Task-optimal data-driven surrogate models for eNMPC via differentiable simulation and optimization
This work addresses the challenge of real-time capable predictive control for dynamic processes, offering an incremental improvement over existing methods by leveraging differentiable simulation for task-specific optimization.
The paper tackles the computational expense of mechanistic models in real-time predictive control by introducing an end-to-end learning method for Koopman surrogate models optimized for specific control tasks, achieving similar economic performance while eliminating constraint violations in a CSTR case study.
Mechanistic dynamic process models may be too computationally expensive to be usable as part of a real-time capable predictive controller. We present a method for end-to-end learning of Koopman surrogate models for optimal performance in a specific control task. In contrast to previous contributions that employ standard reinforcement learning (RL) algorithms, we use a training algorithm that exploits the differentiability of environments based on mechanistic simulation models to aid the policy optimization. We evaluate the performance of our method by comparing it to that of other training algorithms on an existing economic nonlinear model predictive control (eNMPC) case study of a continuous stirred-tank reactor (CSTR) model. Compared to the benchmark methods, our method produces similar economic performance while eliminating constraint violations. Thus, for this case study, our method outperforms the others and offers a promising path toward more performant controllers that employ dynamic surrogate models.