中文摘要：

根据混合气的爆炸极限与混合气各成分的体积浓度之间具有非线性关系的特点，笔者提出采用神经网络非线性方法来计算合有H2，CH4和CO的多元混合气体的爆炸极限。在模型中，H2，CH4和CO的体积浓度作为输入，爆炸上限和下限作为输出。计算结果表明，该非线性模型预测混合气爆炸下限和上限的最大相对误差为3．90％，3．57％，而模型预测值与计算值的相关系数分别为0．971，0．981；非线性模型的预测结果要好于偏最小二乘回归的预测结果。当H2，CO，CH4在混合气中的体积浓度给定时，非线性模型能够准确预测混合气的爆炸极限。

英文摘要：

A non-linear method of neural network model for predicting explosion limits of multi-component gas mixture containing HE, CH4 and CO is put forward according to the non-linear relation between the explosion limits of the mixture and its components volume concentration. In this model, the volume concentration of H2, CO, CH4 in the gas mixture is chosen as the input of the NN model, and the lower and upper explosive limit as the output respectively. The predicting result shows that the maximum relative error for lower and upper explosion limits is respectively less than 3.90% and 3.57% , and the correlation coefficient for lower and upper explosion limits is respectively 0. 971 and 0. 981, which is much more accurate than that by Partial Least Square Regression（PLSR）. This illustrates that the explosion limits of gas mixture can be accurately predicted by the NN model when the volume concentration of HE, CO, CH4 in the gas mixture is determined.