New machine learning framework optimizes industrial energy system design by bridging the gap between architectural planning and real-world operation while reducing costly simulations

Researchers developed an online ML-accelerated optimization framework that handles multiple model fidelities, from system architecture to dynamic operation

The framework estimates upper bounds on system performance while minimizing expensive high-fidelity model evaluations

Tested on a pilot energy system designed to supply a 1 MW industrial heat load

Combines multi-objective architecture optimization for component sizing with ML-accelerated receding-horizon optimal control strategies