Comparison of Advanced Distillation Control Methods, Final Technical Report.

[摘要] Detailed dynamic simulations of three industrial distillation columns (a propylene/propane splitter, a xylene/toluene column, and a depropanizer) have been used to evaluate configuration selections for single-ended and dual-composition control, as well as to compare conventional and advanced control approaches. In addition, a simulator of a main fractionator was used to compare the control performance of conventional and advanced control. For each case considered, the controllers were tuned by using setpoint changes and tested using feed composition upsets. Proportional Integral (PI) control performance was used to evaluate the configuration selection problem. For single ended control, the energy balance configuration was found to yield the best performance. For dual composition control, nine configurations were considered. It was determined that the use of dynamic simulations is required in order to identify the optimum configuration from among the nine possible choices. The optimum configurations were used to evaluate the relative control performance of conventional PI controllers, MPC (Model Predictive Control), PMBC (Process Model-Based Control), and ANN (Artificial Neural Networks) control. It was determined that MPC works best when one product is much more important than the other, while PI was superior when both products were equally important. PMBC and ANN were not found to offer significant advantages over PI and MPC. MPC was found to outperform conventional PI control for the main fractionator. MPC was applied to three industrial columns: one at Phillips Petroleum and two at Union Carbide. In each case, MPC was found to significantly outperform PI controls. The major advantage of the MPC controller is its ability to effectively handle a complex set of constraints and control objectives.

[发布机构] Technical Information Center Oak Ridge Tennessee

[学科分类] 工程和技术（综合）

[关键词] Control systems;Distillation;Configuration;Energy balance;Neural networks