Automating gas lift optimization has many benefits. When considering machine learning approaches, there are too many wells and too few engineers to constantly monitor, analyze, update and evaluate the gas injection rate to establish optimal lift performance. A typical simulation model development followed by parametric analysis to decide a gas injection rate may potentially require close to half a day of work for a qualified engineer. This is due to the efforts in collection of sensor data, completion and lift design data, reservoir, and fluid properties, decision making to select the range, distribution and granularity of modeling parameters, history matching or validation, and deciding the gas injection set-point value.
In the case of unconventional wells, the static bottom hole pressure (SBHP) and productivity index are transitioning and difficult to measure. Hence, extreme discretion is required to perform deterministic nodal analysis. We content that a probabilistic approach which learns from the history of the well using inverse modeling offers a more reliable solution.
Updating the operating gas lift set-point is required several times over the life of the well. This can be due to natural effects such as decline, or due to intervention such as restimulation or workover. Accounting for these factors, the work hours invested can be about 7-10 days per engineer per well annually for human-driven simulation analysis. For a 100 well field, this translates to approximately 2-3 engineering years. With the high volume of wells, it becomes unlikely for production engineers and well managers to undergo this intensive process on a proactive and regular basis.
This leads to a state where the optimization process is performed infrequently, losing out on production, or injection gas or both. Popularly, rules of thumb based on the experience of personnel can dominate the decision making. Such an approach can be subjective to the level of diligence of the individuals managing the wells, who need to astutely evaluate if the conditions under which the rules of thumb were formulated continue to hold. In order to provide a solution which scales up to the challenges of such a problem, automation is necessary.
If these benefits resonate with you, then we would love for you to request our white paper entitled “A Machine Learning Approach to Automating Gas Lift Optimization” written by our Chief Data Scientist, Mike Pennell, and two key contributors and Data Scientists, both Petroleum Engineers, Venkat Putcha, PhD and Nhan C. Le. By clicking the link, you can request our paper to learn more about our human augmented approach to machine learning.