Neural network-based prediction of auto-ignition temperature of binary mixed liquids
Auto-Ignition Temperature(AIT)is one of the crucial parameters in the design of fire and explosion safety measures.However,the current experimental methods used to measure the AIT values of mixed liquids are time-consuming,labor-intensive,and hazardous.This study employs the Quantitative Structure-Property Relationship(QSPR)approach and utilizes a Back Propagation Neural Network(BPNN)and a one-Dimensional Convolutional Neural Network(1DCNN)to establish a predictive model for AIT values of binary mixed liquids.The input parameters of the experiment were molecular descriptors of the binary mixed liquids,and the output parameters were the experimentally determined AIT values.The model's fitting degree,stability,and predictive abilities were assessed and validated using various methods,followed by a determination of its applicability range and a comprehensive interpretation.According to the results,the BPNN and 1DCNN models in the training set have root mean square errors of 4.780℃ and 9.603℃,respectively.The corresponding average absolute errors are 3.775 ℃ and 7.842 ℃,and the average absolute percentage errors are 18.202％and 18.488％.The difference between the goodness of fit and the 5-fold cross-validation goodness of fit are 0.058 and 0.040,respectively.These findings indicate that the BPNN model exhibits excellent fitting capabilities,the 1DCNN model demonstrates good stability,and both models display satisfactory predictive abilities.The leverage method is used to determine the models'applicability range,and it is found that the leverage values in the application domain analysis diagram all fell within the applicable range(within the standard residual range of±3 and to the left of the standard leverage value).The shapley additive explanation method is utilized to assess the impact of nine atom types on the AIT values of binary mixed flammable liquids.The results reveal that both models exhibit the highest predictive accuracy for binary mixed liquids of alkanes and alcohols.
万方数据
维普
