Transfer learning with attention mechanism as predicting tool for dynamic adsorption of ammonia on MCM‐41 matrix materials

Gas adsorption is one important field of gas separation and air pollution control. The modeling prediction of dynamic adsorption can benefit process design and speed up material development. In this article, we try to propose intelligent model for modeling prediction of dynamic adsorption. Transfer learning with attention mechanism is utilized for the first time to capture flow details near breakthrough zone. The predicting target and training source are heterogeneous data from different materials and different operating conditions. The best predicting performance is obtained using one layer of long short‐term memory, linear activation function, and 70% training rate. Transfer learning with attention mechanism obtains the best predicting results for MgBr2/MCM‐41 adsorption at 200°C, 184 kPa is 7.032 × 10−8 in mean squared error (MSE). It also obtains the lowest predicting error of 4.609 × 10−8 in MSE, for predicting CaBr2/MCM‐41 adsorption at 200°C, 176 kPa. The results have proven the capability of modeling prediction of transfer learning with attention mechanism for dynamic ammonia adsorption on MCM‐41 matrix materials.