As the effort to develop unconventional reservoirs takes on a greater pace within the industry, so too does the effort to understand the way in which these unique reservoirs produce. Classical forecasting techniques such as Arps are typically not able to adapt to the characteristic "tail" exhibited by wells producing from these reservoirs. As such, a tremendous amount of investigation and research has been undertaken to better understand and therefore better predict future performance. As a result the literature is full of alternative techniques.
In this paper we explore a number of these techniques, comparing them to the classical Arps method as a form of benchmark. In addition, we take a "backcasting" approach on several large shale fields, including the Barnett, Eagle Ford, Bakken and Montney. This approach is equivalent to going back in time, hiding later production data, forecasting and comparing forecasts to what we now know production rates to have been. We also test the applicability of classical and newer forecasting techniques on big data sets by identifying changes in operating conditions and calculating decline rates for the period thereafter. As seen in this study, each method has its own "sweet spot" in which it yields accurate results.
This study differs from other published studies in that our focus was not so much on obtaining Estimated Ultimate Recovery (EUR) values that matched the actual production of depleted wells, but rather was on predicting production rates in the relatively near term. While EUR estimation is a key performance indicator, it falls into the domain of the long term planners. Conversely, future production rates are of prime importance to operational personnel and to their short term decision making processes, such as timing of infill drilling and remedial activity, along with surface facility planning activities such as system upgrades or reconfiguration.
Another aspect of this study which differs from others published is that we did not preclude wells with noisy data. Rather, we embraced the challenge presented by real data sets. In this way we could ensure that our proposals could be applied universally and not simply to those wells with well-behaved production histories.