Simulation and analysis of dehydration distillation column based on distillation mechanism integrated with neural network

In an industrial solvent dehydration distillation column (SDDC) model, the Murphree efficiency represents the separation ability of a distillation tray and the SDDC model’s performance depends on the value accuracy of the Murphree efficiency. Because there are many operation conditions having nonlinear effect on Murphree efficiency, it is difficult to determine its value. To develop a precise and robust SDDC model, a novel hybrid model combining distillation mechanism with neural network is proposed. In the SDDC hybrid model, the neural network is employed to model the nonlinear relationship between the operation conditions and Murphree efficiency, which is embedded into the SDDC mechanistic model. The results showed a good predicting and robust performance of the hybrid model under different operation conditions. Based on the hybrid model, the effect of the operation conditions on SDDC was analyzed to obtain some useful guiding rules for the SDDC operation.