Statistical Time Estimates for Deepwater Completions in the Gulf of Mexico: Integration of Neural Networking and Distribution Modeling.

摘要

Predicting completion time in deepwater wells is an imminent necessity in the modern well construction cycle. The primary objective of this thesis is to present a novel integrated approach of statistical analysis and neural networking to identify well characteristics for their impact on total time to complete a well.;Using a neural network, fifteen crucial attributes from the Dodson Database were used in this study and analyzed for relative impact with respect to time. These attributes included primary parameters such as well depth and interval number. Wells in the database were assigned a value, depending on their fifteen attributes, that correlated to length of time to complete.;The program designated prospect wells a value using the same time weighted impact system, as well as the same impact parameters. Wells within the database with most similar values to the prospects were used in the statistical analysis for total completion time. Actual data were used for the 15 parameters in the program for "Dark Star," "Liberty," and "Terrapin" to test the reliability of the statistical analysis. Estimates for "Dark Star" and "Liberty," which were completed in 2014, were within 5% of field completion time. "Terrapin" is yet to be completed; however, the programs estimate was within 3% of the Approval For Expenditure (AFE). With access to the data provided by Stone Energy as one of the active operators in GOM, this thesis presents a valuable methodology to estimate completion time.