3D Ore Prediction by Integrating Dynamic Simulation with Machine Learning:A Case Study of the Tongshan Copper Deposit,Anhui Province,China
3D quantitative ore prediction is the most advanced frontier in mineral predictive exploration.Such prediction is also the most challenging due to the complexity of the metallogenic system.Dynamics numerical simulations and machine learning(ML)are two major approaches to complex system prediction.In this study,dynamics simulations and machine learning were combined to quantitatively predict mineral potential in 3D space for the Tongshan Cu deposit in Anhui province.The Tongshan copper deposit has been intensively explored,and discovery of new orebodies is extremely difficult due to the scarcity of easy-identified orebodies.However,the huge geological exploration dataset is favorable for dynamic simulation and ML prediction.By integrating all exploration and geological data of the deposit,we first constructed 3D geological and resistivity models of this ore deposit,which showed the complicated spatial association of the orebodies with geological factors and resistivity.Using the 3D dynamic model of the mineralization system constructed on the basis of 3D geological modeling,we ran numerical simulations of coupled multi-process dynamics by explicitly monitoring the time-step,which replayed the spatiotemporal variation of its dynamic factors and their results.Based on the 3D geological and geophysical model and the numerical simulation results,we selected eight quantitative features as variables for the ML prediction and constructed four ML models with different combinations of the feature variables.The random forest(RF)algorithm was used for ML prediction of this mineralization system.The RF computation results showed that the four models achieved perfect prediction performance for both testing and validation samples.The RF model comprising all the dynamic,geological,and geophysical features exhibited the best prediction performance.The AUC values of the test and validation sets reached 0.998 and 0.999,respectively,and its top 7%high prediction probability domains covered nearly all existing orebodies.The ML prediction results show that there is a high-potential zone at depth in the southeast part of the mining area,which can be the target for further exploration.
3D modelingdynamics simulationmachine learning predictionTongshan copper deposit
万方数据
维普
