Reinforcement learning versus model predictive control: a comparison on a power system problem.

Area of Science:

• Control Systems Engineering
• Machine Learning
• Power Systems

Background:

• Electrical power systems require robust controllers to damp oscillations.
• Model Predictive Control (MPC) and Reinforcement Learning (RL) are advanced control strategies.

Purpose of the Study:

• To compare the performance of RL and MPC in a unified framework.
• To evaluate their application in synthesizing a controller for nonlinear electrical power oscillation damping.

Main Methods:

• Both MPC and RL were formulated as discrete-time optimal control problems.
• MPC utilized an analytical system model and an interior-point solver for open-loop policies.
• RL employed a model-free approach, inferring closed-loop policies from system trajectories and cost values via supervised learning.

Main Results:

• Experimental results were obtained for a nonlinear, deterministic electrical power oscillation damping problem.
• The study provides insights into the advantages and disadvantages of both MPC and RL.

Conclusions:

• Reinforcement learning demonstrates competitive performance against model predictive control.
• RL is a viable alternative even in scenarios where a precise deterministic system model is accessible.