Comparative Study of Nonlinear Approaches for Predicting Explosion Limits of Multi-Component Gas Mixture
Explosion limits are the important indices to evaluate the safety of multi-component gas mixture. In the present study, the effects of the composition (volumetric contents) of various components (i.e., H2, CO, CH4)in the gas mixture on the explosion limits were studied in detailed by the extensive experimental data. A non-linear relation between them was found, in order to predict explosion limits with acceptable accuracy, two novel nonlinear keroel-besod approaches, i.e., support vector machine (SVM) and generalized regression neural network (GRNN), were proposed to establish the relationship between the explosion limits and the composition of the multi-component gas mix-tare. The inherent model parameters of two approaches were extensively investigated so as to obtain the optimal models with the maximum predictive accuracy. A direct comparison between the current methods and the existing back-propagation neural network (BPNN) shown in the previons work was also made. The results showed that the explosion limits predicted by the current three nonlinear approaches were in good agreement with the measured values. For the lower explosion limits, the maximum relative errors for SVM, GRNN and BPNN models between the measurement and the prediction were 5.86％, 8.99％ and 6.28％, while for the upper explosion limits, they were shown to be 5.16％, 10.03％ and 7.65％, respectively. It is concluded that the proposed approaches were superior over the existing method in terms of the predictive accuracy and the ease-to-use feature.
explosion limits multi-component gas mixture nonlinear prediction support vector machine neural networks
ZHENG Ligang YU Minggao YU Shuijun JIA Hailin PAN Rongkun
School of Safety Science and Engineering,Henan Polytechnic University,Jiaozuo 454003,Henan,China
国际会议
The 2008 International Symposium on Safety Science and Technology(2008年安全科学技术国际会议)
北京
英文
1096-1101
2008-09-24（万方平台首次上网日期，不代表论文的发表时间）