Lyapunov-based economic model predictive control for online model discrimination

Abstract

Economic model predictive control (EMPC) is a flexible control design strategy that can be modified to achieve many operating goals while also ensuring safe operation (e.g., by adding Lyapunov-based stability constraints to form Lyapunov-based EMPC, or LEMPC). Prior works have investigated LEMPC capabilities for achieving goals online beyond optimizing process economics, including aiding in model structure selection to benefit model-based control system design since the accuracy and quality of the process model are important for achieving an expected performance from such systems. This work further probes the capabilities of LEMPC to accomplish multiple objectives during process operation, including aiding in the discrimination between mechanistic models online. In particular, several rival mechanistic models may explain the existing data. To discard models from this set that do not fully represent the actual process, a new set of “online experiments” can be conducted to collect more information. However, additional experimentation may be costly and unsafe to be performed. LEMPC can aid in performing online data collection when discrimination between mechanistic models is needed, with the flexibility to ensure safety as the data is gathered and trade off the data-gathering goal for cost considerations. Motivated by this, we discuss how LEMPC can be designed to automatically and dynamically collect data that is useful for the selection of mechanistic models from among a set of possibilities. A chemical process example is used to clarify benefits and limitations of LEMPC for promoting online model discrimination.