Energy A.I. Hackathon 2024

Hosts: Prof. Michael Pyrcz and Prof. John T. Foster

Architects: Elnara Rustamzade and Fehmi Ozbayrak

Home Department Chair and Hackathon Supporter: Prof. Matthew Balhoff

Sponsors

Platinum: Pioneer Natural Resources, Phillips66

Silver: Chevron, Shell, ExxonMobil, Elizabeth Huth CoatesCharitable Foundation of 1992

Bronze: ComboCurve Inc., General Electric

Coordination and Student Engagement: Rowan Halliday, Heba Abdel-Rahim, Gabby Banales

---

Energy A.I. Hackathon 2024 Problem Description

Goal: Develop a data analytics and machine learning workflow in Python to:

• Predict average pump difference during drillout based many possible predictor features for 15 wells. This includes single estimates and an uncertainty model represented by 100 realizations for each well.

Background

When a well is completed, plugs are set to isolate stages. After the well has been hydraulically fractured, those plugs need to be drilled out. For lower-pressure formations, there is an increased uncertainty in bottom hole pressure experienced during drillout. This uncertainty makes it difficult to decide what fluid system to use and makes planning, scheduling, and budgeting post-stimulation operations more difficult. Pump difference is defined as the difference between fluid pumped into the well and fluid returning to the surface. In order to prevent fluid loss we challenge the Energy A.I. Hackathon 2024 teams of The University of Texas at Austin to build a data analytics and machine learning workflow to predict average pump difference with uncertainty.

This will require:

• data analytics and evaluation of multiple data sources over a variety of features
• feature engineering including feature selection, feature transformations, and feature imputation to address missing data
• integration of domain expertise at every step
• selection, training and tuning of robust machine learning prediction models

Your model will be applied to support determination of optimum fluid type to prevent loss.

---

Available Data Files Inventory

You have the following available data:

Well Data

• HackathonData2024.csv - data on 348 unique wells information.

The features include:

Features	Description
Well ID	Unique number assigned to each well as identifier
Avg Pump Difference	Pump difference (gal/min) is the fluid pumped into the well minus the fluid that is returning to surface
Area	The field is divided into smaller areas and wells in the same area are near one-another
Lateral Length	Length of the lateral section of the well in feet
Fluid System	A fluid system represents what type of fluid that was used to drill the well out
TVD	True vertical depth
Development Strategy	How to locate the next well in the reservoir
DELAYED	Whether offset wells exist near the wellbore, from a gun barrel view
BOUND_CODE	Whether offset wells exist near the wellbore, from a gun barrel view
CODEV_POSITION	Within the co-development, the edge wells are the wells that do not have other wells on one side when viewed from above (formation does not matter)
CODEV_FORMATION_POSITION	Within the co-development, the edge wells are the wells that do not have other wells on one side when viewed from above (formation does matter)
PARENT_CODEV_1050_WELL_COUNT	The number of wells that were either put on production (POP’d) more than 60 days before the subject well was POP’d (parent wells) or were POP’d within 60 days of the subject well (codev wells) and have a hypotenuse distance of 1050ft or less
PARENT_IN_ZONE_MIN_HYPOT	The hypotenuse distance to the nearest parent well in the same formation
PARENT_OUT_ZONE_MIN_HYPOT	The hypotenuse distance to the nearest parent well in a different formation
PARENT_IN_ZONE_MIN_MAP_DIST	The map distance to the nearest parent well in the same formation
PARENT_OUT_ZONE_MIN_MAP_DIST	The map distance to the nearest parent well in a different formation
PARENT_1050_AVG_WELL_AGE	Average age (in days) of the wells that were POP’d more than 60 days before the subject well was POP’d and have a hypotenuse distance of 1050ft or less
PARENT_1050_MEDIAN_WELL_AGE	Median age (in days) of the wells that were POP’d more than 60 days before the subject well was POP’d and have a hypotenuse distance of 1050ft or less
PARENT_1050_WELL_COUNT	The number of wells that were put on production (POP’d) more than 60 days before the subject well was POP’d and have a hypotenuse distance of 1050ft or less
PARENT_3000_AVG_HYPOT_DIST	Average hypotenuse distance of the wells that were POP’d more than 60 days before the subject well and are within a 3000ft buffer around the well
PARENT_3000_AVG_MAP_DIST	Average map distance of the wells that were POP’d more than 60 days before the subject well and are within a 3000ft buffer around the well
PARENT_3000_AVG_TVD_DIST	Average TVD distance of the wells that were POP’d more than 60 days before the subject well and are within a 3000ft buffer around the well
PARENT_3000_AVG_WELL_AGE	Average age (in days) of the wells that were POP’d more than 60 days before the subject well was POP’d and have a hypotenuse distance of 3000ft or less
PARENT_3000_MEDIAN_WELL_AGE	Median age (in days) of the wells that were POP’d more than 60 days before the subject well was POP’d and have a hypotenuse distance of 3000ft or less
PARENT_3000_WELL_COUNT	The number of wells that were put on production (POP’d) more than 60 days before the subject well was POP’d and have a hypotenuse distance of 3000ft or less
CODEV_IN_ZONE_MIN_HYPOT	The hypotenuse distance to the nearest well in the same formation (within the same codevelopment)
CODEV_OUT_ZONE_MIN_HYPOT	The hypotenuse distance to the nearest well in a different formation (within the same codevelopment)
CODEV_IN_ZONE_MIN_MAP_DIST	The map distance to the nearest well in the same formation (within the same codevelopment)
CODEV_1050_WELL_COUNT	The number of wells that were put on production (POP’d) within 60 days of when the subject well was POP’d and have a hypotenuse distance of 1050ft or less
CODEV_3000_AVG_HYPOT_DIST	Average hypotenuse distance of the wells that were POP’d within 60 days of the subject well and are within a 3000ft buffer around the well
CODEV_3000_AVG_MAP_DIST	Average map distance of the wells that were POP’d within 60 days of the subject well and are within a 3000ft buffer around the well
CODEV_3000_AVG_TVD_DIST	Average TVD distance of the wells that were POP’d within 60 days of the subject well and are within a 3000ft buffer around the well
CODEV_3000_AVG_WELL_AGE	Average age (in days) of the wells that were POP’d within 60 days of the subject well and have a hypotenuse distance of 3000ft or less
CODEV_3000_MEDIAN_WELL_AGE	Median age (in days) of the wells that were POP’d within 60 days of the subject well and have a hypotenuse distance of 3000ft or less
CODEV_3000_WELL_COUNT	The number of wells that were put on production (POP’d) within 60 days of when the subject well was POP’d and have a hypotenuse distance of 3000ft or less
Pressure Gradient (psi/ft)	Average change in pressure per unit of depth
Soak Time	Hours between the end of the stimulation phase (hydraulic frac) and the start of the drillout
Avg Open Pressure	Pressure at the wellhead right before the frac pumps are turned on at the start of a stage
SD Open Pressure	Standard deviation of average open pressure
Avg Close Pressure	Pressure at the wellhead taken after the frac pumps shut down at the end of a stage
SD Close Pressure	Standard deviation of average close pressure

Problem Set

• solution.csv

The features include:

• Well_ID - anonymized, unique well identifier
• Est Pump Difference - hackathon solution to be entered by the team, indicating average pump difference for each well
• R_ - 100 realizations for average pump difference for each well

Comments:

• all the well names are anonymized (replaced with simple indices).
• NaN in the file indicate missing data.
• the source of the data is confidential. Do not attempt to determine the source of the data.

---

Required Hackathon Submissions

By January 21 at 12:00 noon each team must submit:

• Solution Table - a .csv file with your predictions (estimates and uncertainty) for the 15 wells using the template given in solution.csv in this directory.

  • the file must be named solution.csv with final values in a commit and then pushed to Github for automated scoring.

• Python Workflow and Associated Files - committed to this repository with the workflow as a Jupyter Notebook .ipynb file along with all data files required to reproduce your team's solutions. The submitted workflow Jupyter Notebook should follow the format of the provided template Hackathon_ProjectTemplate for enhanced workflow communication and code readibility.

  • the file must be named xxxx.ipynb and pushed to GitHub for review and scoring (for code readability) by the hackathon architects, where xxxx is your team name.

• Presentation - a PowerPoint slide deck .PPTX file for your team's final presentation to our judges. The submitted presentation should follow the format of the provided example presentation Hackathon_PresentationTemplate.

  • the file must be named xxxx.pptx and pushed to GitHub for review by the judges, where xxxx is your team name.

The Workflow, and Presentation templates are in the PGEHackathon/resources repository. Use the solution.csv file (in this directory) to record your solution.