Parametric analysis is a common exercise employed by production engineers in well modeling and decision making. The figure above displays a set of gas injection performance plots (Oil production rate vs. Gas injection rate) resulting from a parametric analysis of a physics-based simulator. In an example 10-day window of observation, it was observed that the oil production rate of a well varied between 70-90 barrels per day. The other uncertain parameters in the parametric analysis include the unknown reservoir or downhole conditions, and the varying surface parameter measurements. Simulations are generated for a combination of cases varying these parameters, and cases which match the operating production rates are selected.
As demonstrated, there is a non-unique solution resulting from the simulation. In a typical manual review, the engineer is expected to eliminate some of the possibilities to select a configuration which represents the well’s current operating state. As the figure illustrates, the gas lift performance curves can broadly be categorized into either flat or curved lines. The flat lines indicate insensitivity of the well production rate in response to changing injection rate. In such cases, the operator can save gas by reducing the injection rate without any significant impact on production. The curved lines present a possibility to improve the oil production rate by finding an optimum injection rate, and, a potential to lose production by injecting at an incorrect injection rate either by under-injecting or over-injecting.
The underlying conditions representing the individual cases are not always known to the engineer, as explained before. Hence, the major problem in gas lift optimization is model selection under uncertain and transient conditions. In our case example, the operator may make a difference of up to 20% in improvement or under-performance of the oil production rate contingent on the selection of the appropriate model. Thus, customarily, the engineer has to invest significant time to resolve the non-uniqueness of the solution and ultimately select a model based on deduction, conjecture or supposition.
To learn more about how we use machine learning to automate gas lift optimization, request our white paper on the subject. In future weeks, we will be discussing the transient nature of well properties and the uncertainty that makes optimization an ongoing challenge for the production engineering team.