The Research And Implementation Of Production Prediction And Controlling Based On Data Mining Algorithm

At this stage, our country's steel industry is developing rapidly. Considering the production we are on the international advanced level. However, both the accuracy of product and the production efficiency are still a wide gap with the world advanced level. To meet the market demand, improve competitiveness in recent years, major companies are all focusing on producing high quality and high precision steel products. The product of hot rolled steel is an extremely complexity production process in which there is complex changes in physical changes and phase transitions. Hot rolling production process includes slab heating, rough rolling, finish rolling, cooling and coiling. In these processes, the factors impacting the performance result are in each one.The purpose of this paper is seeking for a suitable data mining algorithms and playing a guiding role in hot rolled steel production. The data used in the experiment are all from the actual manufacturers process including data from the steel slab heated to final product process with all information hidden in production.In this paper, we analyze the production data of hot rolled steel with the perspective following and reach the aim that prediction the performance of hot rolled steel and analysis the production data based on timing:1,First of all, we propose to use BP neural network to analyze data in order to predict the production performance. It has good nonlinear approximation ability in dealing with complex nonlinear data. But considering that BP neural network also has its limitations, there are issues still not been solved such as how to determine the number of hidden layer nodes, which training method to be used and longer training time.2,Since the drawbacks of BP neural network, we introduce the radial basis neural network in the area of steel production analysis, as well as the method that combine the radial basis neural network and community discovery algorithm. The general structure of radial basis function neural network is similar to BP neural network, which is network formed by multi-layer neurons connected through weights. The mainly difference between the two network is that the radial basis function neural network uses cluster analysis methods to determine the hidden layer nodes. With this method, it can overcome the disadvantage of BP network such as not stable enough, slow training time. This paper mainly introduces the community discovery algorithm method to determine the hidden layer nodes with the purpose of achieving higher prediction accuracy.3,Consider that those two neural network methods are not considering the time factor. But the process of hot rolling is sequence process. So we use the time sequential pattern mining algorithm to analyze the various production factors on product performance.