Computational Modeling And Machine Learning Prediction Of The Mineralization System In The Shizishan Copper Orefield

The Shizishan orefield is the largest Cu（Au）orefield in the Tongling ore cluster.The dynamics for the location-and scale-controlling mechansim of orebodies are still debatable because of their extremely complicated features and formation processes.For better understanding the dynamic mechanism of ore-formation and key ore-controlling factors,the three-dimensional models of the orebodies and related geological factors in the orefied were constructed by integrating all the survey data;the coupled mechano-thermo-hydrological（MTH）dynamics during the intrusion cooling process was simulated on the FLAC^3D platform;and the ore potentials were predicted quantitatively in 3D space by using of machine learning（ML）algotithms of Random Forest（RF）,Support Vector Machine（SVM）and Artificial Neural Network（ANN）based on the evaluation of association of the major geological factors with ore formation.The 3D modeling results of the mineral system present virtually the complicated 3D geological archeticture in the Shizhishan orefield.The 3D model of the magmatic intrusion,a critical ore-controlling factorin the orefield,shows extremely complicated ariation in its morphology and orientation.The uneven localization pattern of orebodies is related in some extend to such morphological and attitude variation of the intrusion.The dynamic simulation results show that the distribution of dilatant deformation produced during the intrusion’s cooling process is very uneven.The fluid focused high dilation zones,with volumetric strain≥1.2%,are spatially consistent with the know orebodies,suggesting the critical role of dilatant deformation in localization of ores.The dynamic simulation results of hypothetical modeles,with the changed boundary conditions and parameters,show that the extensional tectonic regime,the weaker shear strength of the ore-hosting stratium rocks and the low permeability of the ore-cover bed are also important for forming large-scale ore-hosted high dilation zonesThe 3D predictive models set up by the ML algorithms of RF,SVM and ANN have excellent performances in ore-potential prediction and indicate large scale targets with high-potentials at depth.The results of testing and validation on the predictions from the three different ML predictive models by the proven-mineralied and unmineralized cells show that the RF model is highly robust because of the better prediction accuracy and generalization ablitity.The RF 3D predictive model indicates definitely large-scale high potentials in the C₂₊₃ formations located at the southeast limbs of the ore-contrlolling anticline,which are high potential prospecting target areas.