Prediction of silicon content of hot metal using fuzzy-GA regression
Punya Sheel, Brahma Deo · 1999
The use of fuzzy regression model is generally recommended for industrial processes in which both input and output parameters are fuzzy in nature. However, the application of fuzzy regression (using linear programming) to optimize the regression coe cients in our problem of silicon prediction in blast furnace hot metal showed large deviation from actual values at lower and upper bounds, inspite of the fact that both correlation coe-cient and standard deviation for entire data were acceptable. Analysis showed that linear programming procedure used in fuzzy regression is unable to take care of this anomaly. Therefore, in the present work, a fuzzy-GA model has been developed and it has been found that performance of Fuzzy-GA regression model is far superior to simple fuzzy regression. The spread of fuzzy coe cients obtained by fuzzy-GA regression is reduced signi cantly with the use of GA in the global search for optimized coe cients. It is recommended that simple fuzzy regression be replaced with fuzzy-GA regression for control and prediction in industrial unit operations which generally show a fuzzy behaviour. 1