Inferential monitoring and optimization of crude separation units via hybrid models

The hybrid model of a CDU introduced in this work can be used for both production planning and for RTO, since the model converges rapidly via SLP, it is small in size, its accuracy is comparable to rigorous models, and it can optimize the operating variables, including furnace duty and pumparound duties. This enables CDU part of the planning models to be as accurate as the RTO models, instead of using standard CDU modes of operation. The model includes mass and energy balances and statistical models relating crude distillation curve points, tower operating variables and tray temperatures to product distillation curves. Hybrid model of a sample CDU shows excellent agreement with property predictions from the rigorous model; RMSE of points on the product distillation curves is typically less than 0.5% of the rigorous model predictions. Optimization of the hybrid model via SLP converged to a better point than equation-oriented AspenPlus.

► Model for optimization of CDU operation in multiperiod planning models and RTO.
► Model of a CDU predicts product TBP curves with RMSE 0.5% of the rigorous model.
► Hybrid models consists of mass and energy balances and a PLS model for product TBP curves.
► Interaction of pumparounds and the preheat train included in the model.
► Model of a CDU consists of approx. 100 easy to converge, mostly linear equations.