基于机器学习的成矿背景异常分解关键参数的自动计算

Research on automatic evaluation for key parameter of anomaly decomposition of metallogenic background based on machine learning

分形是地质矿产领域非线性计算的重要方法,GeoDAS软件是实现该方法的有效工具,因受限于软件的紧耦合性,目前难以整合人工智能等在线计算资源.为了充分发挥分形在地学大数据处理与智能分析中的作用,本文以成矿背景异常分解技术中异常下限(阈值)估计为研究目标,提出了一种利用机器学习中线性回归算法拟合数据实现阈值自动求取的方法,在ArcGIS Pro中采用Python基于机器学习框架实现了异常下限智能计算模块.采用GeoDAS软件自带砷(As)示例数据进行两种方法对比实验.实验结果表明:机器学习方法与GeoDAS的能谱密度面积(S-A)异常分解技术所得阈值的平均差小于8%,且最小差为2.35%.此外,根据最佳阈值对异常场圈定效果与软件处理后呈现效果基本一致或完全相同,验证了采用机器学习方法自动计算阈值下限方法的有效性.

Fractal plays an important role on non-linear computation of geological and mineral fields and GeoDAS is an effective software for this method. It is currently difficult to integrate online AI computing resources because its limitation of tight coupling. In order to make up GeoDAS’s drawbacks and enable fractal to act powerful on geoscience big data processing or intelligent analysis as usual or more than before. This paper aims at the key parameter estimation problem of the abnormal lower limit(threshold) in the abnormal decomposition technology of metallogenic background, it proposes an innovative automatic computation method based on machine learning. This way uses loop traversing data to find the best threshold model, and proposes a linear regression algorithm to fit the data. In ArcGIS Pro, Python is used to implement the abnormal lower limit intelligent calculation module based on the machine learning framework. to realize the intelligent calculation module of the abnormal lower limit based on the machine learning framework. Using the arsenic(As) data from GeoDAS example data as experimental input for the new method and GeoDAS. The experimental results show that the average difference of the threshold obtained by the machine learning method and GeoDAS using the Spatial-Area(S-A) abnormal decomposition technology is kept within 8% and the minimum difference is 2.35%. In addition, the delineation effect of the abnormal field according to the optimal threshold is almost the same as the effect after the software processing, which further verifies the machine learning method can contribute much for geological intelligent computing.