광물과 암석 (Korean Journal of Mineralogy and Petrology)

  • 제37권2호
  • /
  • Pages.23-34
  • /
  • 2024
  • /
  • 2734-0333(pISSN)
  • /
  • 2734-0538(eISSN)
한국암석학회 & 한국광물학회 (The Petrological Society of Korea & The Mineralogical Society of Korea)
권호 표지
DOI QR코드

DOI QR Code

기계학습을 활용한 XRD 광물 조성 예측 모델 개발

Development of Prediction Model for XRD Mineral Composition Using Machine Learning

  • 박선영 (한국지질자원연구원 석유에너지센터 ) ;
  • 이경북 (국립공주대학교 지질환경과학과 ) ;
  • 최지영 (한국지질자원연구원 석유에너지센터 ) ;
  • 박주영 (국립공주대학교 지질환경과학과 )
  • Park Sun Young (Petroleum Energy Research Center, Korea Institute of Geoscience and Mineral Resources) ;
  • Lee Kyungbook (Department of Geoenvironmental Sciences, Kongju National University) ;
  • Choi Jiyoung (Petroleum Energy Research Center, Korea Institute of Geoscience and Mineral Resources) ;
  • Park Ju Young (Department of Geoenvironmental Sciences, Kongju National University)
  • 투고 : 2024.05.29
  • 심사 : 2024.06.29
  • 발행 : 2024.06.30
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

퇴적물 내에서 가스 하이드레이트(gas hydrate, GH) 부존 가능성을 파악하기 위해서는 획득된 코어 시료의 광물 조성을 아는 것이 필수적이다. GH 탐사를 진행하며 채취된 울릉분지 코어 시료에서 각 시료 별 488개의 X선 회절 분석(X-ray diffraction, XRD)을 활용하여 광물 조성 값을 확보하였다. 488개의 학습 자료를 기반으로 입력값을 3100개의 XRD 피크 세기로 출력값을 12개의 광물 조성으로 기계학습을 수행하였다. 488개의 데이터를 307개의 학습자료, 132개의 검증자료, 49개의 테스트 자료로 나누어 학습을 수행하였고 랜덤 포레스트(random forest, RF) 알고리즘을 활용하여 결과를 획득하였다. 학습 결과 전문가가 예측한 광물 조성과 기계학습을 통해 예측한 값의 차이인 평균 절대 오차(mean absolute error, MAE)가 1.35%로 확인되었다. 개발된 모델 성능의 개선을 위해 주성분분석(principal component analysis, PCA)을 활용하여 XRD 피크의 핵심 특징을 추출하여 입력자료의 차원을 줄여 추가적으로 기계학습을 수행한 결과 MAE 값이 최대 1.23%까지 감소하는 것을 볼 수 있었고 시간적인 측면에서 학습 효율도 향상된 것을 확인할 수 있었다.

It is essential to know the mineral composition of core samples to assess the possibility of gas hydrate (GH) in sediments. During the exploration of gas hydrates (GH), mineral composition values were obtained from each core sample collected in the Ulleung Basin using X-ray diffraction (XRD). Based on this data, machine learning was performed with 3100 input values representing XRD peak intensities and 12 output values representing mineral compositions. The 488 data points were divided into 307 training samples, 132 validation samples, and 49 test samples. The random forest (RF) algorithm was utilized to obtain results. The machine learning results, compared with expert-predicted mineral compositions, revealed a Mean Absolute Error (MAE) of 1.35%. To enhance the performance of the developed model, principal component analysis (PCA) was employed to extract the key features of XRD peaks, reducing the dimensionality of input data. Subsequent machine learning with the refined data resulted in a decreased MAE, reaching a maximum of 1.23%. Additionally, the efficiency of the learning process improved over time, as confirmed from a temporal perspective.

키워드

과제정보

논문을 심사해주신 익명의 심사위원들께 감사드립니다. 이 연구는 한국지질자원연구원 기초연구사업 심층학습 기반 GH 저류층 분석모델 개발 사업(GP2021-010)의 지원을 받아 수행되었습니다.

참고문헌

  1. Agatonovic-Kustrin, S., Wu, V., Rades, T., Saville, D. and Tucker, I.G., 2000, Ranitidine hydrochloride X-ray assay using a neural network. Journal of Pharmaceutical and Biomedical Analysis 22, 985-992.
  2. Berndt, C., Bunz, S., Clayton, T., Mienert, J. and Saunders, M., 2004, Seismic character of bottom simulating reflectors: examples from the mid-Norwegian margin. Marine and Petroleum Geology 21, 723-733.
  3. Chen, C., Zhang, Y., Li, X., He, J., Gao, F. and Chen, Z., 2024, Investigations into methane hydrate formation, accumulation, and distribution in sediments with different contents of illite clay. Applied Energy 359.
  4. Kim, H.S., Riedel, M., Ryu, B.J., Kim, G.Y. and Bahk, J.J., 2013, Improving gas hydrate saturation estimates using P-wave velocity log data by incorporating XRD-data for detailed matrix-mineralogy definition. Marine and Petroleum Geology 47, 155-167.
  5. Kim, I.-S., Park, M.-H., Ryu, B.-J. and Yu, K.-M., 2006, Tephrostratigraphy and paleoenvironmental variation in late Quaternary core sediments of the southwestern Ulleung Basin, East Sea (Sea of Japan). The Island Arc 15, 178-186.
  6. Kvenvolden, K.A., 2002, Methane hydrate in the global organic carbon cycle. Terra Nova 14, 302-306.
  7. KIGAM, 2021, Gas Hydrate Exploration and Production Research, 26-62.
  8. KIGAM, 2022, Gas Hydrate Exploration and Production Research, 31-56.
  9. Lee, J., Byun, J., Kim, B. and Yoo, D.G., 2017, Delineation of gas hydrate reservoirs in the Ulleung Basin using unsupervised multi-attribute clustering without well log data. Journal of Natural Gas Science and Engineering 46, 326-337.
  10. Lee, J.H., Baek, Y.S., Ryu, B.J., Riedel, M. and Hyndman, R.D., 2005, A seismic survey to detect natural gas hydrate in the East Sea of Korea. Marine Geophysical Researches 26, 51-59.
  11. Lee, J.W., Park, W.B., Lee, J.H., Singh, S.P. and Sohn, K.S., 2020, A deep-learning technique for phase identification in multiphase inorganic compounds using synthetic XRD powder patterns. Nat Commun 11, 86.
  12. Lu, H. and Matsumoto, R., 2002, Preliminary experimental results of the stable P-T conditions of methane hydrate in a nannofossil-rich claystone column. GEOCHEMICAL JOURNAL 36, 21-30.
  13. Mi, F., He, Z., Cheng, L., Jiang, G. and Ning, F., 2023, Molecular dynamics simulation on methane hydrate formation in clay nanopores of edge surfaces. Applied Clay Science 243.
  14. Park, S.Y., Son, B.-K., Choi, J., Jin, H. and Lee, K., 2022, Application of machine learning to quantification of mineral composition on gas hydrate-bearing sediments, Ulleung Basin, Korea. Journal of Petroleum Science and Engineering 209.
  15. Park, W.B., Chung, J., Jung, J., Sohn, K., Singh, S.P., Pyo, M., Shin, N. and Sohn, K.S., 2017, Classification of crystal structure using a convolutional neural network. IUCrJ 4, 486-494.
  16. Ren, F., Ward, L., Williams, T., Laws, K.J., Wolverton, C., Hattrick-Simpers, J. and Mehta, A., 2018, Accelerated discovery of metallic glasses through iteration of machine learning and high-throughput experiments. Science Advances 4, eaaq1566.
  17. Ruppel, C.D. and Kessler, J.D., 2017, The interaction of climate change and methane hydrates. Reviews of Geophysics 55, 126-168.
  18. Ryu, B.-J., Collett, T.S., Riedel, M., Kim, G.Y., Chun, J.-H., Bahk, J.-J., Lee, J.Y., Kim, J.-H. and Yoo, D.-G., 2013, Scientific results of the Second Gas Hydrate Drilling Expedition in the Ulleung Basin (UBGH2). Marine and Petroleum Geology 47, 1-20.
  19. Schuetzke, J., Benedix, A., Mikut, R. and Reischl, M., 2021, Enhancing deep-learning training for phase identification in powder X-ray diffractograms. IUCrJ 8, 408-420.
  20. Schwalbach, J.R., Gordon, S.A., O'Brien, C.P., Lockman, D.F., Benmore, W.C., Huggins, C.A. and Sietronics 1996, SIROQUANT: A quantitative XRD software. Sietronics Pty Limited, Australia.
  21. Surdu, V.-A. and Gyorgy, R., 2023, X-ray Diffraction Data Analysis by Machine Learning Methods-A Review. Applied Sciences 13.
  22. Tatlier, M., 2011, Artificial neural network methods for the prediction of framework crystal structures of zeolites from XRD data. Neural Computing and Applications 20, 365-371.
  23. Torres, M.E., Trehu, A.M., Cespedes, N., Kastner, M., Wortmann, U.G., Kim, J.H., Long, P., Malinverno, A., Pohlman, J.W., Riedel, M. and Collett, T., 2008, Methane hydrate formation in turbidite sediments of northern Cascadia, IODP Expedition 311. Earth and Planetary Science Letters 271, 170-180.
  24. Uchida, T., Takeya, S., Chuvilin, E.M., Ohmura, R., Nagao, J., Yakushev, V.S., Istomin, V.A., Wang, H., Xie, Y., Li, D., Deng, H., Zhao, Y., Xin, M. and Lin, J., 2020, Rapid Identification of X-ray Diffraction Patterns Based on Very Limited Data by Interpretable Convolutional Neural Networks. J Chem Inf Model 60, 2004-2011.
  25. Zhang, Y., Li, X.-S., Wang, Y., Chen, Z.-Y. and Yan, K.-F., 2016, Decomposition conditions of methane hydrate in marine sediments from South China Sea. Fluid Phase Equilibria 413, 110-115.