Abstract
Production water reinjection (PWRI) has been a mandatory practice for waste disposal for economic and environmental reasons. However, PWRI is associated with severe injectivity decline. Therefore, maximizing injection rates under PWRI while keeping wells' injectivity above a minimal level is a real challenge. Preserving the well's injectivities is vital to prevent frequent cleaning operations and avoid drilling new disposal wells. Human-based control strategy has proven ineffective under PWRI, as it gives enough time for severe injectivity impairment to develop. Therefore, the dynamic and potentially nonlinear nature of PWRI is ideal for a data-driven strategy. Fuzzy Logic (FL) can be the natural transition from human to automatic control. FL synthesizes all injection dynamics in simply linguistic interpretable terms and implements control rules easily explainable to human operators. Besides, it can handle nonlinear dependencies intrinsically. FL and other Machine Learning (ML) methods have been applied to regular waterflooding problems, such as optimizing well positions to maximize injected volumes or pressure maintenance. However, the issue of injectivity maintenance under PWRI has not yet tested the power of these technologies. This paper proposes a data-driven strategy based on ML and FL to determine injection rates for wells in a sandstone reservoir. It establishes an automatic intelligent, adaptive, and folly interpretable system to determine the best possible flow rates for any number of wells while preserving wells' injectivity. In addition, we propose proxy models to identify the complex dynamical behavior of the wells instead of time-consuming reservoir simulations. A pilot experiment with several wells evaluates the proposed model. Our results demonstrate the potential of the method to deliver higher injection rates reducing severe injectivity losses.
NSTL