基于ANN-GBC模型的煤液化油临界性质研究

Study of critical properties of direct coal liquefaction oil based on ANN-GBC Model

针对传统基团贡献法和半经验关联式未考虑煤液化油分子中基团键相互作用和缺乏适合其重质馏分油临界性质的计算方法,构建了基于人工神经网络-基团键贡献耦合模型(ANN-GBC),采用3层网络结构,输入层神经元数由煤液化油包含的45个基团键和常压沸点共46个,隐含层最佳神经元数通过试差法优化确定为40,临界性质作为输出层,研究了煤液化油15个窄馏分的临界性质与其分子结构之间的相关性。对20种模型化合物进行了ANN-GBC模型的校核与验证,其计算值与理论值偏离相对误差在2.5%以下,相关系数0.999 69,表明该模型具有较好的模拟推算和精准辨识同分异构体的功能。结果表明:煤液化油的临界温度、临界体积均随蒸馏切割温度升高而升高,临界压力随馏分蒸馏切割温度升高呈先升高后下降趋势。模型预测值与半经验关联式对比结果基本一致,各馏分段不同组成物质的含量差异导致了个别结果的跳跃。该模型不仅揭示了煤液化油临界性质与分子结构之间的定量关系,而且为其他复杂体系临界性质的预测提供了一种新的有效方法。

The traditional Group Contribution method and Semi-empirical Correlation do not consider the interaction of group bonds and lack of calculation methods for the critical properties of heavy distillates in direct coal liquefaction oil（ DCLO）.So Artificial Neural Network and Group Bond Contribution Coupled Model（ ANN-GBC） is constructed to predict critical properties of DCLO.Three layer network structure is adopted,with 45 group bonds and atmospheric boiling point of DCLO,a total of 46 as the number of neurons in the input layer of ANN-GBC Model,the optimum number of neurons in the hidden layer is determined to be 40 by the difference method,and the critical properties is used as the output layer. The correlation between the critical properties of 15 narrow fractions of DCLO and their molecular structure are studied.The critical properties of 20 model compounds are checked and validated by ANN-GBC Model,the relative error between the calculated value and the theoretical deviation is below 2. 5%,and the correlation coefficient is 0. 999 69. The results show that ANN-GBC Model has good function of simulating and accurately identifying the isomer. The critical temperature and critical volume of DCLO increase with the increase of distillation cutting temperature,and the critical pressure increase firstly and then decreased with the increase of cutting temperature.The results of model prediction and semi empirical correlation are basically the same,and the content difference of different components in each narrow fractions leads to the jump of individual results. The ANN-GBC Model not only reveals the quantitative relationship between the critical properties of DCLO and the molecular structure,but also provides a new and effective method for the prediction of the critical properties of other complex systems.