Register Today! Webinar on June 16 | Geopolitics & Energy – Supply Risks on the Rise

Production Forecasting, Predictive Analytics, and Today’s Oilfield


Harnessing the power of Big Data is without a doubt the biggest driver of business success today. For different companies and in different industries the incentives and motivations are as varied as weather. For the “excess capacity” ridesharing companies Uber and Lyft, for example, we as consumers see how good they are at locating a quick and inexpensive solution to our problem of needing to be somewhere else quickly. They mine the same data and see how much it is worth to their bottom line to prevent subjecting their “workforce” to local regulations. And with access to our individual travel patterns – down to the second from their mobile app – who knows what hidden forces they will soon uncover and start to further monetize.

The oil & gas industry is no stranger to big data, in fact the industry has arguably been working with big data longer than anyone else. The extreme data demands of exploration geophysics alone has been a major reason for many advances in computing power, and even today some of the biggest supercomputers spend large amounts of time calculating seismic volumes.

Today, however, computers are faster, programmers are smarter, product people are more keen to solve interesting problems, and investors expect a lot of return for their dollars.

Leading-Edge Multivariate Statistical Analysis for the Oil Patch

Let’s take a look at using big data analysis to identify best practices in a completion optimization plan. In a normal unconventional field the number of geological and engineering factors that can influence production are nearly infinite, with a few of the top-line factors being amount of proppant per foot, length of lateral, number of stages, wellbore spacing, distance to known faults or other impermeabilities, etc. Furthermore, even among the top line items I listed there are opportunities for redundant results – longer laterals are likely to have more stages – but we must be careful to make certain that the co-linearity of the similar data points do not create undue expectations in the final prediction.

Unsatisfied with the one-size-fits-all statistical tools like Spotfire and Tableau, Drillinginfo’s Transform Software has at its core a very sleek and modern multivariate statistical engine that allows for easy management of these industry-specific variables, and as it creates likely prediction models, mitigates the possibilities of false correlations from co-linear data.

This figure shows an area of the Bakken with many factors normalized and predicted production based on additional proppant. If you are investing in additional proppant in the red area, good for you; if you are investing in more proppant in the blue area you might be throwing money away.

Bakken Non-Linear Regression production forecasting

Probabilistic Decline Curves and Production Forecasting

That’s all well and good if you’re a Geologist or Geophysicist or Engineer and have the excess brain capacity to guide inquiries and adjust to the outcomes, but what if you’re a mineral rights owner or an investor or even an interested peer company. We have recently taken the same multivariate methods and applied them to decline curve analysis.

​Brute force curve fitting can be prone to irregular step patterns and exaggerated +/- values based on outliers.​ Using probabilistic decline allows for smoothing the distribution and handling outliers. It also allows users to know the confidence interval for the EUR numbers P10 P25 P50 P75 P90 (P10 values have the lowest probability of occurring; P90 values have the highest probability of occurring).

In the following image, we show the production curve of a well on the upper left. After running multiple regressions vs. a variety of models to determine a best fit decline curve, on the right we see the Linear Gauss Markov (LGM) curve, with 10 identified outliers shown in red. On the lower left we show the P10, P50, and P90 EURs for that well based on the LGM best fit. (This tool is available to all DI subscribers).

Probabilistic Decline Curve Example production forecasting

And Into the Future

As you can see one of the best methods for dealing with the sheer size and interconnectedness of big data is to use multivariate analysis, and we will continue to apply MV stats to even more of the most important (and interesting) issues the industry faces.

Your Turn

What do you think? Leave a comment below.

The following two tabs change content below.

Eric Roach

Eric Roach is the editor of Drillinginfo's blog, which was selected as the Top Oil & Gas Industry Blog based on visibility, engagement and relevance. He also prepares a weekly newsletter of top industry news for blog subscribers, and would be grateful if you would subscribe and tell your friends. (There's a box on the upper right of the page where you can subscribe).