Construction And Application Of Multivariate Information Prospecting Model Based On Deep Learning

Although mineral exploration has developed from shallow mines to deep concealed mines,and from easy-to-recognize mines to hard-to-recognize mines,the difficulties of prospecting in deep mines include less information,greater difficulty,more investment,and greater risks,making the prospecting of mines increasingly difficult.As a cutting-edge technology of artificial intelligence,deep learning has unique advantages for the realization of "smart prospecting",and can provide powerful technical support for the prediction of deep mineral resources.This article is guided by 3D visualization modeling technology and deep learning theoretical methods.First,it analyzes the current status of domestic and foreign research on deep ore prospecting and deep learning applications.Second,it builds a 3D model of geological bodies by collecting data such as sections and boreholes in the study area,And combine previous research results to extract multiple prospecting information;again,optimize the model from several aspects such as the number of hidden layers in the network structure of the DNN algorithm,the number of neurons,the learning rate and the number of iterations,and the gradient descent method.Construct a DNN-based multi-source information prospecting prediction model.Finally,the training set and test set are divided by known ore body information in the study area to verify the effectiveness of the model.The results achieved mainly include the following aspects:(1)A three-dimensional integrated data volume model is constructed.By collecting the existing historical results data of the study area,including drilling,sections,etc.,a three-dimensional solid model of the study area was constructed of rock masses,alterations,ore bodies,and chemical elements,and combined with the Kriging interpolation method to construct a three-dimensional block At the same time,based on previous research results,multiple prospecting information is extracted to form a three-dimensional integrated data volume model.(2)A deep ore prospecting model based on DNN was established.On the basis of clarifying the basic structure of the DNN algorithm,a combination of empirical formulas and experimental verification is used to optimize the core parameters of the model,such as the number of hidden layers and the number of neurons.At the same time,by comparing and analyzing the characteristics of various activation functions,this article The Re LU function is selected as the activation function of the middle layer of the model,and the output layer uses a combination of cross-entropy and Sigmoid functions to accelerate the model's convergence.(3)Realized the prediction of deep mineral resources in typical regions.Applying the model constructed in this paper to the prediction of deep mineralization in the study area,the correct rate,accuracy rate and recall rate of the model predictions were 87.26%,80.21% and 70.32%,respectively.The prediction results were modeled by 3D visualization,and compared with the actual mineralization in the mining area.From the perspective of the prediction effect,the prediction results of the model in this paper are in good agreement with the ore body position,scale and actual conditions of the mining area.At the same time,the algorithm has good convergence,which verifies the validity of the model constructed.