Recent applications in unconventional reservoirs have shown a reverse trend of reducing stage length and perforation spacing, instead, increasing the stage and cluster spacing becomes preferential to free the cash flow and maximize the rate of return. However, the impacts and uncertainties of the key drivers that affected well productivity have remained ambiguous within the GSA. This paper investigates the variable importance including completion parameters and geological factors controlling the recoverable hydrocarbon potentials via unsupervised-manifold-learning algorithms, thereby quantifying and prioritizing the critical productivity drivers. The proposed methodology further enables production forecasting on both a single-well basis and a pad-well basis.
In this study, we utilize the Laplacian Eigenmaps (LE)-based and the Locality Preserving Projection (LPP)-based manifold-learning algorithms to first preserve the intrinsic geometric information inside the completion and geological data, then establish the correlation with the corresponding production. The trained models are cross-validated and optimized using grid-searching. We further conduct the variable importance analysis, and forecast new well production to demonstrate the efficacy, versatility and superiority of the LE and LPP-based methods using three single-well and five pad-well unconventional field cases in Junggar Basin. The outcomes show the specific parameters (e.g stage spacing, proppant intensity) are among the key drivers for both a single-well basis and a pad-well basis analyses.
In summary, we propose the LE-based and LPP-based algorithms under the paradigm of unsupervised-manifold-learning for productivity driver evaluation and production forecasting in unconventional oil reservoirs. We demonstrate the applicability and reliability of the proposed algorithms with total eight field case studies. The results show that the key drivers are identified and quantified in importance hierarchy. The unique value of methodology in this kind further provides the insight into productivity characterization and facilitates the completion parameter optimization thereby improving reservoir development in unconventional oil reservoirs.
