Data-driven generation of multi-modal control programs for continuos-discrete processes

MARIANO DE PAULA; ERNESTO C. MARTINEZ

Cracovia, Polonia

Congreso; European Symposium on Computer Aided Process Engineering 19; 2009

Silesian University of Technology

Multi-modal control is an increasingly used design tool for supervisory control of complex systems by resorting to sequences of modes each one comprising of a feedback law and stopping condition. In this paper the problem of developing multi-modal control programs from a given mode alphabet using data is addressed. By viewing the control space as a set of tokenized instructions rather than as real-valued signals, reinforcement learning algorithms becomes applicable to develop optimal control strategies for continuous-discrete processes using a Lebesgue-sampled finite state machine. A case study related to capacity management of a buffer tank in a petrochemical plant is presented.