Imitation Learning Optimizes Microgrid Energy Management
A new framework uses imitation learning to approximate mixed-integer Economic Model Predictive Control (EMPC) for microgrid energy management. The approach trains a neural network on offline trajectories from an expert EMPC, enabling fast real-time decisions without solving online optimization problems. This addresses scalability issues and variable solution times that hinder real-time deployment. The system manages fuel generators, renewables, energy storage, and curtailable loads.
Key facts
- Imitation learning approximates mixed-integer Economic Model Predictive Control (EMPC).
- Neural network trained on offline expert trajectories.
- Enables fast real-time decision making.
- Avoids solving online mixed-integer optimization problems.
- Addresses scalability and variable solution times.
- Manages fuel generators, renewables, energy storage, and curtailable loads.
- Published on arXiv with ID 2510.20040.
- Announcement type: replace-cross.
Entities
Institutions
- arXiv