What's new in production

Perhaps not as fascinating to some oilfield gear-heads as the nuts-and-bolts side of things, production allocation is certainly of intense interest to operating-company bean counters, reservoir engineers and operations managers. After all the planning, geoscience, drilling, completion and production technology is applied to the well, production allocation is where the petroleum meets the pipe.

Allocated production challenges. A recent white paper by NavPort describes both the landscape of production allocation—heretofore mostly a manual process—and a solution to meeting its fundamental challenges. The company first answers the question: “What is allocated production? In certain states, monthly oil and gas production is reported on a lease or unit level, rather than an individual well level. Leases or units can contain one well or many wells. In instances where many wells are present in one lease/unit, it is a significant challenge for mineral rights owners, operators and engineers trying to evaluate performance on an individual well basis. The method of ‘allocating’ production means to separate out the production values for individual wells from the total production of the group of wells with-in a lease.”

It then details the challenge: “Many companies have developed estimated oil/casing head gas production values, based on the monthly volume from each well on the lease. Most allocation methodologies rely heavily on well test data and pending production data to estimate production of the well. Some states do not require regular testing, while others require reports on annual or semi-annual basis. This results in recently completed wells only having, at most, a single measured data point. This means that the more recently a well has been completed, the higher the uncertainty in the allocated production measurement. In other states, the average time before a production measurement is taken can be upwards of 365 days after the well has been completed. So, for wells with lifecycles of three to five years, this comprises almost a third of the well’s productive life. For lease owners and operators, as well as third-parties that provide supplies and services, a more accurate system is needed to estimate allocated production. This will allow for better resource management and more economical administration of their holdings and customers.”

The company believes its proprietary method for allocating production for oil leases is such a system. This method uses (but is not limited to) well test data, initial production tests, lease production data, completion/recompletion data, lease/well master list, shut-in and plugging data, multiple producing zones (multi-completions), producing dates, pending production, permit data, well information, completion information and reservoir information data.

The process has three fundamental steps:

1. Generate a decline curve for each individual well.
2. Integrate the area under the decline curve to calculate potential production per month.
3. Calculate the allocated production for the well by multiplying the ratio of the production of the well to the sum of the production for all wells in the lease by a production-per-lease value.

The company has patented an interesting trick for decline curve analysis (DCA). It’s applied to “poor test data” wells; typically those wells that do not receive production testing for some time, often a year or more after completion. Here, the only data available may be completion, reservoir and well data.

This solution depends on the level of data available for each well, and is triggered when test data is missing or reported extremely late in the well’s lifecycle. It uses completion, reservoir and well information to predict the most likely type of decline curve and the decline curve parameters.

Machine learning. The company’s idea is based on this field of computer science, which applies a data-driven approach to the development of models. Machine-learning algorithms (MLA) learn from a data set and build a model that fits the data set independently, to make predictions on another data set. A key benefit of the data-driven approach is that human bias, from previous experience, can largely be removed. A computer scientist can choose to provide some direction to the program by adding labels to data fields or by leaving them unlabeled (respectively called “supervised” and “unsupervised” machine learning). Regression analysis is used for estimating the relationships among DCA parameters and predictor variables.

Hundreds of thousands of “rich test data” wells are used to generate reliable DCA parameters and then applied to make predictions of poor test data wells. Dozens of variables are included as input to the model, such as (but not limited to) location, producing formation, field, sub-basin, trajectory, proppant type, fracture date, TVD, MD, lateral length, proppant per lateral foot, proppant mass, total water volume, frac job type, operator, and so on. Each of these variables has different levels of impact on production. MLAs excel at finding the correlations to build a predictive decline model for application to the poor test data wells. As additional well level testing data are reported, wells can migrate from the poor test data category to the rich test data category.

The company reports that for a subpopulation of Eagle Ford shale wells, predictive decline rate values correlated to the actual decline rate. In this case, roughly 80% of the variance of the predictive decline rates is explained by the model. This model effectively makes predictions about the decline trends of poor test data wells, that otherwise would either not be possible or would be based on inaccurate average estimates.

The oil and gas industry must be second, only to astronomy, in the use of indirect measurement techniques. By further developing this capability, another useful tool for that purpose will be added to the industry’s ever-growing toolbox.