Detective And Predictive Research For Blast Furnace Burden Surface Shape

As a pillar industry of our national economy and the important foundation for the other industries, iron and steel industry plays an important role in the social development and economic construction. Blast furnace is the most critical equipment in the process of iron and steel production, which is related to steel production, energy consumption and environmental pollution.The common goal of the iron and steel industry is to maintain the long-term stability and efficient working of the blast furnace.Burden surface distribution in the throat is one of the major factors on BF conditions, which keeps distribution reasonably for gas flow, increases permeability and optimizes operation. In view of the deficiency of tightness in detecting burden surface states, this dissertation develops an "detective and predictive research for burden surface shape" by employing the technique of radars detection, burden surface formation mechanism, swarm intelligence and intelligent computing.Optimal deployment of radar sensors, burden surface detection, prediction model of surface's dropping speed and fault diagnosis are studied in depth, which have guiding significance and application value for a smooth running and energy saving as well.The main research and innovation are included in the following four aspects:(1) An optimal method of radar sensor installation location in blast furnace was put forward. The method can take into account other types of sensors have been installed to achieve unified information. At the same time, in the premise of using the least number of radars, the surface can be covered completely. And, the key points of blast furnace can be K-covered. On the one hand, the annular region of blast furnace was established according to the relationship between the furnace throat radius and diameter of radar coverage, which can reduce the number of radar installation and accelerate optimization speed. On the other hand, the evaluation function of deployment optimization was proposed according to the analysis of surface shape features.The improved artificial fish swarm algorithm was used for solving optimization problem. Finally, the validity of the method is verified by the standard test functions and actual data.(2) A model frame of reconstructing burden surface shape in blast furnace was put forward. The reconstructing model makes use of disciplinarian of burden distribution and multi-radars. Firstly, the description equation of burden surface shape was proposed, and stack angle and heap top position of burden surface were computed according to mechanics principle. The burden surface equation is established. Secondly, the real-time height data collected by radar sensors is extracted, which is used for modifying burden surface equation by means of multi-source information fusion.The coefficient of burden surface equation are calculated iteratively based on the pricinple of the volume constraint.After the test compared with other algorithms, the proposed reconstrcuting model has high precision and the burden surface equation is more reasonable.(3) Precdiction model of burden surface descent speed was proposed. Firstly, the phase space is reconstructed according to time delay and embedding dimension those are calculated by C-C algorithm. The chaotic characteristics of time series about burden surface height are proved with largest lyapunov exponents computed by small-data method.Secondly, the prediction model of online and offline are created with extreme learning machine and online sequential extreme learning machine respectively. Finally, the results from experimental conducted on the actual production data have show that the algorithm is efficient and has a better performance over the same king of algorithms. The chaos method in predicting burden surface descent speed is feasible.(4) Unbalanced sample classification model of burden surface fault was proposed. In view of the rare number of fault samples and unbalanced distribution of training samples, the two stages of samples selection and establishing the base classifier are improved respectively. The classification model was built using the proposed algorithm. The results from experimental conducted on the strandard test sets have show that the algorithm is efficient and increases predictive accuracy of minority class samples. The classification model was established by using the actual production data has higher classification accuracy in burden fault samples.