Development of assets in new frontiers can be quite costly, especially in deep-water and ultra-deep waters. These developments have many uncertainties, and each has an associated risk to both operations and project economics. Leveraging the fact that a significant amount of information has been (and continues to be) gathered in public and/or private domains for many of these assets/wells, a multi-year effort to build a data-mining framework for establishing whether well completion and production performance could have been accurately predicted during various stages of deep-water development ("Pre-Discovery," to "Exploration," through to "Mature Fields") was undertaken. The final results of this work were incorporated into a selection/prediction software tool and database for decision support. The work helps establish priority for data acquisition and data values throughout asset development.
The project created one of the largest privately held databases of GoM well information: five different publicly available and privately owned GoM datasets were integrated to obtain field-scale geologic and reservoir properties, well trajectories, schematics, drilling, completion and production data for thousands of fields and wells in the GoM. Multiple derived attributes were then calculated using Petroleum Engineering relationships to enhance the reservoir characterization and well performance comparisons. Historical well production data, together with periodic well test results, were used to calculate well productivity indices, skin factors and contacted volumes; changes in these values were analyzed over the course of each well's lifecycle to determine its relation to completion designs, production practices and performances, and/or workovers.
Rigorous data mining techniques, including a variant of Principal Component Analysis (PCA) and K-Means Clustering, were used to determine the most prominent attributes governing the intrinsic performance properties of reservoirs ("uncontrollable parameters"). Once subjected to clustering techniques, the uncontrollable parameters indicate several main performance clusters, each of which could be uniquely identified by its paleontology, rock type, reservoir properties and sedimentation/folding regime. Further, each main cluster was observed to be sub-clustered into groups – the wells in each such group having similar completion and production practices (AKA "controllable practices"). Petroleum engineering know-how was used to define well performance attributes and categories to analyze ("objective functions"), which in turn showed a strong relationship between uncontrollable parameters and controllable practices for the performance of each group of wells. Well performance can indeed be predicted with a certainty that depends on the data available: often to within 65%, 80%, and 95%, respectively, depending on the data availability at that stage of a given well's development, from "Pre-Discovery," to "Exploration," through to "Mature Fields".