Adaptive Direct Sampling-Based Approach to Ore Grade Modeling

While gaining recognition, the Multiple-Point Geostatistics (MPS) method faces limitations in its application to mineral resource reserve estimation due to a lack of standardized parameter-setting practices. To address this challenge, this paper proposes an adaptive MPS parameter optimization framework based on optimization algorithms, which is implemented by a particle swarm algorithm (PSO) and direct sampling method (DS) and successfully applied to ore grade modeling. In the framework, PSO is employed to optimize the critical parameters of DS. To ensure accurate ore grade estimation, mean square error (MSE) is used to measure the performance of the DS model under the current parameter configuration. The PSO optimization algorithm is then used to minimize the MSE value and obtain the optimal DS model parameters. The effectiveness of the proposed method is validated using real ore deposit data. The original borehole data is randomly partitioned into training, testing, and validation sets. The training set is utilized for generating MPS training images, the testing set for determining optimal parameters, and the validation set for confirming the method's generalization and stability. The entire ore body is simulated in the final step, and simulation results are comprehensively compared. The experimental results show that the proposed method can automatically optimize the MPS parameters, avoiding the tedious process of manually adjusting the parameters and, at the same time, ensuring the accuracy and stability of the ore grade valuation.