Research On The Computational Models Of Coal Calorific Value For Yili Region Of Xinjiang

In this paper, the computational methods for coal calorific value of YILI inXINJIANG were studied; research works in this thesis had been conducted as follows:1)The relationships between the coal calorific value of YILI in XINJIANG andother coal quality analysis were analyzed and summarized.2)The empirical formulas already existed at home and abroad, for the predictionof coal calorific value were cleared up. According to the characteristics of coalsamples of YILI in XINJIANG, the possibility of whether the empirical formulascould be used to calculate coal calorific values was analyzed. The results indicated theproportion of coal samples that the absolute error of coal sample which was less than0.3MJ/kg calculated by Gmelin formula was14.8%, calculated by Chinese coalresearch formula which was suitable for non-caking coal, weakly caking coal and longflame coal was10.64%, calculated by Chinese coal research formula which wassuitable for all types of coal in China was17.02%. Those empirical formulas couldnot meet the accuracy requirements for the prediction of coal calorific value.3) Based on the coal samples analytical data of coal sample from YILI inXINJIANG, both the linear models and nonlinear models for the prediction of coalcalorific value were established. The results of the optimal linear model which builtby proximate analysis and ultimate analysis show that, the prediction error of95%ofthe coal sample were±0.424MJ/kg and±0.839MJ/kg respectively, that could notreach the national standard.According to the research on the linear models, it'snecessary to build non-linear models. For the non-linear models, in this paper,BP model, the radial basis function network (RBF) model, the minimum support vectormachine model and the genetic algorithm to optimize the BP model were mainlyresearched. The construction principle, modeling methods and techniques of thesemodels were analyzed in detail, and the corresponding MATLAB procedures werewritten. The results show that the calculation accuracy of the nonlinear model is much higher than the linear model. When proximate analysis were used as the inputs ofnonlinear model, the calculate accuracy of model was SVR> RBF> GA+BP> BP.The maximum mean square error was0.0719MJ/kg that was the prediction error of95%of the coal sample was±0.141MJ/kg. And the coal sample mean squareerror of forecast was0.09658MJ/kg.When ultimate analysis were used as inputs ofnonlinear model, the maximum mean error of the nonlinear model was0.15MJ/kg,and each model had good predictive ability except RBF model.Finally the advantagesand disadvantages of these methods were compared, the optimal computational modelto predict the coal calorific value of YILI in XINJIANG was founded, that is theregression support vector machines (SVR) model.4) For the realization of the nonlinear model to predict the calorific value of coalfacilitatly and according to the need to selecte the appropriate empirical formula, anapplication software which can be used to predict coal calorific value of YILI inXINJIANG was developed based on the MATLAB software programming and GUIfunctions, containing the calculated interface, the empirical formula database, coalsample classification and recognition systems, and so on.