E&P Spotlight: Enabling the Vision of Dynamic Gas-lift Optimization

While rod pumps are easily the most visible form of artificial lift and electric submersible pumps (ESPs) are known for their high production rates, many operators are turning to gas lift to produce in unconventional reservoirs.

In October 2018, WTI entered a sharp sell-off period and declined from over $76/bbl to less than $43/bbl by late December 2018. This decline put pressure on LOE, as it continues to do today. Oil settled in by year’s end in the low $40s, matching current prices in second-quarter 2020.

During that steep decline in 2018, Tom Walker, chief production engineer at Occidental Petroleum Corp., said in a previous Hart Energy article, “Right now we’re adding more and more wells in our resource plays to gas lift.” That year, gas lift was used on 7% of Occidental’s producing wells but accounted for 31% of Occidental’s bbl/d production.

Both independent and major producers choose gas lift because the lift type has low failure rates, minimal moving parts, low implementation cost and the ability to handle solids with few problems. Gas lift, having close characteristics to natural flow production, offers a seamless transition if implemented in the early stages of a well’s life. New gas-lift implementations such as high-pressure gas lift (HPGL) have recently been adopted for producing at rates comparable to ESPs. Also, when applicable, gas lift can be an economical type of artificial lift, which comes in handy when the break-even point for a barrel of oil is close to the price of oil.

As a result of gas lift’s wide production rate and gas-liquid ratio (GLR) operability, it is common to start a well on an ESP and switch to gas lift at a later decline stage. The GLR increases while the well is still producing several thousand barrels per day. As bottomhole pressures decline further, and liquid production rates lower to hundreds of bbl/d, operators may implement a combination of gas lift and plunger lift or operate gas lift intermittently with a programmable logic controller (PLC). 

Gas-lift optimization

Gas-lift wells do encounter maintenance issues such as holes in tubing, leaking or clogged valves, frozen injection lines, paraffin or scale obstructions and compressor shutdowns. In most cases, these maintenance issues are relatively infrequent and easy and inexpensive to correct compared to ESP and rod pump failures. Due to these relatively low maintenance costs and minimal downtime, production/field management’s focus shifts toward optimization to minimize LOE. Gas-lift optimization has a targeted nature; it changes to meet the operator’s main objectives as they relate to lease economics, reservoir management and pricing. If injection gas costs and/or sales gas prices are high, the objective will be to maximize the oil revenue per unit cost of net injected gas.

In a low gas price environment, gas-lift optimization will focus on maximizing production. In shale-based or unconventional wells with high decline rates, the optimum gas injection rate to achieve either of these objectives becomes a moving target as a result of frequently changing GLRs, production rates, well pressures and interactions with adjacent and networked wells.

Historically, field personnel monitor individual wells closely, taking daily recordings of pressure, temperature, flow rate and production rate. This quickly becomes unscalable across a field of several hundred wells, especially in unconventionals with highly variable behavior. This makes it common to use generic and static gas injection rates on wells over several weeks or months, while the GLRs, production rates and downhole pressures migrate from values at which the injection rate was last optimized. Unfortunately, this means many wells are not operating optimally, as it takes significant time to analyze a single well and observe the effects of the optimization changes enacted. 

Optimization through digital twins

Nodal analysis that employs physics-based simulations is a tool used extensively for understanding well behavior, history matching, design, analysis and optimization. In the era of Big Data, it is possible to build digital twins of wells that leverage collected sensor, production and well data and run massive scale physics-based simulations. OspreyData has seen early success implementing digital twins within its Production Intelligence system for artificial lift producers seeking unified monitoring, failure detection and mitigation. Within the company’s Production Analytics system, which provides machine learning-based analytics, OspreyData has begun implementing automated gas-lift optimization for producers on groupings of wells.

Production Analytics enables live facility network simulation models that adapt with and reflect changes observed in the field. These simulations are intended to scale field-wide to provide optimization recommendations that respond to ongoing changes in well behavior. The following underlying concepts enable this full vision:

1.    Sensor placement and data frequency. To enable nodal analysis, these are key metrics to understand a well’s flow:

• Inflow performance: flow from the formation into the wellbore;
• Tubing or outflow performance: flow from the wellbore to the wellhead; and
• Surface facilities and pipeline network: flow on the surface from the wellhead to facilities, creating back-pressure from the wellhead that impacts bottomhole pressure.

Nodal analysis software expects inputs related to all three of these components. In OspreyData’s interactions with operators, the company has observed that pressure and flow rate sensors are most commonly placed on the casing, tubing and injection lines close to the wellhead. It is less common to observe downhole pressure gauges and sensors for separators and other surface facilities. There is a spectrum of data collection practices, including early adopters who place and collect data downhole, and from the wellhead, surface facility and pipeline network sensors with 15-second to 1-minute frequency.

This robust level of data collection enables a holistic understanding of a well’s behavior and its interactions with connected wells. More traditional players monitor and record casing pressure, tubing pressure, and injection rate and pressure once a day. Nodal analysis software does provide a method to model and match well performance with this limited input, which requires making certain assumptions related to the missing data. This may result in generic solutions leading to false matches, misdiagnosis and potential mismanagement. Within OspreyData’s Unified Monitoring system, traditional operators can build a stronger data backbone to enable nodal analysis by profiling their data to highlight gaps and opportunities to increase coverage.

2.    Conducting an initial analysis of field data. In this analysis of gas injection rate changes, 72 changes were observed. Data was ingested through OspreyData’s Unified Monitoring system to facilitate monitoring and labeling of historical events. The graphic shows the impact of one injection rate change. The blue bars and shaded regions represent the changes, while the well’s casing pressure and tubing pressure is visualized below.

Gas injection rate changes often overlap with interfering well conditions. It is helpful to perform analysis on time periods where well performance is relatively stabilized and normal. These ideal conditions often last for a few hours or fractions of days which may be affected by noise if the analysis window is not precisely isolated. To capture event details with usable time windows, flow rate and pressure data frequencies should ideally be sent every 15 minutes to a maximum of every two hours. Higher frequency data helps clarify the stability of the well behavior during the analysis period. If the digital twin is also intended to mimic slugging or oscillatory well behavior using a transient simulation, higher frequencies of data may be necessary. 

Operational insights and new knowledge 

The blox plot image (place Plots image here) illustrates the percentage change of wellhead pressure, injection pressure, downhole pressure, sales gas gauge pressure, separator pressure and gas-lift injection rate on a grouping of wells. 
In case of these wells, it can be see that downhole pressure is the least sensitive to gas injection changes, with the median percentage change being lower than 0.3%, while the median separator pressure, wellhead pressure and sales gas gauge pressure change by 4.3%, 10.4% and 16.2 %, respectively, resulting in a median gas-lift flow percentage change of 32%.

One of the popular simplifying assumptions during gas-lift design and nodal analysis is to set the most downstream node (usually the wellhead pressure, sales gas pressure or separator pressure) as a constant. Learnings from the plots above indicate that this may be a questionable assumption. It can also be inferred that bottomhole pressure is less sensitive to gas injection rate changes on these wells.

A typical approach to gas-lift optimization looks to increase production rates by reducing flowing bottomhole pressure by minimizing the fluid gradient. Given the higher sensitivity of wellhead and surface pressure nodes when compared to bottomhole pressure, an alternative opportunity is possible to optimize by debottlenecking the surface equipment to reduce backpressure.

As wells often share surface facilities and pipelines it is necessary to observe the interaction between wells, as a potential gas production or injection increase on a connected well may create additional backpressure. This insight was made possible due to the availability of high-frequency data from sensors placed downhole as well as at the wellhead and surface. 

3.    Using the digital twin as a gas Injection recommendation engine. Beyond sensitivity analysis and identifying bottlenecks, a history matched simulation model can provide timely gas injection recommendations. A recommendation engine would require automated data collection, transformation and clean-up followed by simulation and history matching. The simulation has multiple unknown inputs such as static bottomhole pressure and productivity index because these elements tend to change with time. There are also immeasurable inputs such as the tubing/annulus friction factor that may or may not change with time. Automated history matching of live generated simulations presents the challenge of non-uniqueness with many possibilities being generated early on and multiple parameters having a transient nature including GLRs, water cuts and production rates. 

A fully scalable system

The data scientists and artificial lift experts behind OspreyData’s Production Analytics system are developing machine learning and probabilistic models that adaptively learn from the well’s history to identify zones in which optimization recommendations can be most effective. With the built-in feedback loop its product provides, models continue to learn from the well’s response to subsequent gas injection rate changes. This methodology provides real-time recommendations as well as creates a better understanding of the underlying trends in static bottomhole pressure, productivity index, friction factor and other previous unknowns. This approach will allow producers to optimize gas-lift wells with high variability across the full field and move toward the vision of fully automated field-wide optimization. Producers will benefit from increased production with lowered LOE while achieving a high operational efficiency.