Operational Optimization of Crude Distillation Units with Unknown and Changing Crudes | AIChE

Operational Optimization of Crude Distillation Units with Unknown and Changing Crudes


Yang, X. - Presenter, The University of Manchester
Zhang, N., University of Manchester
Smith, R., The University of Manchester
The operation of crude distillation units (CDUs) plays a central role for refineries. The operational optimization of CDUs is a complex problem. The complexities mainly come from two aspects. One is that the adjustable variables, including the furnace outlet temperature, flowrates of pump-arounds, flowrates of steam and flowrates of products, have strong interactions. The other is that the crudes charged into CDUs are usually unknown and changing because they may be blended from several different types of crudes according to the scheduling process.

In this work, a framework for solving the problem using data generated from optimization models is proposed. First, the process data under optimal operating conditions of different crude scenarios are generated from the optimization model. Next, the adjustable variables are divided into two groups, primary variables and secondary variables through data analysis. The purpose of the grouping procedure is to avoid over-optimization caused by inaccuracies of the model. Primary variables contribute to most of the economic potentials in all scenarios. Third, different strategies are applied to primary and secondary variables. ‘Insight’ variables, which represents common features among all crude scenarios, are extracted from the data and paired with the primary variables using self-optimizing control technique. A multi-input multi-output controller is possible to be designed to adjust the primary variables so that the insight variables are maintained at constant values. For secondary variables, two strategies are proposed. The simplest way is to keep them directly at constant values. The second method is to construct the correlation between the primary and secondary variables through the data. In this way, the secondary variables can be determined after the primary variables are adjusted to near-optimal values.