The absence of proper surveillance during saltwater disposal (SWD) in shallow formations may pose drilling and completion risks for unconventional wells developed below disposal intervals. Predictive models capable of reliable pressure prediction in water disposal formations are essential to mitigate these risks. These models can also be used to optimize future SWD activities such as 1) identifying optimal injection rates and locations of new injectors and 2) minimizing drilling and completion risks by avoiding high-pressure injection locations.

Reservoir simulation has been used for SWD pressure estimation in some assets. However, setting up and maintaining a reliable reservoir model for pressure prediction can be challenging. Complex geology in the disposal formation, coupled with sparse and unreliable pressure data, makes the modeling problem ill-posed and can lead to low confidence reservoir pressure estimates.

We propose a novel SWD modeling technique that leverages the Data Physics framework with simplified but physically rigorous model representations to alleviate the drawbacks encountered in conventional reservoir simulation. These Data Physics models combine reservoir physics and machine learning into unified models that can be built, run, and updated quickly. Additionally, an ensemble-based data assimilation approach accounts for uncertainty and provides more robust pressure predictions.

To test the validity of the approach, a synthetic simulation model was created to serve as the "ground truth." We show that the new modeling technique can effectively use very little data from this simulation model to predict pressure responses due to water injection. The new modeling technique was tested and applied in the Bakken and Permian basins to predict pressure responses due to SWD activities in these areas. The model can also be used to propose new injection strategies to reduce drilling and completion risks.

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