Reinforcement Learning using Gaussian Processes for Discretely Con-trolled Continuous Processes

MARIANO DE PAULA; ERNESTO C. MARTINEZ

Oro Verde

Congreso; XIV Reunión de Trabajo de la Información y Control - RPIC 2011; 2011

Unversidad Nacional de Entre Ríos, Facultad de Ingeniería, Bioingeniería y Bioinformática (UNER)

In many application domains such as autonomous avionics, power electronics and process systems engineering there exist discretely controlled continuous systems (DCCSs) which constitute a spe-cial subclass of hybrid dynamical systems. In this article, we introduce a novel simulation-based ap-proach for DDCSs optimization under uncertainty. In the proposed approach we use Reinforcement Learning (RL) with Gaussian Process models (GP) to learn the transitions dynamics descriptive of mode execution and an optimal policy for mode switching. The proposed approach is based only a few samples obtained from the interaction with the real system and mostly relies to simulated experience. By Baye-sian Inference with GPs the uncertainty in predict-ing state transition following mode execution is ac-counted for in the design of a multi-modal control program. Throughput maximization in a buffer tank subject to an uncertain schedule of several inflow discharges is used a case study to address supply chain control in manufacturing systems.