The main task of the refinery is to convert crude oil into usable end products. going through a distillation process to separate the different fractions. One of the units aiming at increasing the quality of diesel is the diesel hydrotreating unit (DHP). During 2015, TUPRAS purchased approximately 20 different types of crude oil, which were changed approx. three times per week. Each blend changed the feed properties of the DHP, especially T95. Nevertheless, the T95 of the inlet stream is not measured on-line, but in lab experiments in relatively long intervals.
Economic optimization of the DHP unit by optimizing the feed blenc procedure with the help of current Model Predictive Control (MPC). Adaptive learning functions of FUDIPO will be used to enhance MPC. The main target of the demonstration is to produce the maximum amount of diesel at given specifications. Current measurements are to be upgraded, with new soft sensors included in the model, based on NIR and RF, which will allow to know the T95 of the feed to operate the unit more efficiently. A subsequent challenge is to build robust models for prediction of the feed properties from NIR-spectra correlated to data like the T95.