Machine learning-assisted multiscale modeling of an autothermal fixed-bed reactor for methanol to propylene process

The current commercial multistage reactor for methanol to propylene (MTP) process suffers from poor propylene selectivity and catalyst efficiency, mainly because of the low inlet methanol concentration and long residence time. In this work, we proposed an autothermal co-current flow reactor for MTP process, where the reaction heat is continuously removed through heat exchange with cold reactants, thus single-stage reactor can be used with higher methanol inlet concentration. The reactor feasibility was investigated by a three-dimensional multiscale model, in which the diffusion–reaction interaction inside catalyst particle was described by a neural network model trained by machine learning. With the feeding methanol fraction increasing to 30%, propylene selectivity reaches 82.27% while the space velocity approaches 2.68 g_MeOH g_cat^?1 h^?1 at 99.97% methanol conversion, about 1.4 and 3.8 times those of a commercial multibed reactor, respectively. With proper catalyst bed dilution, the reaction temperature is well controlled between 700 and 754 K.