Gas Lift Optimization, what is it and how can you improve it? In this blog, we will discuss three current methods of gas lift optimization, and then discuss how you can move beyond these tedious current methods into Machine Learning. Most operators can agree that optimizing Gas Lift is important for determining the best setpoints, maximizing production, and minimizing costs, but what are some of the popular and current methods for optimization? We will take a look at those manual methods below, which are current standard operating procedures for many engineers, and then discuss a newer, more efficient method.
Method #1: Determining Setpoints from Experience
The first method is Experience or Analog-based set point determination. Here, field operators use their intuition and knowledge of the field to determine the gas injection set points. This may be based on their understanding of a set point value which has worked well for a given well, or a set of wells. Other wells which are considered to be similar to these wells are given a set point value in the same ball park. With this kind of engineering supposition, you might ask yourself: What kind of success rates can you get with this method?
Well, you need testing to understand whether the set point is actually optimal or not. There are multiple interacting factors influencing gas lift optimization. A well may be operating in a sensitive or insensitive zone of a performance curve based on a combination of factors which differ from well to well and may change with time. If the well is in an insensitive zone, the operator may as well reduce the gas significantly. However, this particular method leaves you vulnerable to a lot of missed opportunities to optimize gas lift production.
Method #2: Step Rate Change Testing
The second method is step rate testing. Operators can change injection rates on a well and observe the response of the well. This provides some understanding of its gas injection performance. The step rate changes method is more of a hands-on version of Method #1, but provides an empirical basis for choosing a set point. In regards to feasibility, this method has some drawbacks regarding numbers of wells.
When there are too many wells, and wells declining at all different rates, the number of step changes required across the field are too astronomical to perform comprehensive testing. If the testing is based on too short of a sample time, it may be affected by the variability in the well. If it is based on a large sample time, the decline and other reservoir changes may affect the testing. Further, It is vital to observe if the well operational parameters during the testing period are affected by abnormal slugging or other interferences. It is an incredible challenge. These changes make the optimum rates a constant moving target, and model free methods of optimization can be vulnerable to temporary abnormal behavior.
Method #3: Nodal Analysis / Simulation
To get a better understanding of the well, nodal analysis is a common method to provide a physical basis. Nodal analysis, or simulation, is the most popular method used to provide a physical basis to model a well. These models need to be history matched with the physical well data, which is then used to predict the well’s production at different gas injection rates.
Let’s dive into this a little deeper. The volume of fluids a well produces is dependent on the reservoir pressure and the flowing bottom-hole pressure. This relationship is defined by the inflow performance curve. At the same time, the flowing bottom-hole pressure of the well is affected by the volume of fluids produced, the volume of gas injected and the pressure at the wellhead, this is represented by the outflow performance. The inflow and outflow performance are mathematical equations which need to be solved simultaneously at a node, usually the bottom of the well. The two equations can become interdependent which can be further complicated due to the thermodynamics and fluid mechanics of multi-phase flow. Thus, numerical simulators are used to iteratively solve these equations. Unconventional wells typical show a high degree of variability in production and operation. The variability in parameters such as GOR, Water-Cut, THP etc. combined with uncertainty in reservoir pressure and productivity index of well requires simulation of several cases, and perform parametric analysis subsequently.
Hopefully you are thinking that this sounds powerful, but maybe you see some challenges. For example, you may ask: How long does it take to do the history matching and nodal analysis for each well?
Developing the model does certainly require a big investment of time for training and revision of models. When done right though, each well can take up to one week of time per year per engineer from setup, parametric analysis, history matching to gas injection recommendation. This needs to be coupled with set point change testing to complete the full circle. And all of this is completed for just one well. Our research has shown that the software used for storing data and simulation are usually different, so the engineer also needs to take the load of the input, transformation and matching of data into this time calculation. This is a difficult process. You can imagine what it takes to achieve this for a field, so this means that, unfortunately, all wells cannot be optimized due to a shear lack of time available.
A New Method: OspreyData’s Production Intelligence Solutions
The three methods above are the traditional optimization methods. What we have been curious about and dedicating our research to is how we can utilize machine learning to improve these methods with automation. You might be asking, how does automation help?
Firstly, wells change their behavior. A system is required to keep track of the changes and update the set point. With the help of continuous simulation and machine learning backed history matching, it is possible to leverage the data from these changes to learn the underlying conditions of the well. In the midst of these ever-changing conditions, once created, the automation models become opportunities to gain key insights about the well.
The next reason to automate is to make sure you’re comparing apples to apples in testing. To measure the response of the step rate changes is not as trivial as it seems. If the well is not operating in a normal state, statistically speaking, either before or after the change the test response may be misconstrued. Sometimes, there may be a lag in response, in other cases there can be interference. There needs to be a system which calculates the statistics on the fly, detects anomalies, and can self correct if there majority behavior of the well does not represent the state during the test.
Lastly, automation obviously enables you to respond to a greater number of wells with drastically increased response times. As we all know, well volumes are huge and the number of eyes which can observe them are limited. A machine learning powered system can monitor, maintain and optimize a large field continuously.
For a more detailed discussion on these methods, feel free to request our webcast replay entitled “Increase Gas Lift Production with Artificial Intelligence.” It discusses these topics in conjunction with greater detail about the benefits of Gas Lift Optimization with our Production Intelligence solutions. If you want to start maximizing your oil production and increasing your ROI, contact us to set up an appointment with our sales team to discuss easy implementation on your oilfields. We say easy because Production Unified Monitoring is a cloud-based tool that can be accessed from any web browser without any hardware or software installs, meaning there are no capital expenditures. Contact Us Today!