Intelligent Control Based on Reinforcement Learning for Batch Thermal Sterilization of Canned Foods

S. SYAFIIE; C. VILAS; M. GARCÍA; A. A. ALONSO; E. C. MARTÍNEZ; F. TADEO

Seúl, Corea del Sur

Congreso; IFAC World Congress; 2008

International Federation of Automatic Control

A control technique based on Reinforcement Learning is proposed for controllingthermal sterilization of canned food. The proposed Model-Free Learning Control (MFLC)has the objective of maintaining temperature in the first part of the process (heating up byflowing or manipulating saturated steam), by learning without using a mathematical modelof the process. From the defined state-action space, the MFLC agent learns the environmentby looking up of the tabular state-action mapping. The results show the advantages of theproposed technique for this kind of processes.