Heat level is one of the important factors influencing the stable operation
of blast furnaces, and it is especially important to accurately forecast d
ecreasing heat levels in order to stabilize the heat level.
A forecasting model for decreasing heat levels which occur accompanied with
a sudden rising of wall temperatures has been developed using neural netwo
rk technology. Wall temperatures are measured at various points in the vert
ical and circular directions. Temperature rising points are measured as a d
istributed pattern, and neural network technology is used in order to recog
nize this distributed pattern.
Neural network models are classified into two groups according to their lea
rning style, one is called the supervised learning model and the other, the
unsupervised learning model. The operators notice that a decrease in heat
level sometimes occurs after a rise in wall temperature, but there is no kn
owledge of what patterns cause the heat level decrease, which means there i
s no teaching data for the supervised model. The forecasting model is built
using one of the unsupervised neural network models, the self organization
feature maps model, which recognizes and classifies the wall temperature r
ising patterns. A new method of shift invariant recognition has been develo
ped in order to put circularly shifted wall temperature rising patterns tog
ether in a class.
It has been established that the heat level forecasting model using the cla
ssified wall temperature pattern gives better forecasting accuracy for heat
level decrease than a forecasting model using the total amount of wall tem
perature rising points. Furthermore, this heat level forecasting model, whi
ch uses a classified wall temperature pattern and solution loss C, has suff
icient accuracy for heat level operation guidance.