A regression model for estimation of dew point pressure from down-hole fluid analyzer data

Abstract

Accurate knowledge of dew point pressure is important in understanding and managing gas condensate reservoirs. Without a correct assessment of dew point pressure, an accurate description of phase changes and phase behavior cannot be achieved. Numerous models for predicting gas condensate dew point pressure from surface fluid data have been proposed in the literature. Some of these require knowledge of the full composition of the reservoir fluid (based on laboratory experiments), while others only need field parameters that are relatively easy to obtain. This paper presents a new model for predicting the dew point pressure from down-hole fluid analyzer data. Such data are now measured (usually in real time) while obtaining down-hole fluid samples. The new model predictions give a quick estimation of dew point pressure for wet gas and gas condensate reservoirs. Since it relies only on down-hole measured data, the model provides an estimate of dew point pressure without the need for laboratory analyses. During down-hole fluid sampling, the model can be used to ensure whether the sample is still in single phase, or whether the dew point was crossed during the sampling operation. An early estimate of dew point pressure is also valuable in designing further tests for gas condensate wells. The new model, constructed using a fluid database of nearly 700 gas condensate samples, was devised using sophisticated statistical/machine learning methods, and attained a mean absolute relative error value of 2% for predicting the logarithm of pressure. In comparison with other dew point estimation models (that use surface fluid data), the chosen model was found to attain a similar level of accuracy when tested on samples not used in the model building phase.