基于图卷积神经网络汽油单体烃辛烷值的预测

GRAPH CONVOLUTION NEURAL NETWORK FOR PREDICTING THE OCTANE NUMBER OF PURE HYDROCARBONS IN GASOLINE

基于图卷积神经网络的神经指纹方法,引入了池化操作,建立了改进的神经指纹方法;进而采用改进的神经指纹法建立了汽油单体烃辛烷值的预测模型,作为对分子级汽油辛烷值调合模型的支撑。通过用单体烃沸点和临界温度数据集对预测模型进行验证,发现池化操作的引入对神经指纹法的预测能力有明显提升,改进神经指纹法模型可自动选取对辛烷值有利和不利的结构特征,双键结构对单体烃马达法辛烷值的影响比芳环结构的影响更大。该预测模型对研究法辛烷值和马达法辛烷值的预测达到了同等水平,取得了良好的预测效果。

Based on the graph convolution neural network,an improved neural fingerprint method was established by introducing pooling operation.A model for predicting octane number of pure hydrocarbons in gasoline was established by this method as an important part of blending model of gasoline on molecular level.Through the verification on the boiling point and critical temperature data set,the prediction ability of neural fingerprint method was improved obviously by introducing pooling operation.The favorable and unfavorable features could be selected automatically by the improved neural fingerprint method.It was also found that the effect of double bond on motor octane number was greater than that of aromatic ring.The improved neural fingerprint method also had good results in the prediction of both research octane number and motor octane number of pure hydrocarbons in gasoline.