Deep learning prediction modeling of blast furnace condition based on principal component analysis of temperature field

The development of in-depth information mining and modeling of blast furnace process big data is an important content of blast furnace informatization and intelligent construction. Aiming at the shortcomings of traditional shallow neural network in the representation of blast furnace condition information,a logical model of furnace condition parameters was constructed with the furnace condition temperature field information mining as the center. The inputs of the model are composed of temperature field,operation and index parameters. A set of 108-dimensional temperature field data was reduced dimension by principal component analysis method,and a 20-dimensional principal component characteristic parameters was obtained under 86% information extraction rate. The instantiation of logic model forms three models, namely convolutional neural networks ( CNN), long short-term memory ( LSTM) and CNN-LSTM hybrid model. The results show that among them CNN-LSTM model has the best effect,and the gray correlation degree of the prediction result reaches 0. 89. The furnace condition prediction model constructed by deep learning is helpful for the information analysis of furnace condition big data.