Research On The Method Of Soft-sensor For Mill Load Based On SVM

The monitoring of ball mill load is a very important part of the process of pulverizing. However, it’s impossible to keep the ball mill long-term stability running in the optimization area because of the difficult detection. In this paper, the current widely used double-ball coal mill in domestic power plants will be researched. In the second chapter describes the Double-ball coal mill structure and its working principle have been studied, and the process variables associated with mill load were selected as the pre-selected set of auxiliary variables. In the third chapter describes the soft-sensor model structure and it’s implementation process. Covers the secondary variable selection method proposed the combination of space-based reconstruction and chaotic combination of correlation analysis method to determine the auxiliary variable of the soft-sensor model. Analysis the correlation size of auxiliary variables and dominant correlation among variables through the analysis of correlation, and then based on the chaos space reconstruction methods to determine the auxiliary variable dimension. From the easily measured variables related with the mill load through the way of mechanism analysis got, through the combining of the correlation analysis and the mechanism analysis ultimately determine the soft measurement model of auxiliary variables. Then introduced the auxiliary variable data preprocessing methods, and finally wavelet noise reduction method is selected as this soft sensor data pre-processing method. Based on the Study of the basic theory of Support Vector Machines, make an analysis of the support vector regression used to solve the forecasting problem of the soft-sensing model. Using the soft-sensing mode based on SVM to perform an off-line modeling of for the Pulverizing System history data in a thermal power plant. After several tests and comparison, the nuclear function and other parameters of the model are determined. Through the training of the historical sample data, the soft-sensor model of the mill load is obtained. Use the model to calculate the test samples so as to forecast the mill load, the feasibility of this model in this paper is proved.