Research On Performance Prediction And Technological Design Of Iron And Steel Product Based On Data Mining

As a big country of iron and steel manufacturing,the technology and quality of iron and steel products in China are still lagging behind of the developed countries.It's urgent to optimize the product mix and improve the product quality of iron and steel products.With the continuous improvement of the information construction of iron and steel enterprises,and the continuous development of information technology,it has become a trend to realize the process of deep optimization and quality control by means of modern information technology.According to the problems existing in the present research,this dissertation proposes a research system for the storage and analysis of the quality data of the whole process of iron and steel enterprises,which is driven by actual production data,including performance prediction,composition system design and data warehouse model design,etc.The main contents and innovation works of this dissertation are as follows:(1)On the basis of the traditional mechanical performance prediction model,a genetic neural network prediction model with high-dimensional multi-input layers is proposed,in accordance with the multi stage characteristics of the hot rolling production process.In order to construct the multi-input layers neural network,the main process parameters are distributed into four input layers,corresponding to the four stages of hot rolling production process.The high precision network connection weights are obtained by using two step training method which combines the genetic algorithm and BP algorithm.The mapping between components,process parameters and mechanical properties of hot rolling product is obtained,and the prediction model has high accuracy and versatility.(2)According to the attributes and inherent links of process parameters in hot rolling process,an ELM prediction model based on attribute reduction is proposed.In view of different kind of steel,the model uses the attribute reduction method,which combines information entropy and Gram-Schmidt orthogonal transform,to form an effective attribute feature set.It reduces the dimension of the process parameters.And then a corresponding ELM prediction model is established.Compared with the traditional ELM model and the other methods,the structure of the ELM prediction model based on attribute reduction is simpler and the prediction accuracy is higher.(3)In view of the demand of multi variety,small batch and cusomization in current production,and the problem caused by management of steel grades in steelmaking process,a DBSCAN clustering algorithm improved by grouping algorithm is proposed.The grade and chemical composition of iron and steel products are clustered to realize the integration of steel species driven by actual production data,in order to design a more adaptable steelmaking composition system.It can guide the flexible rolling and customized production,and also improve the management difficulties caused by various kinds of steel.(4)On the basis of the characteristics of the production process and its data of iron and steel enterprises,the data storage model and the corresponding OLAP analysis method are designed for the purpose of quality analysis.The model is for the whole process analysis of iron and steel enterprises.The data warehouse model,throughout the entire life cycle of the production process,can achieve the multi granularity analysis of the quality and heritability of steel products,as well as the needs of information preprocessing and multi angle observation.