Water flooding is an established method of secondary recovery to increase oil production in conventional reservoirs. Analytical models such as capacitance resistance models (CRM) have been used to understand the connectivity between injectors and producers to drive optimization. However, these methods are not applicable to waterflood fields at the initial stage of life with limited data (less than 2 years of injection history). In this work, a novel approach is presented that combines analytics and machine learning to process data and hence quantify connectivity for optimization strategies.

A combination of statistical (cross-correlation, mutual information) and machine learning (linear regression, random forest) methods are used to understand the relationship between measured injection and production data from wells. This workflow is first validated using synthetic simulation data with known reservoir heterogeneities as well as known connectivity between wells. Each of the four methods is validated by comparing the result with the CRM results, and it was found that each method provides specific insights and has its associated limitations making it necessary to combine these results for a successful interpretation of connectivity.

The proposed workflow is applied to a complex offshore Caspian Sea field with 49 production wells and 8 injection wells. It was observed that implementing the diffusivity filter in the models while being computationally expensive, offers additional insights into the transmissibility between injector producer pairs. The machine learning approach addresses injection time delay through feature engineering, and applying a diffusive filter determines effective injection rates as a function of dissipation through the reservoir. Hence, the combined interpretation of connectivity from the different methods resulted in a better understanding of the field. The presented approach can be extended to similar waterflood systems helping companies realize the benefits of digitization, in not just accessing data, but also using data through such novel workflows that can help evaluate and continuously optimize injection processes.

You can access this article if you purchase or spend a download.