Research For Multi-scale Feature Of BF And Prediction Of Silicon Content

Blast Furnace (BF) ironmaking, which is the foundational industry of the national economic development, has been played an important role in the progress of steel industry and energy-saving. During the production process, the change of hot metal silicon content is usually used to reflect the variation of temperature in BF. Hence, accurate prediction of silicon content is conducive to controlling the temperature and maintaining stable operation of BF.Based on the dynamic characteristics and multi-scale feature of blast furnace ironmaking process, this paper centers around the prediction of silicon content, which has certain clinical value. In view of the dynamic characteristics, a recursive neural network-Elman neural network is selected; while for multi-scale features, Hilbert-Huang Transform is used to analyze the time series.Firstly, the multi-scale characteristics of ironmaking process is identified using the Hilbert-Huang Transform. After empirical mode decomposition and Hilbert transform, the time series of silicon content is decomposed into a series of intrinsic mode function components and a residual, the frequencies of which lower gradually and the average instantaneous frequencies differ from each other. The result shows that the blast furnace ironmaking process is a multi-scale process.Based on the analysis of the multi-scale feature, a multi-scale based model-the improved EMD-Elman neural network model is proposed for the prediction of silicon content. Firstly, time series is decomposed into a finite number of components by EMD, obtaining relatively stationary sub-series from original data set. Second, for each component of the output variable, Elman neural network is set up. To further improve the accuracy of prediction, the result of each sub-model is multiplied by a weight and then summed up to obtain the final silicon content. Here, all the weights are optimized by particle swarm optimization (PSO). Models based on process variables and quality variable are discussed respectively, which were applied to the prediction of silicon content in a steel mill and obtained higher hit rate. The results proved the effectiveness of the proposed method.