My research studies the optimal control of flexible energy assets and energy-cost-driven production systems, such as batteries and industrial process lines, under market and physical uncertainty.
In industry, there is an increasing interest in the smart control of flexible assets and energy-cost-driven production systems. They generate more often their own (renewable) energy. Because both this generation and the energy prices vary, the moment at which energy is consumed matters. By managing flexible assets such as batteries and production processes intelligently, companies can use more of their own renewable energy, shift consumption towards cheaper moments, and avoid wasting surplus generation. This can lead to substantial savings.
Controlling these assets well requires decisions under uncertainty, subject to physical and operational constraints. Operations research methods explicitly model these constraints and provide interpretable, reliable solutions, but become difficult to apply under high uncertainty or at larger scale. Reinforcement learning handles uncertainty well and yields fast decisions once trained but offers few guarantees on feasibility and interpretability. This research investigates hybrid approaches that combine both, aiming to retain the strengths of each.
