An Operator Behavior Model for Thermal Control of Blast Furnace

Abstract

To realize precise thermal control of a blast furnace, an operator model that imitates the behavior of skilled operators was developed using a convolutional neural network (CNN). Conventional thermal control systems suffer from large control errors when large disturbances occur, e.g., when low-quality materials are used. Despite these adverse conditions, capable operators are still able to take appropriate control actions by making the best use of sensor information. Such operators’ control actions were simulated by the CNN, and the validation results showed that the accuracy of the developed operator model was 71%. The operator model was then incorporated in the operation guidance system at actual furnaces. It was found that the operator model can cope with the severe situations where the material characteristics change or abnormal descent of the burden materials occurs.