Employing First Principles Model-Based Soft Sensors for Superior Process Control and Optimization

This paper describes the development and implementation of soft sensors at the world’s largest crude oil stabilization plant (Abqaiq) for superior process control and optimization. The soft sensor uses a rigorous steady-state model combined with dynamic synchronization to compute real-time stream properties (i.e., hydrogen sulfide amount (H2S), Reid vapor pressure (RVP), and true vapor pressure (TVP)) so corrective action can be taken immediately. Crude oil is characterized by 19 pure light components and 17 heavy pseudo-components. A non-Random Two-Liquid (NRTL) model is used for predicting thermodynamic properties of the liquid phase while an ideal gas model is used for the vapor phase. Unlike traditional data driven methods, the soft sensor adopts the first-principles modeling approach so current operating conditions can be correctly reflected in the online model based on sound engineering principles. Fault detection, sensor validation, and calibration with laboratory data are also performed online to ensure reliable and accurate predicted H2S, RVP and TVP values.