pycnocline, which has a great influence on underwater glider's navigation as well as application. In this work, an operation flow of glider hovering movement in the pycnocline was discussed in detail in order to improve the observation performance of glider. We firstly developed an LSTM (Long Short-Term Memory) neural network to identify the pycnocline boundaries. Then the accurate buoyancy adjustment of glider hovering movement was calculated based on the obtained pycnocline information. Finally a hard ware in the loop simulation was implemented to verify the feasibility of the method, and the simulation results showed a reasonable performance of glider hovering in the pycnocline.