Online Controls of Mineral Processing Plants by Neurocomputing

1992 Fiscal Year Final Research Report Summary

• Project/Area Number: 02452226
• Research Category: Grant-in-Aid for General Scientific Research (B)
• Allocation Type: Single-year Grants
• Research Field: 資源開発工学
• Research Institution: The University of Tokyo
• Principal Investigator: INOUE Toshio The Univ. of Tokyo, Faculty of Eng., Professor, 工学部, 教授 (80010742)
• Co-Investigator(Kenkyū-buntansha): OKAYA Katsunori The Univ. of Tokyo, Faculty of Eng., Instructor, 工学部, 助手 (80134493) ; NONAKA Michio The Univ. of Tokyo, Faculty of Eng., Instructor, 工学部, 助手 (70010981) ; OKANO Yasuhiko The Univ. of Tokyo, Faculty of Eng., Associate Professor, 工学部, 助教授 (30011092)
• Project Period (FY): 1990 – 1992
• Keywords: neural network / process control / programming / back propagation / process identification

Research Abstract

The following items have been investigated in order to assess the potentiality of process control utilizing the learning function of artificial neural networks(ANN): (1) programming language environments fitting the ANN operations, (2) learning methods of the ANN, (3) process identifications by the ANN, (4) controls by the ANN and so on. The details for each item are as below. For (1), a programming language has been successfully developed,which can process a large amount of data and is also provided with high speed array operators. Concerning (2), the calculation formulae for the hierarchical ANN have been arranged and unified. At the same time it has been found that the serial moment method is effective in ANN learning, considering the options for the parameters as well as the learning methods of the ANN. About (3), the two types of research have been conducted; a trial in direct learning of process dynamics and a development of the ANN for estimation of process parameters. In the former, however, any stable solution has never been obtained and there remains a future problem. In the latter almost favorable results have been given except under the disturbance by observation noises. In consideration of (4), the two methods have bee investigated, one of which learns characteristics of the conventional controllers and the other utilizes an inverse model of the process considered. The first method, however, could yield no satisfiable results. In the second one, an optimal control could be attainable with a suitable reference model. Some pieces of useful information as well as some practical methods have been thus obtained for online controls of mineral processing plants by neurocomputing.