Developing Strategy for Coastal Marine Debris Disposal Network Using Deep Reinforcement Learning

The process of collecting and disposing of coastal marine debris isn’t usually economically feasible without government subsidies. However, Ishida et al. reported this process can be made profitable by linking Styrofoam waste oil liquefaction equipment with decentralized energy systems. Examples of these energy systems include cogeneration systems (CGS) and photovoltaic power generation. In this study, a model was developed to examine which part of a coastal debris disposal operation system should be assisted with profits generated from a coastal marine debris disposal network linked to CGS. This is expected to dispose of as much coastal marine debris as possible in a short period of time. Specifically, it was modeled using deep Q-learning (DQN), which is one of deep reinforcement learning methods. On the premise that a certain amount of marine debris reached the coast, the model incorporated the combination of three elements, namely greater profits gained by increasing the capacity of CGS, stocking profits for a certain period of time and marine debris collection. The model developed herein made it possible to handle the amount of coastal marine debris that wouldn’t have been treated if profits were spent on the coastal marine debris disposal network on a yearly basis. This DQN-based model resulted in the development of a strategy, which reflects the view that “the larger the initial amount of litter becomes, the longer the stock period of profits from CGS should become.”