This paper details the results of Pioneer’s pilot study with Novi on automated PDP forecasting, and their vision for a fully-automated, ML-based forecasting workflow.
|- Wednesday, July 28th at 10:50 AM | Room 361 - Theme 7: Data-Driven Production Forecasting & Optimization - AUTHORS:: I. Gupta1, O. Samandarli1, A. Burks1, D. McMaster2, V. Jayaram1, D. Niederhut3, T. Cross3 (1Pioneer Natural Resources; 2Pioneer Resources; 3Novi Labs)|
Objectives/Scope: Production forecasting is the foundation of reservoir engineering workflows, a requisite for critical tasks across an asset lifecycle. Machine learning (ML) can improve accuracy and ease of automation for production forecasting. This paper details the process and results of a ML-based forecasting pilot test. Our ML-based approach shows accuracy that parallels or exceeds that of a Modified Arps method for both pre-drill and post-drill wells. In this study, we elucidate the advantages of the ML-based approach, as well as our conception for an integrated, automated production forecasting workflow.
Methods/Procedures/Process: We compared traditional Arps-based Decline Curve Analysis (DCA) forecasts against ML-based forecasts. The latter utilized autoregressive (AR) and extra randomized trees (ETR) algorithms. This multi-variate ML forecasting workflow utilizes geology, spacing, completions variables, and historic production to forecast future production. We independently hyper-parameterized and trained models to forecast at 5 or 30day intervals, for both pre-drill and post-drill wells, and for each production stream (oil, gas and water). For post-drill wells, production data prior to the budget cycle date was split into two datasets, training and test. From those sets we built robust models to generate a series of forecasts aligned with budget
cycles. Back-testing with production acquired after the budget cycle date provided evidence that ML-based forecast accuracy outperforms or is at equal with Modified Arps method, on average.
Results/Observations/Conclusions: At the asset level in the pre-drill wells, the Modified Arps method results in superior forecasting for the first sixty days; however, ML-based forecasts result in improved accuracy beyond that timeframe. In post-drill wells, the ML-based models produce more accurate oil production forecasts. For gas and water product streams, the ML forecasts are more accurate over the first 90-120 days, though prove less accurate beyond that time frame. The higher accuracy of ML-based approach is largely attributed to its multi-variate approach, of which DCA methods are incapable. The results from this pilot study provide confidence to integrate ML methods into forecasting and reserves estimation processes.
Applications/Significance/Novelty: ML-based solutions for production forecasting can match and in certain cases exceed the performance of simple curve fitting DCA method. ML-based methods for production forecasting, pad optimization, and budget planning are promising not only for their accuracy, but additionally for their speed, automation and low cost. Reservoir engineers can utilize this application to maximize optimization workflows, critical processes in reducing costs and improving profitability of shale assets.
Interdisciplinary Components: This interdisciplinary work includes contributions from reservoir engineering, geoscience, data science, and software engineering.