A Hybrid Modeling Approach for Catalyst Monitoring and Lifetime Prediction

Abstract

In this work, we present a hybrid fundamental-empirical model to monitor and predict the catalyst lifetime of an operating industrial reactor. The hybrid model combines a fundamental adiabatic reactor model to calculate the activity of the catalyst bed with an empirical partial least-squares model to predict the catalyst activity at different operating conditions. A baseline model was trained on process data and validated separately using analytical data, showing good agreement between the measured reactant breakthrough of the reactor train and the predicted values from the model over 18 years of continuous operation of four industrial production reactors. To implement the model for catalyst activity monitoring, the model must closely match the current catalyst charge performance. Therefore, the baseline model parameters are updated automatically with new plant data using a filter algorithm.