The transient nature of well properties contributes greatly to uncertainty and makes optimization an ongoing challenge for the production engineering team. The figure above provides a snapshot of a few key operating parameters of a well on an unconventional formation, over na two-month range. During this period, the oil production rate declines sharply from 500 bbl/d to 150 bbl/d, the Water Cut varies heavily between 0.5-0.9 (50% to 90%), the casing pressure drops from about 800 psi to about 600 psi, the tubing pressure changes from about 170 psi to 110 psi and the calculated operating gas lift valve changes from the 5th valve (relatively shallow valve) to the 2nd valve (deeper valve).
Further, the figure demonstrates that around January 30th to February 7th, the well is subjected to interference from an external activity which can potentially be a frac hit. This can be observed by temporary increases in water cut, oil rate, casing pressure, tubing pressure, and operating gas lift valve (GLV) estimates. This is an example of potentially several other changes to the state of the well which demand constant attention for adaptive set-point changes.
Scalability of the well management problem is limited by the volume of wells coupled with uncertain parameters and transitionary behavior. Solving this purely with human input may require the engagement of several engineers with a great understanding of the well, reservoir, and surface conditions. In the context of unconventional wells with extremely low permeability, measuring the accurate representative Reservoir or Static Bottom-Hole pressure becomes an additional challenge. In the event of a well shut-in for testing, the bottom hole pressure in an unconventional well may remain transient for months.
Therefore, the true representative pressure to be used as a static bottom-hole pressure becomes uncertain and transitionary. The productivity index of the well also changes significantly with time, unlike a typical conventional well. This situation makes it arduous to effectively utilize nodal analysis on unconventionals using engineering estimates. Further adding to the problems of scale and accuracy, it can be argued that manual engineering efforts follow a reactive approach. This can be improved with a proactive approach which is probabilistic and learns from history and self-updates.
Our white paper, entitled “A Machine Learning Approach to Automate Gas Lift Optimization” presents an automation solution which constantly monitors the well, generates candidate models through simulation, selects the appropriate model and provides a gas injection recommendation. To request this white paper and learn more about our approach to automating gas lift optimization that assists with the transient nature of well properties, click here.