An Analysis Of Blast Furnace Anomalies Condition Based On Data Mining

Adequate molten iron of high quality should be supplied from blast furnace Ironmaking for the subsequent working procedure of steelmaking. For iron and steel enterprises, ironmaking consumes 60% of energies as required by the whole procedures. Meanwhile, blast furnace is the significant equipment to which accidents happens frequently. Currently in China blast furnace anomalies are usually monitored manually. A small number of expert systems have been introduced from abroad, but such technology import is not only costly but also poor in transportability. This thesis was completed when its author was fortunate to participate in"Research and Application of Blast Furnace Anomalies Forecast", a research program for one of the large-scale iron and steel enterprises in China. The major research results are summarized as follows:①The effect and significance of blast furnace operating parameters are explored on the basis of practical blast furnace operation states. In order to solve the bottleneck problem of expert systems on knowledge acquisition, it is proposed that data mining be introduced into expert systems, as complementary to the later and making them work more effectively. The study also distills the eigenvalue of parameter changes, which come to be the data source for the knowledge and information database of blast furnace expert systems, from field investigation and statistical analysis of historical data of blast furnace operation.②CTP (Cross Temperature Device) demonstrates gas temperature changes of materials. Through analysis of the effect of blast furnace operation on CTP, the study distills eigenvalue as the reflection of channeling and simulates the structure change of CTP curve in order to prove the validity of algorithm given.③Clustering analysis of blast furnace operation parameters is carried out by K-Means Clustering. Analysis and comparison of practical data is conducted to determine the optimal cluster number and iteration number of this algorithm for blast furnace parameters analysis, and thus yield the ideal operating value for the parameters. The optimal threshold for blast furnace parameters is determined through statistical analysis, repeated experiments and field assessment, and the difference between blast furnace state as estimated and the practical one analyzed.④Extended Kalman filter, singular value decomposition method is introduced to adjust the weight of BP (Back Propagation) Neural Network, to effectively overcome the its shortcomings of slow convergence and easiness to fall into local minimum, as well as to enhance calculation speediness, filter numerical stability and the output precision of neural network. Simulation test proves the improved algorithm's higher convergence effectiveness and speediness.⑤A Cooling-Hotting Classified Recognition Model is established to predict blast furnace thermal state. It recognizes blast furnace cooling and hotting by BP neural network based on extended Kalman filter, singular value decomposition method, and complement blast furnace expert systems in their operation. The orderly change of blast furnace CTP curve is analyzed and 8 temperature points predicted by BP neural network which is based on extended Kalman filter, singular value decomposition method, in order to predict the happening of channeling. Results of simulation tests corresponding to classification and prediction shows that the classification and prediction by the algorithm in the present thesis is of higher precision, and matches the practical blast furnace operation states.