Facilitate Geoengineering and Completions with Machine Learning Methods: Case Study from Ordos Tight Oil Field

Abstract

Abstract Geoengineering is an important integrated approach for oilfield development. Conventional geoengineering focuses on geology and reservoir studies using analytical or numerical methods. A new geoengineering approach uses data mining and machine learning as input to completion optimization. Compared to conventional practice, this innovative approach allows field development optimization to be based on statistics instead of numerical models. The application of geoengineering to completion means converting geology and reservoir understanding into a form that supports the well completion engineering. Conventional geoengineering work is heavily associated with building geology models and populating reservoir property grids; these require a huge amount of data and human effort to set up and sustain. In addition, in conventional geoengineering for completion, the engineered completion design, production modeling, and pressure matching can show high levels of uncertainty. The data-driven approach incorporates data mining and machine learning methods to improve the understanding of the reservoir from wireline measurements, geological modeling, and completion practice. Because this approach is based on data and algorithms, the potential exists for improving efficiency and accuracy. In a case study, data from 130 existing horizontal wells in the Ordos tight oil field were collected for analysis. These data included logging data, well engineering parameters, and production performance data. Beyond data mining, which focused on the data cleanup and visualization, the typical machine learning methods, including analysis of variance, linear discriminating analysis (LDA), and artificial neural network (ANN), were used to improve the logging evaluation, to define the effective sand body and geology quality, to calculate effective lateral length, and to demonstrate trends leading to a completion optimization strategy. The high efficiency and accuracy of data mining and machine learning contribute to upgrading our understanding of the logging data, reservoir properties, completion engineering, and production performance. With the support from those powerful tools, we can use geological and reservoir input quantitatively for completion engineering, which enables better well planning and better production evaluation. This case shows examples of data mining and machine learning used in geoengineering for completions. Compared to conventional geoengineering, which is based on analytical and numerical modeling, this innovative approach built on data, statistics, and algorithms has the potential to be more efficient and accurate.