Deep Multi-Agent Reinforcement Learning for Cost Efficient Distributed LoadFrequency Control

The rise of microgrid-based architectures is heavily modifying the energy control landscape in distribution systems. Decentralised control mechanisms are needed to ensure reliable power system operations. We propose using Reinforcement Learning to implement load frequency control without requiring a centralised authority. Specifically, we approximate the optimal solution using Multi-Agent Deep Deterministic Policy Gradient (MADDPG) at all levels: primary, secondary and tertiary. Generation units are characterized as agents that learn how to maximize their long-term performance by acting and interacting with the environment to balance generation and load, thus restoring frequency. We prove numerically that our Reinforcement Learning methodology can be used to implement the load frequency control in a decentralised way, even when more than one balancing authority is considered