基于BP神经网络的二元混合液体自燃温度预测

发布时间：2018-04-05 21:34

本文选题：安全工程　切入点：二元混合物　出处：《安全与环境学报》2017年04期

【摘要】：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。
[Abstract]:In order to provide an effective way for industry to predict the spontaneous combustion temperature of binary mixed liquid, the spontaneous combustion temperature of 168 binary mixed liquids with different components and ratios measured in the experiment is regarded as the expected output.Based on the theory of electric topological state index (ETSI), the mixed ETSI corresponding to nine atomic types calculated by introducing the concept of mixed ETSI is used as input.A BP neural network model for predicting spontaneous combustion temperature of mixed liquids based on atomic type mixed ETSI is established by using three-layer BP neural network technology. The improved Garson algorithm is used to analyze the sensitivity of multiple parameters.After model evaluation and stability analysis,The determination coefficient 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. it is found that the prediction performance of this model is better than that of the existing multivariate nonlinear regression model.It is shown that the BP neural network model has better fitting ability and prediction ability, and the prediction accuracy of spontaneous combustion temperature of alkane and alcohol mixture system is the best.
【作者单位】：南京工业大学安全科学与工程学院江苏省危险化学品本质安全与控制技术重点实验室;
【基金】：国家自然科学基金项目(21436006,21576136) 江苏省高校自然科学基金重大项目(12KJA620001)
【分类号】：X932

【相似文献】