Fast Time Series Similarity Matching And Its Application Research In Molten Iron Silicon Content Modeling

Molten iron silicon content can reflect the heat state in blast furnace and the quality of molten iron. It is very important to set up silicon content predicting model for monitor the running state of blast furnace and improving the quality of molten iron. At present, the modeling method based on Neural Network learning, which has been widely applied in practice, has many advantages compared with traditional pure mathematics model and rule based reasoning model. However, Neural Network has the disadvantages of over-fitting and depending on experience excessively, which make it limited in molten iron silicon content prediction, especially in the condition of large fluctuation. In recent two years, Support Vector Machine(SVM) was attached importance in molten iron silicon content modeling. The SVM based predicting models have been presented. But its accuracy and efficiency need to be improved further. For these problems, the thesis proposes a molten iron silicon content predicting model based on time series similarity matching, which can enhance modeling speed while ensuring precision.Firstly, the thesis proposes a Supporting Vector Regression(SVR) modeling method based on time series similarity matching. It chooses the matching historical data to predicted data as training set, which effectively removes the influence of inconsistent historical data and enhances modeling accuracy.Secondly, for the problem of low efficiency of the method, the thesis proposes a fast time series similarity matching algorithm based on bit operation, which greatly improves the efficiency of time series similarity matching and modeling speed while ensuring stated precision.Lastly, in order to decrease precision loss, the thesis proposes a dynamic time twisting algorithm based on bit operation, which maintains high model accuracy while improving efficiency.The experimental results show that the SVR modeling method based on time series similarity matching is effective for molten iron silicon content prediction. The fast time series similarity matching algorithm and dynamic time warping algorithm based on bit operation greatly enhance modeling efficiency, while ensuring model accuracy.