摘要：AbstractGiven the complexity and isolation of the blast furnace (BF), field engineers generally operate the system upon their former experience and the operating manual. Harsh environment and equipment shortage have made the testing of silicon content a prevailing method for the detection of temperature within BF. As the silicon content is a comprehensive performance of internal thermal state, knowing the exact value in advance can be very helpful for operators to keep the furnace temperature at a reasonable extent. Thus, an improved gated recurrent unit recurrent neural network (GRU-RNN) is proposed to predict the silicon content of hot metal, indicating a competitive performance at 92.4% hit rate among several deep learning methods.
