用神经网络模型预测催化裂化原料油结构族组成的研究

PREDICTION ON GROUP ANALYSIS OF FCC FEEDSTOCK BY NEURAL NETWORK MODEL

利用人工神经网络模型 ,根据催化裂化原料油的碳元素质量分数 (ωC)、氢元素质量分数 (ωH)、密度 (ρ2 04 )、运动粘度 (ν10 0 )四参数对催化裂化原料油的结构族组成进行预测 ,获得了较高的精度 ,计算值与试验值相比 ,CP、CN、CA 的平均绝对误差分别为 0 .36 %、0 .16 %、0 .15 %。用该模型对未参加训练的 4种油样进行检验 ,预测结果的平均相对误差为 4.0 5 % 。

Prediction on group analysis of FCC feedstock with 4 properties, carbon m%, hydrogen m%, density at 20 ℃ and dynamic viscosity at 100 ℃, using neural network was developed. Comparing the calculation data with the experimental data, the average absolute deviations of CP, CN and CA were 0.36%, 0.16% and 0.15% respectively. The accuracy of the prediction method was promising.