The objective of this paper is to enhance the theoretical and industrial validity of the type well production profiles used to evaluate, predict, and optimize development strategies for unconventional reservoirs. Current type well construction methods perform poorly when the individual wells used to construct the type well deviate significantly in performance characteristics. The paper introduces a new methodology to construct type wells using cluster weighted modeling, which can inherently capture the underlying reservoir heterogeneity and hence solve the critical challenges associated with the choice of wells used to construct a type well and each individual well's relative importance.
Common industry practice is to average all the ‘analogous’ wells in a given area of interest in a field to construct a single type well. The current process of selecting the analogous wells is entirely subjective adding to the unrecognized uncertainty associated with the type well forecast. In the new methodology, individual wells are modelled as a time series smoothed using cubic splines. Then the wells are separated into clusters and each cluster is represented by a local natural cubic spline model and the type well is represented as a global weighted probabilistic superposition (mixture) of the local models.
We tested our methodology on a group of wells from two counties expanding over the Barnett Shale and concluded that the proposed method has following advantages:
Better type well forecasts for younger wells. This has been one of the key problems in production forecasting using type wells, as relatively younger wells usually do not perform as predicted by type wells.
Better representation of reservoir heterogeneity. There will be no need to choose the analogous wells as the use of clustering creates groups of similar behaving (analogous) wells and the weighting probabilistically makes the forecast of the input well closer to the observed behavior of the analogous wells.
In some fields, the performance of wells improves with experience and in other fields the best wells are often drilled first. This change of development pattern can induce a bias in the type well forecast, but our methodology inherently removes this bias by using clustered weighting.