2023 SPWLA PDDA Machine Learning Competition

Automatic Well-Log Depth Shift with Data-Driven Methods

Winners

	Winner Team	Contact
1st Place	Dreamstar	Fan Meng [email protected]
		Siyuan Chen [email protected]
		Yingying Ye [email protected]
		Hailong Jiang [email protected]
2nd Place	WELL-LOGGED	Hyunmin Kim [email protected]
		Heesung Kong [email protected]
		Inwook Baek [email protected]
		Jinhyeon Baek [email protected]
3rd Place	ARC_CNLC	Xuekai Sun [email protected]
		Hao Sun [email protected]
		Siyi Li [email protected]
		ZiMeng Zhao [email protected]
4th Place	All For A Dream	Yuxin Ke [email protected]
		Ankang Feng [email protected]
		Wei Hu
		Yifan Wu
5th Place	IGPL	Byunghoon Choi [email protected]
		Hojin Kang rkdghwls1010@naver
		JunHa Park

Leaderboard

Rank	Team Name	Best Score	Best Solution	Notebook
1	Dreamstar	550	MDWD+RNN(GRU)+DTW	Github
2	WELL-LOGGED	551	Random Forest+DTW	Notebook
3	ARC_CNLC	552	A Pick-Pair-Based DTW	Matlab
4	All For A Dream	559	Ridge regression+DTW	Notebook
5	IGPL	564	CNN+DTW	Package

Challenge Description

The objective of this contest is to develop a data-driven model to shift misaligned well logs to a reference GR log. Well logs of nine wells from the same field that have been depth shifted by a petrophysicist. Participants' models will be trained/validated with these nine wells and tested on misaligned well logs from other three wells from the same field.

The submission should include the shifted well logs and the shifted depths

Evaluation and Scoring Website

Please register your team at Codalab with your team leader's email. A link to the competition will be sent to you to access the scoring system. You will be given nine wells with well logs aligned, then you will train a model to perform depth shift for three hidden wells with misaligned well logs. The gamma ray log (GR) is used as the reference log, you are required to shift the other well logs (Density (RHOB), Neutron (NPHI), and Resistivity (RD)) to align with the reference GR log. You will predict the values of corrected well logs (RHOB_pred, NPHI_pred, RD_pred), and the asssocited depth shift (RHOB_dept_pred, NPHI_dept_pred, RD_dept_pred) for the three test wells. You will be scored based on the normalized mean squared error (NMSE) of your well-log prediction and the Mean absolute deviation (MAD) of your depth shift prediction.

Submissions are evaluated based on normalized mean squared error(NMSE) of corrected well logs and mean absolute deviation (MAD) of depth shift prediction, the final score will be rank transformed and averaged to avoid the scaling of different metrics.

The metrics for scoring are as follows:

$$NMSE = \frac{1}{mn}\frac{1}{Var(\mathbf{y})}\sum_{i=1}^{m}\sum_{j=1}^{n}(\hat{\mathbf{y_{i,j}}} - \mathbf{y_{i,j}})^{2}$$

where

• $\hat{y_{i,j}}$ the prediction (RHOB_pred, NPHI_pred, RD_pred) of the values of shifted well log j for sample i, $y_{i,j}$ is the actual values of the well log j for sample i shifted by a petrophysicist.
• $m$ is the sample size.
• $n$ is the number of well logs (RHOB, NPHI, RD): 3.
• $Var$ is the variance.

$$MAD = \frac{1}{mn}\sum_{i=1}^{m}\sum_{j=1}^{n}|\hat{\mathbf{d_{i,j}}} - \mathbf{d_{i,j}}| $$

where

• $\hat{d_{i,j}}$ is the prediction (RHOB_dept_pred, NPHI_dept_pred, RD_dept_pred) of depth shift of raw well log j for sample i, $d_{i,j}$ is the actual depth shift of raw well logs by a petrophysicist.
• $m$ is the sample size.
• $n$ is the number of well logs (RHOB, NPHI, RD): 3.

Note:

• Please remember to use random_state for all randomization steps, so the results are reproducible.
• RD is transformed with log10 for RMSE calculation.
• The two scores for your prediction of different well logs will be rank transformed, and averaged, e.g., if the NMSE of predictions for the values of RHOB, NPHI, and RD ranks 1st, and the MAD of assosiated depth shifts ranks 3rd, your final score would be $$score = \frac{1+3}{2}=2 $$ The team with the minimum score win the competition
• Understanding and optimizing your predictions for this evaluation metric is paramount for this competition.

Notes for submission:

1. Only one user can register for the competition per team.
2. The user name has to be exactly the same as the team name. If space is not allowed, please replace space with underscore '_'.
3. The submission file must be a zip file with three .csv files, which include your predictions of corrected/shfited well logs and their depth shift. Note: the three csv files in your submission should have the same file names as the three files in the test set, see name convention below.

• Zipped csv file name {team_name}_submission_{number of submission}.zip, which should contain (aligned_well_01.csv, aligned_well_02.csv, and aligned_well_03.csv)
• Each csv file should contain aligned well logs (NPHI_pred,RHOB_pred,RD_pred), which is calculated by shifting raw well logs (NPHI, RHOB, RD) according to reference depth (DEPT) and reference log (GR). The depth of raw well logs after depth shift (NPHI_dept_pred, RHOB_dept_pred,RD_dept_pred) also need to be included in the csv file for MAD calculation. Please refer to submission example "pdda_submission_1.zip" and don't change variable names.

4. The submission status might need a couple minutes to be updated, don't refresh the page too often.
5. The user needs to manually submit their best results to the leaderboard. Click "Participate", " Submit / View Results", click the "+" symbol in your submission. See the red circles in the attached figure.
6. Since Codalab doesn't support two scores or randk transformation, the score you see in the leaderboard is scaled for better display, the first 4 digits represent the scaled NMSE, and the last 4 digits represent the scaled MAD, the final ranking may be different from the leaderboard in Codalab after we collect all submissions. To see the actual scores for depth shift and well-log predictions, please Click "Participate", " Submit / View Results", click the "+" symbol in your submission, and Click "View scoring output log"
7. Please use version-control properly, as we need to validate your code and reproduce the results of the final submitted score in order to rank your team in the final scoreboard.
8. Max submissions per day: 3
9. Max submissions total: 100

Timeline

• March 26, 2023 - Registration starts. You must email Wen Pan ([email protected]) with team information (team name, member names, affiliations, and emails) to get the submission URL.
• March 31, 2023 - Competition starts and data releases on github.
• August 26, 2023 - Submission deadline.
• August 31, 2023 - Announce winners.
• September/October, 2023 (tentative) - Award ceremony and presentations in the special session of the SPWLA Spring Topical Conference - Petrophysical Machine Learning.

All deadlines are at 11:59 PM UTC on the corresponding day unless otherwise noted. The competition organizers reserve the right to update the contest timeline if they deem it necessary.

Competition Rules

1. Contestant can be an individual or a group with the maximum size of 4.
2. The contest focuses on data-driven methods, the use of additional data or petrophysical equations is not allowed.
3. Privately sharing code or data outside of teams is not permitted. However, it's okay to share code if made available to all participants on the competition Github repository via submitting issues or pull requests.
4. A contestant will submit the predictions for both corrected well log and the depth shift of each type of well logs for each testing wells
5. A contestant will submit the source code and a brief report documenting the accuracy achieved in a few plots.
6. The judges will review the source code.
7. A leaderboard will be updating the rank of submissions from each team, but the score and rank is subject to scaling and change.
8. The contestant with the best quality source code and the best performance will be declared the winner for this competition.

Prize Policy

• 1st Place - $500
• 2nd Place - $400
• 3rd Place - $300
• 4th Place - $200
• 5th Place - $100

Top 5 winning teams will be awarded with prizes(NOT in cash).

Note: The winners will additionally be required to provide a detailed description of their method in order to claim the prize (minimum of 2 pages double-column) by October 1, 2023.

Novel and practical algorithms will be recommended for a submission to the a SPWLA special issue by PDDA or a journal paper.

Contest Committee

Wen Pan (Architect), Michael Ashby (align well logs), Lei Fu, Yanxiang Yu, HyungJoo Lee, Chicheng Xu, Jaehyuk Lee