Thesis

PhD Thesis

Multi-Agent Reinforcement Learning for Distributed Solar-Battery Energy Systems

Supervisor

Prof. Kenji Doya

Thesis committee

Prof. Hiroaki Kitano,  Prof. Gail Tripp

Examiners

External

Prof. Zoltan Nagy (The University of Texas at Austin) (Intelligent Environment Lab)
Prof. Ziang (John) Zhang (Binghamton university SUNY) (Zhang'S Research Group)

Chair

Prof. Franz Meitinger (OIST) (Cell Proliferation and Gene Editing Unit)

Highlights

  1. A case study in local actual Direct current open energy system (DCOES) with reinforcement learning methods to save more energy and realize intelligent control policies.
  2. Achieved flexible energy exchange rules based on the energy storage systems (battery RSOC) in different agents/houses in a multi-agent system.
  3. Experience replay especially prioritized experience replay in DRL further boosts the performance. States' options prove that knowing community RSOC allows the most energy sharing and further reduces the purchasing power.
  4. One successful application implementation with open sourced energy emulator Autonomous Power Interchange System (APIS) developed with SonyCSL. (Link).

Download

PDF (Thesis cover)