Modeling and control of an industrial hydrocracker using the discrete and continuous lumping methods and model predictive control

Abstract

Hydrocracking is a catalytic chemical process which converts high-boiling heavy petroleum fractions such as vacuum gas oil into lighter and more valuable products like naphta, diesel, kerosene, gasoline, and LPG. Hydrocracking takes place in the presence of rich hydrogenat elevated temperatures and pressures. The products are free of sulphur and nitrogen compunds which are hydrogenated into hydrogen sulfide and ammonia and which are subsequently removed. The aim of this work is to develop a model for an industrial hydrocracking reactor for optimization and control purposes. Our research is centered around the hydrocraking unit (HCU) of TUPRAS refineries which consists of four catalytic beds with interstage cooling by hydrogen quench. The overall reaction is exothermic and tight control of bed temperatures is crucial for achieving the optimal product distribution. For modeling purposes the methods of discrete and continuous lumping are used. In continuous lumping the reaction mixture is treated as a continuum in which the reaction rate constant kis a continuous function of the true boiling point of the mixture. A yield distribution function p (k,K) is introduced to formulate the amount of species with reactivity k formed from cracking the species with reactivity K. The existing HCU models of this type are steady state models and they do not explicitly include the heat effects. Our continuous lumping model is a pseudohomogeneous non-steady-state plug flow reactor model which includes both the material and energy balances. As such the model is original. In the case of discrete lumping the reaction mixture is characterized in terms of pseudocomponents that are defined for the lumped species boiling in a particular temperature range (i.e. cut). The two sets of lumping methods are fundamentally different approaches, and their joint development provides additional insight into understanding the behavior of HCU and arriving at a final reactor model for optimization and control. Model parameters were estimated using parameter estimation methods. With optimally tuned parameter values, the predicted reactor bed temperatures, hydrogen consumption, conversion and product distributions match TUPRAS' actual plant data very closely. This is demonstrated for different feedstocks in our training and validation data sets. The developed HCU model is used for two purposes: 1) To compute the optimal product distribution and the optimal reactor inlet temperature setpoint values. This economic optimization is performed under steady state conditions. 2) The dynamic model is used by a Model Predictive Controller (MPC) to control the product amounts at their optimal set points. When the product amounts (light naphta, heavy naphta, diesel, kerosene and bottoms) deviate from their optimal values due to feedstock changes, catalyst deactivation and other disturbances, MPC makes the necessary adjustments in the reactor beds inlet temperature setpoints. Temperature increase in each bed and the weighted average bed temperatures (WABT) are the addional variables which are kept within limits by MPC to maintain a desired level of conversion and uniform catalyst deactivation across the beds. Reactor inlet temperatures are changed to their new setpoints by the regulatory PID loops which adjust the hydrogen quenches between beds. This cascade arrangement of MPC and PIDs provides both optimizing and regulatory actions as shown in the figure below. In the presentation, modeling, optimization and control simulations will be given and compared with the present status in the plant.