THE WOODLANDS—Gone are the days when looking back on situations to determine why plans did not unfold as expected was enough for oil and gas operators aiming to cost-effectively boost production.

“We need to get to the point where it is telling us something,” Scott Raphael, vice president of operations for P2 Energy Solutions, said of big data before turning to a technology application that is gaining attention. “Predictive analytics is the science of using facts from the past to analyze the present and predict future performance. It’s the basis for what we know as reservoir simulation.”

With big data and predictive analytics, operators can use production surveillance by exception, comparing predictions to new actuals, to determine how a particular oil or gas asset is performing. In response, adjustments can be made to improve production. Raphael spoke on this topic during the SPE Digital Energy Conference last week.

“More and more we rely on predictions. The more representative the results are the better we can trust the outliers that it generates for us to analyze,” he said.

During the presentation, Raphael explained how a computer-based production forecast algorithm uses oil and gas production figures to forecast future volumes. The model he described had two parts. One explained the data in terms that are physically plausible, generally showing how production changed over time. The statistical part, he said, is used to explain normal deviations from physical behavior such as if a well gets shut in, worked over, is stimulated or recompleted. All of these things create a need for an engineer to look at the forecast to identify what is really happening, so knowing the well’s history is important, he said.

Once samples are drawn, they can be pushed forward to make a probable forecast. The production forecast algorithm also can be used to set thresholds, enabling alarms to be triggered when production falls to a certain level.

For one example, Raphael showed how a well’s production fell more than 50% below the 30-day moving average, triggering alarms.

“It initially triggers but that quickly becomes the new norm and we don’t have any alarms until we trigger a new low,” he said. “It’s sort of a just-in-time analysis. Whereas if we have a true probabilistic approach to forecasting, and probabilities can be trusted, then we can base our alarms on a probability threshold.”

Possible applications for predictive analytics for production surveillance are wide-ranging.

“It’s everything that we want to know about the future related to our production,” Raphael added. “We can do a most likely [forecast] based on the history of the well and you can also extrapolate from the historical data points what the full capacity production would have been had [the well] produced every day of the month.”

In addition, the forecast can be used for baseline economic to help determine costs associated for refracs or other EOR efforts, production losses and for examining common systemic problems, he added.

“The key to surveillance is knowing what to expect,” Raphael said. “The better forecast you have the better every other analysis that is based on that forecast will be. ... It’s’ an innovative technique that takes some of the bias out of forecasting.”

Contact the author, Velda Addison, at vaddison@hartenergy.com.