Research Of The Statistic Pattern Recognition-neural Network Optimization Program In The Metallurgical Process

In this thesis, the artificial intelligence optimization software in the process of iron and steel making was developed. Considering of the features of metallurgical process and basing on the MATLAB language circumstance, several common-used methods for pattern recognition were employed, which were primary characteristics analysis(PCA), partial least squares(PLS), optimal discrimination plan(ODP) and shared K-nearest neighbors(SKNN).Artificial neural network technique was also used. Taken the practical producing hot-roll silicon-steel plate as optimization research object, the developed software was validated.The research technique procedure was: pretreated the collected sample data, gained the classified maps of PCA,PLS and ODP and discussed the qualitative analysis based on the furthest optimal map of the three, then quantitative analysis was made with class centers of good data and bad ones calculated by SKNN method; lastly, artificial neural network was employed to predict the optimized values and validate the acquired optimized project.The developed software was designed with modular structure. It was readable, extensible and easily adjusted and tested.The study showed that the employed method in this thesis, from the qualitative to quantitative analysis of optimization, and then using neural network to predict andvalidate optimized project, was scientific, reliable and practicable for the metallurgical process which with multi-affected factors and complicated reactive mechanism.