Wells are usually completed and fractured regardless of the formation's geological variations along the lateral in unconventional resource plays. Engineered completions intend to improve the well productivity by minimizing the geological variance within a stage, better locating each cluster. Targeted perforation friction can still be achieved when the number of clusters vary by stage but using difference hole size or number of shots per cluster, and different pumping schedules can be applied to achieve a uniform frac load in difference length stages or to achieve higher frac efficiency by formation types. Compared to a simple geometric design with a constant spacing and cluster number by stages, numerous engineered strategies can be considered providing more insights to formation heterogeneities and comprehension on what a well-by-well economic optimization and offset frac hit optimization and making completions more sustainable by reducing water and hydraulic horsepower needs.
Using a field database with both pre-frac rock data and post-frac production data, a machine learning process was developed to predict the post-frac production distribution along the lateral. In this data set, the pre-frac data is free/low-cost Gamma Ray (GR) and RockMSE (Mechanical Specific Energy calculated from drilling data) which is manipulated inside the tool to estimate the geological variations along the laterals. The post-frac data available is proppant tracer logs, but a similar process could be used to optimize with fiber, imaging, WT-log, or other post-frac data. Once the relationships between facies and cluster-production are learned for the given reservoir, the tool can be used to optimize the placement of the clusters along subsequent wells and defining how they should be grouped in stages to maximize production. Early data indicates hydrocarbon production is higher from wells completed with engineered completions when compared to geometrically completed wells with a similar frac load.
The semi-automatic tool integrates both machine learning and completion optimization, and it has been used in a United States shale play to quickly use MSE derived from drilling data, to quickly output recommended perforation depths. Any low-cost data acquired along the lateral before fracking compared with productivity proxies obtained after fracking at cluster/stage scale can be used to feed the supervised data learning workflow. Multiple engineered designs are then generated for each well accommodating formation heterogeneities along the lateral in function of the desire fracturing intensity. An understanding of the well productivity versus fracturing intensity can also be incorporated, so it provides more insight for selection of more optimized completion choices that impact economics and hydraulic fracturing sustainability.
Geometrical design optimization relying only on clusters spacing modification, increasing, or decreasing the fracking intensity in function of the economic environment, do not allow the understanding of the highly dynamic, nonlinear relationship of the well productivity versus completion choices. Improving the geological knowledge of unconventional formations and its impact on well production can reduce the environmental impact by generating wells with similar productivity to geometric analogues, with lower fracking intensity (and at a lower cost). Reviewing various engineered designs proposals based on low-cost data learning is a novel way to reduce fracking intensity while maintaining desired productivity, through informed choices.
