Reproduce “Hybrid Reinforcement Learning and Model Predictive Control for Adaptive Control of Hydrogen-Diesel Dual-Fuel Combustion” on OpenResearch

About this paper

Julian Bedei, Murray McBain, Alexander Winkler, Charles Robert Koch, Jakob Andert, David Gordon

Researchers from RWTH Aachen University and the University of Alberta developed a hybrid control framework combining Reinforcement Learning (RL) with Machine Learning Integrated Model Predictive Control (ML-MPC) for hydrogen-diesel dual-fuel engines. This approach improves load tracking accuracy by adapting to system drifts, such as simulated injector aging, while ensuring safe operation.

Key insights

• Combining RL's adaptive learning capabilities with ML-MPC's precise, constraint-aware control effectively mitigates the performance degradation caused by model-plant mismatch in a real-world engine.
• The ML-MPC's ability to enforce safety constraints is crucial for enabling safe deployment of RL in safety-critical environments, preventing hazardous conditions during the RL agent's exploration.
• The RL agent can learn to introduce compensatory offsets and an 'overshooting' feature to the ML-MPC reference, significantly improving load tracking under changed operating conditions.

Summary by alphaXiv.