FORECASTING AND REINFORCEMENT LEARNING STRATEGIES FOR EFFICIENT ENERGY EXCHANGE IN PEER-TO-PEER ENERGY TRADING GAME AMONG NANO/MICROGRIDS: EMPIRICAL ANALYSIS

Date

2023-9-7

Author

Güneş, Rabia Şeyma

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

113
views

38
downloads

Cite This

New technologies included in distributed energy systems have created solutions that allow the management of demand and generation variability in the electricity grid and the costs arising from this variability. Trade between these small grids has enabled the sale of excess energy between each other and the purchase of needed energy, thus reducing costs and system constraints. The purpose of this trade is modeled as a game of agents mentioned in reinforcement learning, enabling the creation of the market that offers those benefits from each peer. Each peer provides its electricity demand with both internal resources and other peers. The aim of this thesis is to comply with system constraints while providing the demand of each peer in this game aiming at maximum benefit. A novel Multi-Agent Reinforcement Learning model to facilitate very short-term energy trading among peers is suggested in this thesis. The key contributions of this thesis lie in incorporating very short-term load, generation, and price forecasts into the framework to enable more accurate decision-making by individual agents. To evaluate the performance of the proposed model, it is conducted extensive simulations using real-world data collected from various peers. The results compared with rule-based working agents. The experiment shows incorporating very short-term forecasts significantly enhances the ability of agents to adapt to rapidly changing conditions, thereby leading to more efficient and stable energy trading decisions. The use of very short-term forecasts empowers prosumers to make informed decisions in response to dynamic energy market conditions, ultimately contributing to increased grid reliability, energy efficiency, and sustainability.

Subject Keywords

Energy Trading, Multi-Agent, Reinforcement Learning, Peer-to-Peer Trading, Very Short-Term Forecasts

URI

https://hdl.handle.net/11511/105540

Collections

Graduate School of Informatics, Thesis