Distillation models presently used in real time optimization are rigorous tray to tray models. For instance, a crude distillation unit models for real time optimization predicts accurately product distillation curves and other bulk properties (e.g. sulfur, density, etc.) by using approx. 10k to 20k of nonlinear equation. The model is capable of predictions which are within 1% of the actual equipment performance. On the other hand, models used for planning and scheduling are much simpler (e.g. swing cut model) but they provide a significantly lower accuracy. Discrepancies between these models make it difficult to reconcile the results between the planning and scheduling objectives and the objectives for RTO.
Our goal is to develop accurate and yet small in size, easy to converge distillation models which include mass and energy balances, as well as prediction of composition related properties of the product streams. Such models will enable determination of optimal crude selection and associated optimal operating conditions in production planning. In addition, these models are a step towards optimizing operating conditions while scheduling the operations. Being able to use the same model for RTO will ensure that the operating targets are consistent across planning, scheduling, and RTO.
In this work, we present methodology for development of hybrid models of distillation units. The goal is to achieve accuracy on par with detailed tray to tray models and to have much smaller number of equations. In addition, the model needs to be linear or nearly linear to make it easy to include it in the overall plant model for production planning and for scheduling.
Crude unit model (about 200 mostly linear equations) removes commonly made assumption that the front end/back end of the adjacent product from the CDU are equidistant from the crude distillation curve. Model performance is demonstrated on a crude unit consisting of a preflash tower, atmospheric tower, and of vacuum tower. Model predictions are within 1% to 2% of the rigorous tray to tray model which has been used as a surrogate for the plant.
We also present models of several additional types of distillation towers (about 30 to 40 equations per tower) and illustrate their accuracy by comparing them with rigorous simulations.
This work provides a basis for consistent determination of optimal operating conditions in production planning, production scheduling, and real time optimization.