Reaserch On Cement Grinding System Modeling And It’s Control Method

The development of China’s cement industry is very quick. From the mid-eighties of lastcentury, the production of China’s cement has ranked first in the world. In2013the totalproduction of cement was to reach2.41billion tons, accounting for over50%of the world’scement production, but the passing rate of the cement is low. This is due to fundingconstraints, the majority of our cement plants are using the off-line testing method, not usingthe on-line granularity monitor, the measurement results of cement particle size are seriouslylagging behind than the production, and this increasing the probability of failure of theproduct. The appearance of soft measurement technology can solve the above problems, andalso can ensure the measurement credible, reasonable and more accurate.This article is based on the combined grinding system of the pingyin cement plant ofshanshui group. Though the analysis of the cement manufacturing process, the current ofroller press, the volume of feed, circulating fan speed, the current of out-mill lifting machine,the out-mill load pressure, the speed of separator, the speed of big fan, the temperature of thematerial are chosen as the grinding auxiliary variable of soft measurements. By removingthe abnormal pull up data standards, and removal of abnormal data sample data werenormalized after treatment.Since the combined grinding system is a nonlinear, large delay and strong couplingsystem, it is difficult to use the method of mechanism modeling to make the model, therefore,after analyzing various modeling methods, the choice of BP neural network and squaressupport vector machines (LS-SVM) are modeled on the cement particle size. For BP neuralnetwork method, the Levenberg-Marquardt (L-M) optimization algorithm selection method,select the method of BP neural network hidden layer and the hidden layer nodes to establish acement particle size soft measurement model based on BP neural network are discussed indetail. Simulation results show that the selection and number of nodes in the hidden layer BPneural greatly affect the performance of the model, and the time of model training is too long,The over-fitting phenomenon is always disturbing. For the least squares support vectormachines. The basic theory, the choose of kernel function, model parameters, radial basisfunction, and use cross-validation method parameter optimization are discussed in detail. With the use of the least squares support vector machines, the cement particle softmeasurement model is established. Simulation results show high prediction accuracy, goodlearning ability and good generalization performance. By comparing the training error, themean relative error of the training, testing, and root mean square error of the mean relativeerror test and other performance indicators, obviously, the least squares support vectormachine is better than BP neural networks.The Microsoft Visual Studio2005is used for programming. Through OPC interface, thesoftware and the industry control machine are connected. The software has been successfullyused in cement production site and it can be a good response to the cement particle size. Theon-line measurement of cement particle size is achieved. At the same time, the software isable to guide the operator change the key parameters of some important machines.Furthermore, this is a good reference for better control.