An Improved MST Based Key Feature Extraction Method And Its Application On Finish Rolling Temperature Modeling

With the rapid development of steel industry,the quality and performance of steel are becoming more and more important.Hot rolled strip is an important one among the steel products.During the hot rolling production,finish rolling temperature is a key factor which influence the performance and organization of strip steels.In order to predict the finish rolling temperature accurately,it is necessary to construct the prediction model for the finish rolling temperature.However,there are so many kinds of factors which affect the finish rolling temperature and couple with each other.In order to decrease the complexity and increase the prediction accuracy of the prediction model,it is important to select the key factors which affect the finish rolling temperature.Therefore,an improved minimum spanning tree based key feature extraction algorithm for high-dimensional data is proposed in this paper,which can remove the irrelevant features and the redundant features in one extraction process.In this paper,a minimum spanning tree(MST)based key feature extraction algorithm is introduced.In order to deal with the single-point-divergence problem which emerged in the process of processing the actual finish rolling process data,the algorithm is improved.Then the improved algorithm is adopted to extract the key features which influence the finish rolling temperature from the actual finish rolling process data.Then a method to construct the finish rolling temperature prediction model is proposed by combining the improved feature extraction algorithm and the support vector regression.The prediction effect of the prediction model is satisfactory after the parameter optimization.Finally the support vector regression model is compared with back propagation neural network model.