基于BP神经网络的SCR蜂窝状催化剂脱硝性能预测

Performance Forecasting for SCR Honeycomb Catalyst Based on BP Neural Network

选择性催化还原(SCR)法是目前烟气脱硝应用最广泛的技术之一,其脱硝效率与催化剂有紧密联系。在自制性能测试试验台上,以某商用蜂窝状催化剂为研究对象,选取空速、温度、氧气含量、氨氮摩尔比、NO浓度5组参数进行脱硝性能测试,分析其对脱硝效率的影响。在实验数据的基础上,应用BP神经网络建立了预测模型,并与实验数据进行了对比分析。结果表明,当BP神经网络拓扑结构为5×7×1时,训练结果较好,利用其进行脱硝性能预测时,绝对误差绝对值的平均值为8%,相对误差绝对值的平均值为11%,证明BP神经网络拟合效果较好。

Currently, the selective catalytic reduction （SCR） is one of the most widely used technologies for flue gas denitification. However, its De-NOx efficiency relies heavily on the catalyst. In this paper, take the example of a commercial SCR honeycomb monoliths, five sets of parameters, i.e., GHSV, temperature, oxygen content, NH3/NO ratio and initial NO concentration, are chosen in the De-NOx performance testing on a homegrown testing bed to study its effect on the De-NOx efficiency. Based on the testing data, the prediction model of De-NOx efficiency of SCR honeycomb monoliths is established by applying BP neural network. The results show that the convergence results are satisfactory when the topology structure is formulated as 5×7×1; When it is used in the performance forecasting, the absolute error is about 8% by average while the average relative error is 11% respectively, which proves the good fitting effect of BP neural network.