基于机器学习的多金属矿成矿预测和评价  

Machine learning based prediction and evaluation of polymetallic mineralization

作　　者：孙富海 SUN Fuhai(Hubei Bureau of Coal Geology,Wuhan 430075,Hubei,China)

机构地区：[1]湖北煤炭地质局,湖北武汉430075

出　　处：《矿产勘查》2024年第S2期78-84,共7页Mineral Exploration

摘　　要：针对成矿预测领域中负样本开采难度大、工作量大,导致成矿预测效果不理想的问题,本文将机器学习技术引入成矿预测研究中,构建了基于PU算法的多金属矿成矿预测模型。首先对模型整体框架进行设计,然后构建数据集对模型进行训练与测试。测试结果表明:基于PU的成矿预测模型收敛速度快,损失值仅为0.08,在测试集上的平均准确率为92.5%,AUC值为0.96,成功预测34个矿点,且不存在非矿点样本,最接近实际矿点数量,仅有1个矿点未成功识别。与基于PCA、RF、SVM的成矿预测模型相比,整体性能良好,识别能力强、分类效果最佳,预测结果最可靠,在样本数据欠缺的情况下,可以获得良好的成矿预测效果。In response to the difficulty and heavy workload of negative sample mining in the field of mineraliza-tion prediction,which leads to unsatisfactory mineralization prediction results,this paper introduces machine learn-ing technology into mineralization prediction research and constructs a multi metal mineralization prediction model based on the PU algorithm.Firstly,design the overall framework of the model,and then construct a dataset to train and test the model.The test results show that the PU based mineralization prediction model has a fast convergence speed,a loss value of only 0.08,an average accuracy of 92.5%on the test set,and an AUC value of 0.96.It suc-cessfully predicted 34 ore points without any non ore point samples,which is closest to the actual number of ore points.Only one ore point was not successfully identified.Compared with mineralization prediction models based on PCA,RF,and SVM,the overall performance is good,with strong recognition ability,the best classification effect,and the most reliable prediction results.In the case of insufficient sample data,good mineralization predic-tion results can be achieved.

关 键 词：机器学习 PU算法 成矿预测 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]