Bayesian optimization for monitoring the dynamic environment

How to sample the data in an optimization algorithm is important in an environmental monitoring problem. Ensuring the sampling method practical while obtaining useful information as much as possible to reduce time and energy cost during optimization is the key. This thesis focuses on the implementation of Bayesian Optimization (BO) to monitor a time-varying three-dimensional environment. The BO algorithm is based on the Gaussian Processes (GPs) surrogate models which are non-parametric regression methods, and uses the reward function for decision making. An uniquely designed kernal function is used in GPs to learn the underlying pattern of spatial and temporal variations. A seies of theoretical but less practical experiments are developed to prove the capability of BO, together with presenting the importance of temporal information. A continuous path planning is designed to replace the waypoint planning for a real path design in the environmetal monitoring. Furthermore, this planning is effective to balance the trade-off between the exploration and the exploitation in the optimization problem.M.S.Includes bibliographical reference