Stochastic Multistage Satellite Constellation Reconfiguration for Tracking Uncertain Targets

Natural disasters are highly impactful and unpredictable planetary phenomena that occur on various spatial and temporal scales. The modeling and prediction of such disasters heavily depend on understanding local natural processes, which requires extensive data collection, especially through satellite-based remote sensing. Constellation reconfigurability is a leading-edge operational concept for Earth observation satellites, with the potential to enable high spatial and temporal resolution data acquisition by reconfiguring a constellation of satellites into a more optimal configuration via satellite maneuvers. Leveraging orbital maneuverability, constellation reconfigurability has been demonstrated as a promising solution to respond to dynamic events. Previous literature assumes a deterministic setting, in which either a case study or the entire formulation is considered to have a priori knowledge of target properties. This is not accurate to reality, as the progression of natural disasters is governed by stochastic processes that are only discernible in retrospect. This unpredictability complicates the decision-making process, often resulting in suboptimal reconfiguration strategies. In response to this gap in the literature, we present a stochastic variant of the Multistage Constellation Reconfiguration Problem (MCRP) leveraging the use of a Monte Carlo Tree Search (MCTS) algorithm to account for target uncertainties, which builds upon the authors' prior work on a deterministic variant. Additionally, the use of MCTS allows for near-optimal decisions regarding the reconfiguration of a given constellation with respect to a number of potential probable target scenarios. A hyperparameter search is conducted with respect to the number of iterations and simulation rollouts to determine the most effective hyperparameters in obtaining the most rewards. To verify the stochastic MCRP variant, a case study is conducted with Hurricane Irma, a deadly Category 4 hurricane that struck the Southeastern United States in September of 2017. The results of the case study depict that a stochastic MCRP can outperform a baseline constellation of non-reconfigurable satellites and a random policy wherein a constellation of satellites reconfigures in a random feasible manner. Simultaneously, a stochastic MCRP accounts for the stochastic nature of a target while obeying visibility, feasibility, and maneuver budget constraints.