【企画講演／募集型企画】 資源探査の動向と大規模データプロセッシング

[1K0301-05-04] 教師なし学習アルゴリズムによる鉱脈型金鉱床での富鉱部の抽出

○デサ ビトーリベロ¹、村岡 俊希¹、小池 克明¹ (1. 京都大学)

司会：久保大樹（京都大学）

キーワード：階層クラスラリング、条件付き地球統計シミュレーション、切頭型ガウスシミュレーション、金鉱化、地質モデル

The main sources of uncertainty in geological studies are scarcity and biased distribution of available data as well as the geological heterogeneity in the area of interest. Understanding this last factor is paramount for any project aiming at modelling mineral resources. To succeed in such goal, the team of specialists usually puzzles over to define the geological units and their geometries to estimate the extensions of mineralized zones in subsurface. Machine learning techniques offer a wide range of tools, e.g., clustering analysis, to optimize and speed up workflows in mining planning. The study tries to combine an unsupervised learning algorithm (agglomerative hierarchical clustering) and a conditional geostatistical simulation (truncated-Gaussian simulation) to consider the spatial correlation of the samples and provide a framework for checking existing domains. Such workflow is employed to data collected in a low-sulfidation epithermal deposit, where gold is the target resource. It is hosted in veins and poses an additional challenge because they are conditioned by geological structures. The proposed methodology satisfactorily highlights the mineralized zones as well as their extensions in the depth direction and shed light on the mechanisms of formation of the studied deposit.