Contact Info
San Juan Capistrano, CA
[email protected]
(844) OSPREY-NOW (677-7396)
Follow Us

Be An E&P Company That Operates By Exception

Written by Jon Snyder

I was visiting an oil company a few weeks ago, and I overheard an engineer say he was looking for 50 barrels of “missing” daily production. Fifty barrels may not sound like a lot to a large producer, but to a small “mom & pop” company it can be a significant fraction of their total production. Having worked for a very large global producer and then making the transition to small operators myself, I looked back at my past career and started thinking: How do operators find “missing” oil?


I was the production engineer for a platform, offshore California. One nice thing about a platform is all our facilities were concentrated. Our operators, who were the eyes and ears of the platform, were never more than a few minutes away from our 30 wells in the well bay. The platform was set in the 1980’s and unfortunately, we didn’t have a lot of automation on our wells. The ESPs had downhole sensors, but our first indication that production was “missing” was our aggregate flow meter through our LACT coming off the platform. From there, the operators would immediately start checking tubing and casing pressures in the well bay. If they didn’t see anything out of the ordinary, they would start cycling wells through our test separators. We had four ESP wells at the time, so they would run those first. They were our largest producers and had very stable flow rates that didn’t require a long test. The rest of our wells were on gas lift. We had a couple “bad actor” wells so the operators would cycle through them next, often requiring up to 12 hours per well. They would also take water cut samples to see if any wells had any significant increases. In most cases, we were able to find the main contributor to our production loss in a matter of hours. Other times, it would take a day or two. It wasn’t the best system, even though we made it work. Looking back, we were simply looking for problems.


In a field with limited automation, operating by exception can be rather difficult. Often, an operator will drive their normal route, going well-by-well, visually inspecting, recording pressures, checking valves, etc. Most operators have access to a production variance report each morning. Unfortunately, this data is a day old.


How did a well produce or flow yesterday compared to the day before? Sometimes we can compare the well’s production to a rolling average, e.g 7-days, 14-days, 30, days. Taking it step further, we can compare a wells actual production to its forecast rate.


“Production is king” is an expression I have heard and used often in the oil field, and it will remain king, but it is only one variable. All we’re doing is comparing production to production. To make matters worse, it’s hardly ever in real time. We’re essentially reading yesterday’s newspaper. How can we do better?


Let’s learn to operate by exception! We can take all the sensors available to us and utilize the more advanced tools at our disposal. When an engineer does a historical lookback, or well review, we can often find a trend or variation in the data prior to the well losing production. With the advent of data science and machine learning, we can label these early indicators and train an algorithm to identify these signals and notify us when something is awry!


A lot of oil and gas companies already use threshold-based notification systems to alert them when a signal goes above or below a certain level. Thresholds have their place, but unless appropriately managed, these notifications can lead to nuisance alerts, or alarm fatigue, to the point operators simply turn them off. At best, thresholds only look at one or two variables during a single time window.


Machine learning algorithms can look at all the available sensors over multiple durations creating thousands of features. Each feature can then be weighted based on importance.


There is another issue with thresholds: they are not “set and forget.” Wells change over time. Thresholds should too. With machine learning, models are individually calibrated to the well and changes with it over time.


With advanced analytics and anomaly detection we can spend less time finding problems, and more time solving them.


When you’re ready to start operating by exception, take a closer look at our platform and solutions.

Get in touch with us today to start the conversation about how we can help you get there!