Operational decision systems / Learning and control
Real-Time AI and Model Predictive Control
Combining machine-learning predictions with model predictive control and constrained optimization.
Problem statement
Translate evolving measurements into decisions while accounting for uncertainty and constraints.
My contribution
Design of machine-learning-augmented model predictive control for real-time operational decisions.
Technical challenges
Domain-level considerations, without implementation-specific details.
- Connecting predictions to decisions rather than treating them in isolation.
- Accounting for model uncertainty and changing conditions.
- Evaluating decisions under the constraints of real-time operation.
Methodology
Use machine learning within a model predictive control formulation to support repeated, constrained decisions.
- Use measurements to inform a predictive model.
- Combine predictions with an explicit control objective and constraints.
- Reconsider decisions as new feedback becomes available.
Conceptual architecture
Measurements
Prediction
Constrained decision
Feedback
New measurements inform subsequent predictions and decisions.
Technologies and methods
- Machine Learning
- Model Predictive Control
- Optimization