Research On Predicting Model Of Burning Through Point Using Artificial Neural Network

Sintering provides raw materials for blast furnace. Quality, production and energy consumption in iron-making are directly influenced by the production and quality of sintering ores. The state of burning through point (BTP) is the end of the sintering process, which is one of the important parameters in judging the state of sintering process. The prediction of BTP is a multivariable, nonlinear and longtime-delay problem, and neither traditional modeling by mechanism analysis nor conventional control theory can perform the task of prediction and control of BTP. The artificial neural network (ANN) has many advantages, such as self-adaptation, self-organization, parallel computation and associative memory, which make ANN particularly suitable to describe uncertain and complicated nonlinear system. Therefore, this paper attempts to predict BTP by modeling of ANN.Based on a lot of researches both domestic and abroad on correlative literature, the paper introduces the current status of BTP prediction firstly; then it analyzes the basic principles of sintering and the basic theory of ANN; four processing parameters, which can be monitored online, are determined by analyzing the sintering process, including the suction pressure of main chimney flue, air input, velocity of the sintering machine and ignition temperature.Here prediction models are built for BP NN and RBF NN respectively. 95 groups of data are collected from plant, and 70 groups of them are used to train networks separately as training sample in all data, while the rest are used to predict. The results show that RBF model has a distinctive advantage over BP model in network practice speed and accuracy of prediction.At last, the paper proposes an online training method, and 30 groups of serial data collected from plant are used for simulation test. The result shows that the method is effective, indicating the potentials of sintering process.