Hydraulic fracturing is a well-known production enhancement technique that achieves increased production rates when properly planned, performed and followed up after execution. Throughout the life of Hassi-Messaoud field operating companies have aggressively used this technique. The results of those jobs vary significantly from well to well and from one zone to another. In addition, the reporting of these results is scattered across several databases and servers of both operator and service companies in operational records and reports This work focuses on structuring and mining the data of the almost 500 fractured wells in this field, to get a clear understanding of the major parameters affecting hydraulic fracturing in a field like Hassi Messaoud.
Almost 80% of engineers’ time is consumed in gathering the data from the many sources before using the data to produce the answers. To reduce this time spent on gathering the data, the first step was to define the variables to be included in the dataframe. This was then followed by creating a workflow to set a battery of routines in R language to extract the information and generate this dataframe from the different repositories. Once the data was structured, it was clear what questions should be asked of the data (calculated variables), how to ask these questions (correlations, evolutions, distributions), and what to expect from future inputs (unsupervised or supervised machine-learning techniques).
From the almost 500 fractured wells in the field, 78% were successful with a range of post-job gains between 100 and 3800 BOPD. The most successful periods were 1993 – 1998 and 2000-2006 in Zones 15, 9, 25, and 8. Among the most representative zones, operational success (highest immediate post-job gains) is 30% and technical success (fracs that have sustained production through time) is 45%. Out of 105 re-frac jobs, 60% were operationally successful with gains between 14 and 1500 BOPD, 30% of them sustained through time. Since 2002, ten multi-stage frac attempts on horizontal wells have not yielded the desired results due to high field geological and reservoir complexities. This study briefly discusses the technologies and screening criteria to deal with such complexities. Another important conclusion is that successful jobs in this field are time independent, i.e. fluid distribution and reservoir characterization are important factors in successful frac operations, but they are not as determinant as job design and execution parameters. A map with the distribution of the best candidates for a frac/re-frac/multi-stage frac is also presented after applying machine learning techniques.
Lessons learned from, and best practices for, hydraulic fracturing in a mature tight sandstone reservoir and analysis reproducibility are presented from the both long and data-rich history. Task automation and data management habits across the involved staff are the most immediate and effective tangible benefits with the available resources that can be extended to the petroleum engineering community.
