卷积神经网络及其在矿床找矿预测中的应用——以安徽省兆吉口铅锌矿床为例

大数据人工智能地质学刚刚起步,基于大数据智能算法的地质研究是非常有意义的探索性实验。利用大数据和机器学习解决矿产预测问题,有助于人们克服不能全面考虑地质变量的困难及评估当前模型在已有数据中的可靠性。元素地表分布特征量主要受原岩成分、成矿作用影响和地表过程的影响,它们携带某些指示矿体就位的信息,即矿体在地下空间就位时在地表的响应,且未在地表过程中消失。以往的地球化学勘查工作仅仅识别异常,但未能发现矿体在地表响应的成矿特征量。本文以安徽省兆吉口铅锌矿床为例,通过机器学习,利用卷积神经网络算法,不断挖掘元素Pb分布特征与矿体地下就位空间的耦合相关性。经过1000次训练后,可以得到准确率0. 93,损失率0. 28的卷积神经网络模型。这种神经网络模型就是矿体在地下就位时元素在地表分布的响应,可以用来进行矿产资源预测。应用该模型对未知区进行预测,结果显示第53号区域具有很大概率存在尚未发现的矿体。

Application of Convolutional Neural Network in prospecting prediction of ore deposits: Taking the Zhaojikou Pb-Zn ore deposit in Anhui Province as a case

The development of big data and artificial intelligent geology has just started since geological researches focusing on big data intelligent algorithm are still in a stage of exploratory experiment with significant meaning. Mineral resources prediction is one of the core tasks of artificial intelligent geology. By using big data and machine learning to solve mineral resources prediction problems, it will help us overcome the difficulties brought by the unablilties of the fully considerations upon geological variables, and assess the reliability of current models in existing data. This research takes the Zhaojikou Pb-Zn ore deposit in Anhui Province as a case to use the Convolutional Neural Network (CNN) and the logistic regression to reveal the relationships between the surface distributions of element Pb and the occurrences of the ore-bodies through machine learning. Then a CNN prediction of mineral resources model was constructed. Finally, the unknown areas were evaluated via this model. After 1000 trainings, the CNN model can reach an accuracy of 0.93 and a loss of 0.28. The prediction results of unknown areas showed that Area 53 has a large probability to find potential ore-bodies.