Inferential Models Using Vapor-Liquid Equilibrium Equations of State

Vapor-liquid equilibrium properties, such as dew point and enthalpy, can be more indicative of plant performance issues than raw online measurements. This article demonstrates how to use these properties to create inferential models.

Inferential (or soft-sensor) models are usually generated by regressing a set of plant measurements against an infrequent lab analysis; these models can be used to obtain frequent estimations of real-time operating conditions. In this way, inferential models are an indirect measurement of the state of a system. In many applications, operators can monitor a stream’s composition from an online analyzer, but may require a stream property that is not apparent from the composition or other plant measurements.

These properties indicate either an equipment condition or distance from an operating constraint. They are not linear combinations of simple plant measurements such as temperature, pressure, etc., but can only be ascertained by employing plant measurements along with a vapor-liquid equilibrium (VLE) equation of state.

Generally, the Peng-Robinson equation of state is used to determine these properties for hydrocarbon systems, although other equations of state can be used if other components are present (e.g., H₂S, CO₂, and particularly water). Solving for these properties is moderately numerically intense — iterative algorithms and numerous correlations must be employed to determine equilibrium properties. But standard algorithms have been published, and the required correlations with parameters for most compounds are readily available.

This article provides three examples of inferential models using a VLE equation of state. These models provide timely and intuitive information about the state of the operating plant...