This challenge aims to use time series injection information and monitoring data on a carbon capture well to predict carbon capture well injection rates deltas. Correlating the change in injection rate to the behavior of other parameters in the well can be used to provide a checkpoint against carbon migration from the well or other losses during the process.  The code developed to predict injection rate deltas based on monitoring well data can be used to validate carbon containment throughout the injection of the well.

If you missed the challenge launch event on April 7th, you can catch up here: Challenge-Launch-Event-April-7

Background

The Illinois Basin - Decatur Project is aimed to demonstrate the capacity, injectivity, and containment of carbon storage in the Mount Simon Sandstone, which is the main carbon storage resource in the Illinois Basin and the Midwest Region. Injection began in 2009 into the Mount Simon Sandstone at a depth of approximately 7,000 feet (2,100 meters). During this three-year period, a substantial amount of data was collected.

To monitor temperature changes, a distributed temperature sensor (DTS) fiber optic cable was installed in the tubing, extending to a depth of 6,326 feet and recording temperature every 1.624 feet every 5 seconds. Utilizing the first two years of injection data (rate, pressure, temperature) from the injector and the corresponding fiber optic DTS temperature profile from the observation well, we aim to apply machine learning methods to predict the injection rate delta. The predicted injection rate deltas determined from the monitoring well, fiber optic data and other injection well attributes can be compared to the actual injection rate as an additional way to verify reservoir integrity.

The table 1 below has a list of the meta data describing the data captured from the injection well, reservoir using fiber optic cables and the monitoring well. Contestants are to use these data to predict the injection rate delta (inj_diff). Where: IBDP_RTAC_1hrData (i) - IBDP_RTAC_1hrData (i+1) = inj_diff

Table 1 - Variables in Train and Test Datasets

The monitoring wells are located at different locations from the injection well.   Contestants can consider other factoring that may impact the accuracy of the prediction. 

Topsoil analysis was done to verify that no carbon migration occurs to the surface. 

The diagram below shows a screen image of the Illinois Basin Decatur Project system overview with the list of connected sensors.  This diagram and additional details about this diagram can be found in a white paper attached to the Starter Notebook files on the Data Tab.

Figure 1 - Overview Illinois Basin Decatur Project subsurface.

Observations and suggestions based on this simplified machine learning challenge are highly encouraged to be shared with the committee to enhance our understanding of the effect of CO2 injection and migration in the subsurface.

Data Source

All data and documents reference in this challenge were provided by Department of Energy - https://edx.netl.doe.gov/reference-shelf/faqs/

UPDATE: APRIL 10

An issue was discovered with the variable "Avg_PLT_CO2InjRate_TPH," which could be used to back out the answer key. The Train and Test data have now been updated. If you've downloaded the Train and Test before today, please delete your copies and download a fresh copy from the Data Tab.

Solutions that use the variable "Avg_PLT_CO2InjRate_TPH" will not be considered for final evaluations. Please ensure your algorithm isn't trained on this variable.

Registration

Contestants must be registered at SPE GCS link below to be evaluated for the challenge. 

https://www.spegcs.org/events/ 6655/

Final Submission

Contestants selected for final submission must upload their submission to a personal GitHub account and designated it with an Apache 2.0 license. The contestant must then must allow access to the SPE Gulf Coast Section's Github page (https://github.com/SPEGCS). Contact SPEMLChallenge@gmail.com if there are any questions.  The results are submitted to the XEEK.org platform for preliminary scoring.