【摘要】 把混合气体分析的问题转化为模式分类的问题来处理的新途径已被相关文献证明是可行的。然而,当感兴趣体积分数区被分成更多的子区域或混合气体组分数多于2时,模式类别数目的急剧增加使网络的输出节点数也大增。这不仅使计算工作量大增,而且,更难使误差收敛到全局极小。对二元混合气体的数值模拟结果表明:利用概率神经网络(PNN)可解决上述的问题。PNN不仅易于使用,而且,总能给出一个合理的分类结果,样本数足够多时,可给出高达95.4%的识别率。此外,它的计算负载比反向传输网络轻得多。更多还原

【Abstract】 A new approach to analyzing gas mixtures by transforming the problem into a pattern classification one has been proved feasible by related literature.However,when the volume fraction area of interest divides into more sub-areas or the component number of gas mixture is more than two,the number of pattern classification will increase rapidly,and cause the great increase of output nodes of ANN.This will cause not only a larger amount of computation,but also more difficultly converging to a global minimum.The numerical simulation results for a binary gas mixture indicate that a probabilistic neural network(PNN) is able to solve the problem.The PNN is easy to use,and always able to give a reasonable classification result,such as recognition rate up to 95.4 % when samples are enough.In addition,its computing load is much lighter than that of the back propagation network.更多还原