数学地球科学跨越发展的十年:大数据、人工智能算法正在改变地质学

The Great-leap-forward Development of Mathematical Geoscience During 2010-2019:Big Data and Artificial Intelligence Algorithm Are Changing Mathematical Geoscience

近十年是科学研究从问题驱动向数据驱动转变的转折时期,科学研究的第四范式—数据密集型科学发现应势而生。这期间,大数据与人工智能算法的引入使数学地球科学实现跨越式发展,并正在改变地质学。机器学习是使计算机具有智能的根本途径。深度学习,即多层神经网络的方法,是一种实现机器学习的技术,是过去几年大数据与数学地球科学研究的最重要的热点。贝叶斯网络是贝叶斯公式和图论结合的产物,可用来建立矿床地质的成因网络,进而理解矿床成因。地质大图形问题可以转化为大型的复杂网络空间问题和社区结构问题,社区分析技术可用于地震预报、地质网络分析、特殊地质现象识别、矿床预测。关联规则和推荐系统算法在地质研究中已有成功的应用实例。化探数据及其异常经常包含复杂和非线性模式,深度学习在智能识别与提取复杂地质条件下地球化学异常具有优异的能力,卷积神经网络、堆叠自编码机等是较为常用和有效的方法。非线性矿产资源预测、基于GIS和三维地质建模的三维成矿预测及相应的软件系统得到持续改进。三维虚拟仿真建模技术的应用实现了多模态、跨尺度地学虚拟现实与多维交互,地质过程数值模拟等已有创新性进展。区块链技术以及OneGeology、玻璃地球、深时数字地球等大地质科学计划,将在整合全球地质大数据、共享全球地学知识、推动数学地球科学学科发展方面起到重大的推动作用。

The last decade was a historic transitional period for the transformation of scientific researches from the question-driven to data-driven. The fourth paradigm of scientific research, namely data intensive scientific discovery, has emerged in this period.The introduction of big data and artificial intelligence algorithms into the mathematical geoscience has resulted in a great-leap-forward development for mathematical geoscience in the period and is changing research patterns of geoscience. Machine learning is a fundamental way to endow computer with intelligence. Deep learning, as the multi-layer neural network algorithms, a technology for the implement of machine learning, was the most important hotspot in the study of big data and mathematical geoscience in the past few years. The Bayesian network from the combination of the Bayesian formula and the graph theory can be used to establish genetic network models of deposit geology for further understanding the mechanism of deposit formation. Big geological graph data problem can be converted to large complicated network space and community structure issues. Community detection algorithms can be applied to the earthquake prediction, geological network analysis, special geological phenomena identification and deposit prediction. The association rule and the recommendation system algorithms have been successfully applied in the geological research. Geochemical exploration data and the related anomalies generally contain complex nonlinear patterns. Deep learning is very powerful for the intelligent recognition and extraction of geochemical anomalies from various data under complicated geological conditions. Convolution neural network and stacked autoencoder are among effective methods frequently used. The methods for nonlinear mineral resources prediction, 3 D metallogenic prediction based on GIS and 3 D geological modeling, and corresponding software systems have been continuously improved. The 3 D virtual simulation modeling technology has been applied to have realized multi-modal, trans-scale virtual reality and multi-dimensional interaction in geoscience. Innovative progresses have also been made in the numerical simulation of various geological processes. The Blockchain technology together with big geoscience programs, such as the OneGeology, Glass Earth and Deep-time Digital Earth, will play very important promoting roles in fields of integrating global geological big-data systems, sharing global geoscience knowledges, and promoting the development of mathematical geoscience discipline.