Research On Data Analytics Based Production Parameters Prediction For Raw-Material And Finished-Product Stages In Steel Production

Accurate prediction of operation parameters can provide an important basis for the iron and steel enterprises to achieve meticulous production.The raw materials stage and the cold-finished-product stage are important head and tail production processes respectively.The sophisticated raw material and fuel supply operations in raw material production stage can guarantee sufficient material for the subsequent production unit and make the production process smooth.While,the sophisticated production operations in the finished-product stage can help steel plant improve production efficiency and reduce production cost.This thesis studies prediction problems for two key operation parameters in the raw-material and finished-product stages,respectively.The first one is the demand prediction for raw materials and fuels,and the other one is the cycle prediction for cold rolling production process.By using least squares support vector machine and the estimation of distribution algorithms,the prediction model for the raw materials and fuels and the prediction model for the production cycle of cold rolling production process are built.Finally,based on the production cycle prediction,a performance evaluation system for cold rolling process is designed and developed.The research can provide scientific basis for production operations management,and hence is meaningful for improving the sophisticated management of raw-material feeding operations and finished-product production operations.The main outlines are summarized as follows:(1)Taking raw material and fuel purchase process for the blast furnace production as a background,the demand prediction problem for the raw material and fuel is studied.First,the data are preprocessed by interpolation and normalization methods,and then the input variables are selected by making correlation analysis.Using the least squares support vector machines,the demand forecasting models for raw material and fuel are built respectively.Then,the distributed estimation algorithm is adopted to optimize the parameters of the models.Experimental results show that the average relative error of the raw material and fuel demand predicted by our proposed model and algorithm is less than 3.3%,and it is satisfactory and can provide a reliable basis for procurement of raw materials and fuels.(2)Taking the cold-finished-product production process as a background,the production cycle prediction problem is studied.Based on the data preprocessing,the input variables are selected by making correlation analysis.Using the sparseness least squares support vector machine,the production cycle prediction model for each production unit is built.Then,the estimation of distribution algorithms is adopted to optimize the parameters.In the estimation of distribution algorithms,the sampling probability in the last iteration is considered to improve the efficiency of the algorithm.Experimental results show that the average prediction error of the production cycle for each production unit predicted by the proposed model and algorithm is less than 1 day,and the proposed improvement strategies can significantly improve the prediction accuracy.(3)Based on the practical requirement,a performance evaluation system for cold rolling process is designed and developed.Based on the production cycle prediction result,the system calculated the key performance indicator of stock turn cycle.It can help planner find out production bottleneck and adjust the production line.In addition,the system can also evaluate the staff’s performance,and hence improve the production and management efficiency.