Hybrid Models for Monitoring & Optimization of Hydrocarbon Separation Equipment

Separation equipment such as distillation, absorption, and stripping are the fundamental processes found in refineries and petrochemical industries. Accurate prediction of the performance of the separation equipment is essential for optimization of real-time operation, production planning and scheduling. Over the last two decades, rigorous models (tray to tray distillation models, detailed kinetic models or reactors) have been introduced for use in on-line optimization. Models of crude separation units describe crude feedstocks (a complex hydrocarbon mixture) in terms of pseudocomponents, predict product composition in terms of pseudocomponents and then compute distillation curves of the product streams. This enables model use for predicting properties such as 5% or 95% point on a distillation curve and comparison to the corresponding specifications. The resulting models are quite large, nonlinear, computationally intensive, and requires a specialized expertise to be maintained. Hybrid models of separation equipment developed in this work consist of two parts: material and energy balances that determine external and internal material and energy flows in the equipment; and linear Partial Least Squares (PLS) models that predict product properties based on the properties of the feed and the internal equipment flows. The models are highly accurate, capable of predicting product properties and internal equipment flows with approx. 0.5% error with respect to the rigorous models of the same equipment.