Research On Adaptive Observation Of Ocean Environments With Underwater Vehicles Based On Data-driven Approach

Observation is fundamental to understanding and prediction of the status,process and variation of ocean environments.The significant technical framework of realizing adaptive ocean observation is based on the closed-loop feedback system architecture,which integrates underwater vehicles and predictive ocean models.In the framework,the in-situ data collected by underwater vehicles is used to establish an ocean model,and the adaptive data collection of ocean environments and dynamics ocean features required for optimal ocean state estimation and prediction are guided based on the running ocean model.Integrated numerical ocean models are important predictive ocean models,but they still have a number of technical challenges in many aspects and applications.Therefore,data-driven ocean modeling and prediction,and adaptive ocean observation with underwater vehicles based on data-driven ocean models are important issues that need to be investigated and resolved for implementing adaptive ocean observation with underwater vehicles.This thesis conducts research on data-driven adaptive ocean observation with underwater vehicles.And the main research work and results in this thesis are as follows:(1)Research on strategies and methods of estimation and prediction of ocean environments with in-situ sensing data collected by a network of underwater vehicles,which are sparse relative to the high-dimensional ocean states.Dynamic mode decomposition method and ocean modeling and prediction method based on dynamic mode decomposition are studied.Learning of dominant dynamic modes are implemented with data sets produced by a numerical ocean model.The lowdimensionality and multi-resolution characteristics of the temperature,salinity and currents in a selected local area in South China Sea are analyzed,and estimation and prediction of ocean states based on the learned dominant modes are computed,and proposes a data-driven ocean environment adaptive modeling strategy.(2)Research on method of optimizing the location of a network of underwater vehicles for accurate estimation of ocean states,based on the results of data-driven ocean modeling research.The mathematical model of the optimization problem is established,which optimizes the locations of a network of underwater vehicles to enable the network to collect best data for the aim to accurately reconstruct the ocean states with the dynamic mode decomposition-based ocean model.To solve the NP-hard optimization problem,an effective method is proposed which is based on the approach of reinforcement learning.Computer simulation is conducted,and the results demonstrate that accurate reconstruction of high-dimensional ocean states could be achieved with sparse data via the proposed method.(3)Research on strategies and methods of adaptive observation of ocean front-like gradient features with underwater vehicles,to meet the requirements of further accurate prediction of the features.Based on the above-mentioned data-driven ocean modeling,a motion planning strategy for an underwater vehicle to track and map an ocean gradient feature transported by ocean currents is proposed.And a behavior-based motion planning algorithm to track and map an ocean gradient feature is developed to implement the strategy,which is based on a behavior-based motion planner.In addition to theorical analysis,computer simulations are conducted in this thesis.And the computer simulation results demonstrated that the results of developed strategies,methods and algorithms are effective.