Well construction is the most expensive stage of oil field development. Oil and gas companies carefully calculate all possible expenditures to assure that the project of well construction and exploitation will be profitable in time. The time for every operation of well construction is strictly regulated. However, in different oil and gas companies time allotted, for example, cementing surface casing or tripping operations can differ significantly. Therefore, some oil and gas companies are unable to calculate their well construction expenses accurately and lose a significant amount of money, which can be reduced. Also, engineering teams spend a substantial amount of time developing a drilling plan for the drilling crew, which may also be fully or partially developed by automated means. We present a data-driven approach for automatic planning and scheduling of drilling operations on wells with similar design and geological characteristics, suggesting also an improved operation classification based on the IADC dictionary of drilling operations.
The model for drilling operations planning is based on the directed graph traversal and process mining techniques. The algorithm takes as an input an undetailed plan, the section which is planned to be drilled, and the phase of well construction (drilling, cementing, logging). The algorithm selects the shortest path between two adjacent operations, gathers all paths together, and outputs a detailed plan with the corresponding time for each operation. For directed graph construction, we processed about 15 drilling reports from wells of a particular oil field which have similar well design and geology conditions by virtue of their geological proximity.
Algorithm performance was estimated by comparing graph time against similar plan time which was calculated based on the median time of every operation in a whole dataset. Median time (P50 percentile) was used because it demonstrates objective time in terms of well construction operations. We conjecture the above based on the fact that in some cases time for tripping can be faster or slower in some wells due to geological or other conditions, and the median provides an outlier-robust estimate of the average value. Also, graph time was compared between the engineering team's proposed plan and the actual time from drilling reports. Graph construction quality was estimated using three principal metrics: the Jaccard coefficient, structural distance, and fitness similarity.
