Study On Soft-Sensor Of Coal Fill Level In Ball Mill Based On Evidence Theory

Ball mill is an important part of the coal pulverizing system which is widely used in advanced small thermal power plants. But most pulverizing system does not run in its economic conditions because of the difficulty of the material measurement in the drum. Since there is still not a precise and reliable coal fill level measurement method so far, the operators have to make it running in conservative conditions without considering their economy thus results in great waste of the energy. It is of significant importance to scientifically solve the problems of the drum level measurement not only for the control, economic and safe operation, but also for the performance tests.In this paper, the soft measurement methods for coal fill level of a ball mill are studied based on evidence theory.Firstly, on the basis of the review of domestic and abroad status and trends for level measurement in ball mill pulverizing system, it is pointed out that the related variable joint monitoring method and soft sensing modeling technology is an inevitable trend. The working principle of ball mill is studied which is taken the actual research object B coal mill of a power plant as an example. Through the analysis, the secondary variables associated with the material level of ball mill are determined. In addition, the data of the secondary variables is acquired by designing the experiment, pre-processed and analyzed based on the grey entropy theory to acquire the correlation with the material level, which is the basis of the model simulation.Secondly, the application of evidence theory in cognitive modeling of uncertainty reasoning, prediction and data fusion in thermal process field were studied. The limitations were analyzed and a new robust adaptive K-NN classifier was proposed based on the improvement of parameter distance, structure parameter optimization and the optimization criterion which improved the classification accuracy.Thirdly, a soft sensor modeling method based on evidence theory, that is the multi evidence regression soft sensor modeling method, is proposed which is with the model of evidence K-NN classifier. In order to verify the accuracy and usefulness of the proposed method, a simulation experiment was carried out based on experimental data. Experimental results of the real data on site show that the method proposed in this paper not only has a better ability to data processing & acquirement, but also can apply it to the changeable conditions in complex situations and has a higher measure and prediction accuracy.Finally, a combined soft measurement method based on D-S’s rule of combination is proposed. The partial least squares regression analysis software measurement (NLPS), BP neural network and support vector machine are adopted to respectively measure the level of a ball. Based on the analysis of measure errors, D-S’s rule of combination is applied to fuse the three models. In addition, the prediction accuracy in comparison with the soft measurement model of partial least squares is analyzed. Experimental results of the real data on site show that the method proposed in this paper has a higher measure and prediction accuracy.