| The silicon content in hot metal tapped from blast furnace is a very important factor which should be controlled in iron making process. It is not only a very important criteria in evaluating the quality of hot metal, but also an index of the thermal level in blast furnace. So, it is essential to keep its stabilization for the blast furnace manufacturing. However, existing models till now which used for the prediction and controlling of silicon content in hot metal couldn't meet the requirement to control such a complex processing of blast furnace. In this thesis, a new model used for predicting and controlling silicon content in hot metal based on Artificial Neural Networks (ANNs) and Expert System has been investigated by developing prototype of the model with software engineering methodology, after inquiring about the parameters which could affect silicon content in hot metal with the operators of blast furnace. And the model also be tested with the process data of blast furnace in Chongqing Iron & Steel Co.,Ltd.(Cisco).The predicting and controlling model of silicon content in hot metal tapped from blast furnace is composed of following components or sub-models: ANNs prediction model, Expert model and interface. After training, the ANNs sub-model could be used to forecast, at two-hour interval, the possibility of silicon content in the next two hours and figuring the non-linear mathematic relationship between silicon content and the other parameters as the foundation of the numerical operating guidance. Expert system is mainly used to analyzing the result of the silicon content prediction ANNs and giving blast operators some reasonable operating guidance which can insure the silicon content is in a certain range. In order to make the communication between the operators and software, a slinky and appropriate interface was also developed. As a result the prototype that was function satisfactory and user friendly, has developed with VB and Matlab commercial software. Because the numeric calculations are all done by ANNs, the software is able to transplanted and embedded in different blast furnaces and different software.Seven of input and state variables of blast furnace, i.e. silicon content, coal injection, blast temperature, blast volume, blast pressure, pressure difference, coke rate and gas permeability, are considered as input of the prediction model. When the designation of the software was finished, it was trained and tested by 57 sets ofpretreatment data (the process data of blast furnace No.5 in Cisco which were standardized and noise cut off). The ANNs were trained on 30 from 57 data sets and tested by the rest. The results show that when the prediction error is less than 20%, the hit rate of the ANNs is 96.29%, the prediction error is less than 15%, the hit rate of the ANNs is 85.18%, the prediction error is less than 10%, the hit rate of the ANNs is 66.66%, the prediction error is less than 5%, the hit rate of the ANNs is 37.37%. After training, the ANNs can be used to predicting the silicon content. When the predicted silicon content is out of the goal range, the expert system will be activated and through analyzing, it will give some reasonable operation guidance, which can maintain silicon content in the prescript range, to the blast furnace operators. These guidance can be qualitative or quantitative. It showed that in the same condition, the guidance made by expert system model made can match with the controlling strategy made by human expert well.Furthermore, a few promising ways to improve the prediction and controlling of model also have been discussed.
|
PDF Full Text Request