基于神经网络的催化裂解产品产率模型

摘要

Based on artificial neural networks,a model predicting yields ofDCC products was established.The model used Levenberg-Marquardt algorithm to promote convergence and handle part extremism.The results showed that average errors of LPG,gasoline,diesel,propylene,coke plus loss are 1.8%,2.4%,5.7%,5.8%,6.3% respectively,meeting the commercial requirements.%采用神经网络方法，构造了一个催化裂解产品产率的BP神经网络，并利用Levenberg-Marquardt算法来提高收敛速度及克服局部极值。模型预测结果液化石油气、汽油、柴油、丙烯和焦炭加损失产率的误差分别为1.8%，2.4%，5.7%，5.8%，6.3%，能够满足工业应用需求。