These days study on a learning method for getting a dynamic characteristic model without the knowledge of an object in the field of a hierarchical supervisory control system has been actively being conducted. However, no method of applying it to existing production processes by speeding up Q-learning convergence has been proposed yet.
Kim Song Ho, a section head at Faculty of Automation Engineering, proposed a two-layer intelligent supervisory control system to adjust the online optimum setting point of the control system and set forth a new method to improve Q-learning convergence.
Unlike previous methods, the core of his suggestion is that it helps overcome trial and error Q-learning process, the weakest point of this learning, and achieve the fastest Q-learning convergence by setting optimally the way of automatically extracting experience rules of process operation from historical operation data and the initial phase of Q-learning.
His suggestion is estimated to have great practical significance in making large-scale continuous industrial processes unmanned and intelligent as it is supposed to minimize the effect of environmental changes and to guarantee product quality and stability of process operation.
He presented his essay titled “On-line set-point optimization for Intelligent Supervisory Control and Improvement of Q-learning Convergence” to SCI journals “Control Engineering Practice”.
© 2021 KumChaek University of Technology