论文部分内容阅读
为了给工业界提供一种快速预测二元混合液体自燃温度的有效途径,将试验所测不同组分及配比的168个二元混合液体的自燃温度作为期望输出,将基于电性拓扑状态指数(ETSI)理论、引入混合ETSI概念而计算出的9种原子类型所对应的混合ETSI作为输入,采用三层BP神经网络技术建立了根据原子类型混合ETSI来预测混合液体自燃温度的BP神经网络模型,并应用改进的Garson算法进行多参数敏感性分析。经模型评价验证及稳定性分析,得到训练集的决定系数R2为0.965,平均绝对误差MAE为11.892 K,测试集的交叉验证系数Q2ext为0.923,平均绝对误差MAE为15.530 K,发现该模型的预测性能优于已有的多元非线性回归(MNR)模型,表明BP神经网络模型具有较好的拟合能力和预测能力,对烷、醇类混合体系自燃温度的预测精度最佳。
In order to provide an effective way for industry to predict the autoignition temperature of binary mixed liquid rapidly, the autoignition temperature of 168 binary mixed liquid with different components and proportions tested is expected to be output. Based on the electrical topological state index (ETSI) theory, a mixed ETSI is introduced by using the hybrid ETSI concept and a BP neural network model based on atom-type hybrid ETSI is proposed to predict the auto-ignition temperature of the mixed liquid using the three-layer BP neural network. , And apply the improved Garson algorithm for multi-parameter sensitivity analysis. After model evaluation and stability analysis, the coefficient of determination R2 of the training set is 0.965, the average absolute error MAE is 11.892 K, the cross-validation coefficient Q2ext of the test set is 0.923 and the average absolute error MAE is 15.530 K. The model predictions The performance of the model is better than the existing multivariate nonlinear regression (MNR) model, which shows that the BP neural network model has good fitting ability and predictive ability, and has the best predictive accuracy for the auto-ignition temperature of alkane and alcohol mixtures.