Virtual Reactor and Dynamic Reaction Analysis by Using Neural Networks

本研究将人工神经网络作为“虚拟反应系统”应用于碳- nox反应动力学建模。采用固定床流动反应器，在不同的实验条件连续变化下进行了非稳态动力学实验。采用气相色谱法和NOx测定仪测量了O2、CO2、N2、NO和NO2浓度的变化，作为反应动力学的时间序列数据。反应动力学的时间序列数据（催化剂和碳的用量、温度和反应物质浓度随时间的变化）用于神经网络的学习。利用训练好的神经网络模型，在虚拟实验的基础上讨论了碳- nox反应的机理。

In the present study, artificial neural networks have been applied to modeling the reaction kinetics of carbon-NOx reaction as a "virtual reaction system". The unsteady-state kinetic experiments were carried out using a fixed-bed flow reactor at various different experimental conditions altered continuously. The changes in the concentrations of O2, CO2, N2, NO and NO2 were measured as time series data of reaction kinetics by gas chromatography and a NOx meter. The time series data of reaction kinetics (the amount of catalyst and carbon, temperature and concentrations of reaction species versus time) was used for learning the neural networks. The mechanisms of carbon-NOx reaction are discussed based on the virtual experiments by means of trained neural network model.