Model-predictive control and real-time optimization of a cat cracker unit

Abstract

A project for control and optimization of the Residual Catalytic Cracking Process at the Mongstad refinery is near completion. Four model-predictive control applications have been successfully implemented, using the IDCOM control software from Setpoint Inc. The most attractive feature of the controller is the well-defined control prioritizing hierarchy, and the linear impulse-response models have proved to give satisfactory performance on this process. Excitation and identification of the dynamic models proved to be a difficult task, and careful design and monitoring of the tests was mandatory in order to produce good results. Multi-variable Pseudo Random Binary Test Sequences were used for the excitation. Technical performance and operator acceptance of the new control functions have been good, but it is realized that a continuing effort is needed to fine-tune and maintain such functions. An on-line optimization function based on a static yield and quality model combined with an economic model is currently under work. An extensive program of planned experiments, with careful logging of measurements and laboratory results is now completed, and a hybrid model established by a combination of theoretical knowledge and regression techniques fits the process responses surprisingly well. The chosen model complexity seems to be very cost-effective for this purpose. The optimization is only local to the cracker plant. Prices for feeds and products have to be in accordance with the over-all refinery objectives to contribute to a global optimum. The new functions automate and integrate work and information transfer that were previously handled by procedures. This fact represents new demands for co-ordination and co-operation in the organization if the potentials of the new functions are to be fully exploited.