Load Frequency Control: A Deep Multi-Agent Reinforcement Learning Approach

The paradigm shift in energy generation towards microgrid-based architectures is changing the landscape of the energy control structure heavily in distribution systems. More specifically, distributed generation is deployed in the network demanding decentralised control mechanisms to ensure reliable power system operations. In this work, a Multi-Agent Reinforcement Learning approach is proposed to deliver an agent-based solution to implement load frequency control without the need of a centralised authority. Multi-Agent Deep Deterministic Policy Gradient is used to approximate the frequency control at the primary and the secondary levels. Each generation unit is represented as an agent that is modelled by a Recurrent Neural Network. Agents learn the optimal way of acting and interacting with the environment to maximise their long term performance and to balance generation and load, thus restoring frequency. In this paper we prove using three test systems, with two, four and eight generators, that our Multi-Agent Reinforcement Learning approach can efficiently be used to perform frequency control in a decentralised way.