Research Of Blast Furnace Gas Flow Prediction Algorithm And Its Application In Furnace Condition Analysis

Blast furnace(BF) ironmaking is a key link in the process of iron and steel smelting. The stable and smooth BF status is a guarantee for the optimal BF operation. The BF gas carry abundant and timely information, which is the important reference for the BF operation. Therefore, BF gas flow information can be used to analyse the BF condition, which is valuable in BF condition adjustment. In the light of BF condition analysis problem in ironmaking production, the gas flow is studied, and the “Blast furnace gas flow prediction and operation auxiliary analysis system” is developed using bayesian techniques, particle swarm optimization algorithm, echo state network, expert knowledge methods. The system has been tested on the #2BF, Liuzhou iron and steel company. The prediction results show that the system can predict the trend of BF gas flow accurately. This system can offer help for BF condition analysis and prediction, and ensure the BF steady running. In this paper, the main research work and conclusions are presented as follows:Firstly, the characteristics of gas data and the BF condition are deeply studied, and the relationship between the gas and the BF condition is established. Based on the grey correlation algorithm, compute the degree of grey correlation in the indexes and select the ones that have significant effect on gas abnormal BF condition. This study provides the basis for the prediction model in the paper.Secondly, aimed at the disadvantages that the echo state network(ESN) model easily falls into ill-condition in the blast furnace gas flow prediction, two kind of improved ESN model are proposed. When the training sample is small, particle swarm optimization based on Bayesian method is proposed, and applied in ill-posed ESN model. When the training sample is large, the L-curve method is applied to ill-posed ESN. Two methods are effective in relieving the ill-condition of the ESN model. Experimental results of BF bosh index show that they have very good training precision.Then, in view of the disadvantages that internal structure of ESN is unknown, Improved Hammerstein model are proposed. The Bayesian technology is used to identify model parameters and order in Hammerstein model. The markov chain monte carlo(RJMCMC) algorithm is employed in approximating solution of model parameters Then the amount of calculation and computational complexity are reduced greatly in the algorithm. Simulation results of the four top gas flow indexes show that the method have very good training precision and stability.Later, consider the disadvantages that the gas flow information is not used sufficiently, an adaptive block orthogonal matching pursuit(AOMP) algorithm is proposed and the rules number is computed more reasonably. The consequent parameters of the T-S fuzzy system are identified using the bayesian techniques, which reduces the complexity of the model meanwhile maintain the accuracy of the prediction. Seen from the prediction of wind pressure, the model improves the ability to resist noise in prediction. Simulation results of two classic data set and wind pressure show that the model has good ability in prediction and noise immunity.Finally, on the basis of the above theory research, the “Blast furnace gas flow prediction and auxiliary analysis system” is developed. In Liuzhou iron and steel company, the experiment results show that the system can provide real-time prediction of the BF gas flow indexes, the prediction accuracy has been affirmed by the technical staff. The prediction results have important practical value to analyze BF condition further.