The Application of the Supervised Descent Method in the Inversion of 3D Direct Current Resistivity Data

Three-dimensional direct current resistivity inversion is vital for mineral exploration, offering detailed electrical property distribution data of subsurface resources. To overcome the traditional algorithm's reliance on initial models, its tendency to become stuck in local optima, and its low inversion resolution, this paper introduces an SDM-based 3D direct current resistivity inversion method. The Supervised Descent Method (SDM) is primarily used to solve optimization problems in nonlinear least squares, effectively capturing subsurface structural details and identifying electrical anomalies through the integration of machine learning and gradient descent techniques, thereby precisely revealing the complex electrical characteristics underground. Moreover, this study incorporates smooth regularization to enhance the stability and reliability of the inversion results. The paper demonstrates the feasibility and generalization of the SDM in 3D resistivity inversion through two sets of model calculations. A comparative analysis with traditional methods further proves the advantages of the SDM algorithm in improving inversion resolution and efficiency. Finally, applying the SDM algorithm to the Xiagalaoyi River mining area in Heilongjiang Province fully proves its optimized data processing capabilities and sensitivity to complex geological structures, providing a more precise and rapid technical approach for mineral resource exploration and assessment.