Structured PENN Controller for MIMO Process

JOURNAL OF CHEMICAL ENGINEERING OF JAPAN

Masaru Ishida¹  Takehiro Ohba¹ 

[1] Research Laboratory of Resources Utilization, Tokyo Institute of Technology

关键词: Process Control;    Neural Network;    MIMO Process;   

DOI  :  10.1252/jcej.31.545

来源: Maruzen Company Ltd

【 摘 要 】

References(5)This paper presents a neural net controller for a multi-input, multi-output (MIMO) process. The controller is based on the PENN (Policy- and Experience-driven Neural Network) method. The PENN uses two types of knowledge for learning: local experiences obtained from the control results; and global policies based on rule-based control strategy. The structured PENN that is proposed for a MIMO process in this study has a structure in which small controllers for SISO processes are assembled. By use of such a structured assembly, both learning and control for a MIMO process can be accomplished. To prevent the effect of noises observed in the controlled variables, a noise reduction mechanism modifying the error function in the learning algorithm is proposed. The simulation results for a crystal growth process indicate that the proposed controller has the ability to learn the interactions between control variables and to get satisfactory control results.

【 授权许可】

Unknown