Path Planning For Multi-Glider Cooperative Adaptive Sampling Using Gaussian Process Regression

Underwater gliders are an important part of ocean observation network.Multi-glider formation and cooperative observation have great potential in large scale,long endurance,continuous observation.Since current underwater gliders can only carry limited energy,multi-glider cooperative adaptive sampling strategy becomes essential for efficient field sampling.This thesis presents multi-glider path planning methods based on Gaussian process regression and Markov decision processes,and aim at improving the efficiency and accuracy of ocean observation.An informative sampling strategy is proposed and adaptive sampling has been achieved under uncertainties caused by ocean currents.The main contributions of this thesis are listed as follows:(1)The Gaussian Process Regression method is used to plan the cooperative sampling path of multiple underwater gliders.In the path planning strategy,the goal is to maximize the sampling information,and the sampling position is selected based on the standard deviation of sampling.At the same time,the problem of asynchronous surface of multiple underwater gliders is considered.The simulated temperature field and the real temperature field in the South China Sea are carried out,and the feasibility of the method is verified.(2)Ocean current is the main external interference that affects the path planning and sampling accuracy of underwater glider.How to effectively reduce the interference of ocean current on the sampling planning of underwater glider is the key to improve the sampling efficiency and observation quality.In this paper,according to the interference of ocean current,under the framework of Markov decision process model,combined with the glider motion and ocean current interference design state transition probability,the state transition function obeying Gaussian distribution is established,and the state transition distribution is updated in real time.At the same time,the sampling strategy of greedy method is distinguished,and the sampling points are selected by far-sighted method according to the infinite field value function The optimal strategy guides the glider to complete the sampling,and realizes the optimal decision of adaptive sampling planning under the interference of current.(3)Based on the above Markov Decision Process model,combined with the Gaussian Process Regression method,the path planning of multiple underwater gliders is carried out,and the adaptive sampling of marine environment is completed.Taking the ocean temperature field as the background,the path planning results under the influence of different ocean flow fields are simulated,including the sinusoidal and gradient changes of current velocity,as well as the flow field and vortex field with uniform size and different directions.The simulation results show that the path planning method based on Gaussian Process Regression and Markov Decision Process model can achieve less repeated sampling area and high reconstruction quality of observation environment field.Finally,the optimal placement and the number of gliders are discussed.