Model Predictive Control Of Underwater Gliders Based On Neurodynamic Optimization

With the depletion of land resources, ocean resources are becoming increasingly important for the future of mankind. In order to be effectively and efficiently explorer ocean resources, it is necessary and rewarding to monitor oceanic environments in real time. The ability to carry out environmental surveillance determines the success of ocean exploration.Underwater gliders consist of a class of autonomous underwater vehicles that use the change of buoyancy forces to propel themselves. Underwater gliders have several attractive features for oceanic engineering, such as the low cost, high efficiency, high endurance, and low energy consumption. Underwater gliders are expected to play important roles in exploring the sea environments. One important issue in developing underwater gliders is to develop high-performance control systems. As the model of an underwater glider is generally nonlinear, time-varying, and underactuated, off-line designed control methods will probably result in poor performances. Model predictive control is a control strategy that generates control signals via online optimization. But due to the limited computation resources in an unwater glider, conventional optimization methods may not be efficient for solving the sequential optimization problems encountered in model predictive control. Neurodynamic optimization offers promises for addressing such issues.In this thesis, motion control of underwater gliders in vertical plane is studied. First, its model predictive control is formulated as a sequential quadratic programming. Then, a neural network called the simplified dual network is applied for solving the quadratic programming online. Simulation results show that the proposed control scheme is efficient and effective for soling the motion control problems of underwater gliders.