Abstract
Production forecasting in unconventional reservoir systems is an important problem for the industry. While "averaging" observed production profiles is often used to construct type wells, model-based methods have significant advantages, including observance of basic fluid-flow principles. Model-based approaches for type well construction involve using analytical and numerical models for a representative well in a group of wells. Scaling factors are used to apply the model to the rest of the wells in the group. However, there is uncertainty in the estimates of these scaling factors that is not accounted for when we use point estimates. In this work, we calculate the uncertainty in the scaling factors in a group of analog wells. We then account for this uncertainty in the estimation of the P10, P50 and P90 type wells within the group of analog wells, using a Bayesian hierarchical model.
We select a group of analog wells and fit a model to a representative well in the group. We then calculate the scaling factor for each well in the group as a probability distribution. We pool these distributions to estimate the variation in scaling factors between different wells. This represents the uncertainty in the scaling factors across the group. We translate this into uncertainty in the production rate profile across the group. Finally, we generate P10, P50 and P90 production profiles from the distribution. These represent type wells for the group of analog wells.
We show that our technique is superior to point estimates of scaling factors traditionally used in the industry, due to pooling of distributions. Point estimates ignore possible variation in the scaling factor due to noise and presence of outliers. We show that this could accumulate and have a significant impact on the estimated type well. This propagates to the calculation of estimated ultimate recovery. We validate our method with simulated data, with and without added noise.
Forecasting production for producing wells and estimating production on proved undeveloped reserves are problems of engineering and economic importance. Existing methods using empirical averaging of production profiles or point estimates for scaling factors fail to use information on the uncertainty of the scaling factors. By discarding this information, we discard uncertainty that could influence reserves estimates. We present a method that incorporates this uncertainty into the analysis transparently. Our method can also be used to pool uncertainty across various well groups.
