چكيده به لاتين
Due to the advantages of decentralized energy production compared to
centralized production, distributed energy resources, including renewable
energies, are expanding. These renewable energy resources have a random nature,
leading to uncertainty in production. On the other hand, electric loads within the
78
grid are not constant, posing new challenges for system control and the growing
need for the development of new controllers. In this thesis, we address the issue
of energy management for the Renewable Energy Laboratory of Heliocentris
Energy Solution AG, located at the University of Staffordshire. This laboratory
comprises solar cells, wind turbines, fuel cells, battery energy storage systems,
and electrical loads. To manage energy in this thesis, we have employed Q-
learning, SARSA, and Monte Carlo learning methods. In this multi-agent control
problem, all producers and consumers are modeled as autonomous agents capable
of learning. They interact with the environment to learn the optimal policy
without having information about other agents and aim to maximize their
profits.The goal of designing this energy management system is to meet power
balance constraints, increase producer profits, reduce consumer costs, and
decrease the dependence of the microgrid on the main grid. The Double Auction
algorithm is used to determine the domestic market price. After energy exchange
in the domestic market between producers and consumers, unmet loads and
surplus energy from producers are exchanged with the main grid. In this thesis,
three algorithms are compared using real data on wind speed, solar radiation, and
ambient temperature, among which SARSA outperforms the others
