分子筛上苯与丙烯烷基化反应催化性能的人工神经网络模拟  

The catalysis performance simulation for alkylation of benzene and propylene on zeolite using the artificial neural network

作　　者：孙晓岩[1] 李建伟[2] 项曙光[1] 

机构地区：[1]青岛科技大学炼油化工高新技术研究所,山东青岛266042 [2]北京化工大学化工资源有效利用国家重点实验室,北京100029

出　　处：《计算机与应用化学》2011年第3期351-355,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(20576012);青岛科技大学科研启动基金项目(0022430)

摘　　要：针对分子筛上苯与丙烯烷基化反应速度快,难以获得准确动力学方程的特点,通过探索构建人工神经网络模型来关联分子筛催化剂本征性能、反应条件和催化性能之间的相互关系。研究分别以单输出和多输出参数为预测目标分别建立2种不同形式的BP神经网络模型,从预测数据和实验结果的比较上可以看出,无论单输出还是多输出网络,两者之间的平均相对误差均较小并且具有较高的相关系数,说明所建立的神经网络模型可以较准确的预测苯与丙烯烷基化反应性能。比较的结果还表明,单输出网络由于其针对性较强,较多输出网络具有更加精确的预测能力。该神经网络的建立可以为实际生产提供理论指导,也可以应用于催化剂的设计与开发,确定适用于反应最佳的催化剂织构特性和反应条件。The accurate kinetic equations of the alkylation of benzene and propylene on the zeolite catalyst could not be obtained easily because of the quick reactions. The usability of Back-Propagation （BP） artificial neural networks （ANN） for the correlativity of the essential capabilities of catalyst, the reaction condition and the catalysis performance were investigated. Two kinds of BP artificial neural networks were established with the aim of the single and multi-output parameters. The results showed that for the both of ANN model, the average relative deviations of the predict data and experiment data were small. And the regression coe^cient of determination showed a good correlation between estimated and experimental data sets for both train and test data sets. They were indicated that the ANN for estimation of the product distributions of alkylation reaction had a good predicate performance. Moreover, the ANN with a single output had a better predicate performance than the ANN with the multi-output. The establishment of the ANN offered the theories of practical production and the design and development of the catalyst guiding to find the feasible essential properties of catalyst and the reaction conditions.

关 键 词：催化剂 BP神经网络 预测 催化性能 

分 类 号：TQ015.9[化学工程]