Research Of SVM To Predict Silicon Content In Iron Smelting Process

The process of ironmaking is highly complicated, whose operating mechanism is characteristic of nolinearity, time lag, high dimension, strong noise and distribution parameters, etc. As the Committee of National Natural Science Foundation of China, Automatic Science and Technology-The Research Report on Development Stratagem of Natural Science Subject pointed out, modeling and control of the complicated ironmaking system is a foreland of corrent development of automatic science and technology. Meanwhile, it is also an important and difficult issue of automatic system of the ironmaking process. The prediction model of silicon content in molten iron is the kernel mathematic model of this automatic control system. The key problem is to improve the accuracy of prediction. This dissertation focuses on building the prediction model based on support vector machine technique. The contributions are listed as follows:1. Having read lots of literatures (both English and Chinese), the knowledge of ironmaking process is established. The correlation between silicon content in iron and other variables is analyzed, the time lag these variables is also discussed. Besides, a review of support vector machine and independent component analysis is given. Detail descriptions of support vector machine and independent component analysis are also shown in chapter 2.2. Based on the modified dynamic independent component analysis (ICA) and the support vector machine (SVM), a novel modeling method for prediction of silicon content in iron smelting process is proposed. In order to eliminate the correlations of production variables, the dynamic ICA is used for feature extraction. A dynamic recursive model is then built for prediction of silicon content, using least square SVM which has low computational complexity. An application study is carried out on real process data. The experimental result shows that the prediction-hit-ratio of our proposed method is greatly improved compared with the other existing methods3. An incremental support vector machine for online modeling of silicon content in iron is proposed to solve the slow time-varying characteristic of the ironmaking process. Traditional methods always become invalide when process changes. The proposed method can improve the adaptive ability of the prediction model. Therefore, the prediction of silicon content in iron can adaptively change corresponding to thechange of process operating modes. Simulation results show feasibility and efficiency of the proposed method, the prediction ability is greatly improved.