机器学习方法在矿产资源定量预测应用研究进展

回顾了国内外在矿产资源定量预测研究领域的发展历程，对近十年来国外相关方向的文献进行了统计对比分析，结果显示机器学习方法已经成为矿产资源定量预测研究领域的热点方向，并主要在如下3个方面发挥了积极的作用:①提取和挖掘复杂数据中隐藏的难以识别的矿化信息；②致矿异常信息关联与转换；③多源地学数据的致矿异常信息融合、预测和发现矿床。对逻辑回归、人工神经网络、随机森林与支持向量机等主要机器学习算法与模型在矿产资源定量预测实践中的应用效果进行了评述，并探讨了在实际应用过程中存在的样本选择、错分代价、不确定性评价以及模型性能评价等主要问题及目前的解决方案。最后提出基于大数据与机器学习的矿产资源定量预测是未来发展的重要趋势。      

Estimation of potential distribution of gas hydrate in the northern South China Sea

Gas hydrate research has significant importance for securing world energy resources, and has the potential to produce considerable economic benefits. Previous studies have shown that the South China Sea is an area that harbors gas hydrates. However, there is a lack of systematic investigations and understanding on the distribution of gas hydrate throughout the region. In this paper, we applied mineral resource quantitative assessment techniques to forecast and estimate the potential distribution of gas hydrate resources in the northern South China Sea. However, current hydrate samples from the South China Sea are too few to produce models of occurrences. Thus, according to similarity and contrast principles of mineral outputs, we can use a similar hydrate-mining environment with sufficient gas hydrate data as a testing ground for modeling northern South China Sea gas hydrate conditions. We selected the Gulf of Mexico, which has extensively studied gas hydrates, to develop predictive models of gas hydrate distributions, and to test errors in the model. Then, we compared the existing northern South China Sea hydrate-mining data with the Gulf of Mexico characteristics, and collated the relevant data into the model. Subsequently, we applied the model to the northern South China Sea to obtain the potential gas hydrate distribution of the area, and to identify significant exploration targets. Finally, we evaluated the reliability of the predicted results. The south seabed area of Taiwan Bank is recommended as a priority exploration target. The Zhujiang Mouth, Southeast Hainan, and Southwest Taiwan Basins, including the South Bijia Basin, also are recommended as exploration target areas. In addition, the method in this paper can provide a useful predictive approach for gas hydrate resource assessment, which gives a scientific basis for construction and implementation of long-term planning for gas hydrate exploration and general exploitation of the seabed of China.     

基于贝叶斯优化机器学习超参数的滑坡易发性评价

利用机器学习模型进行滑坡易发性评价时,不同的超参数设置往往会导致评价结果的不同。采用贝叶斯算法对4种常见机器学习模型(逻辑回归LR、支持向量机SVM、人工神经网络ANN和随机森林RF)的超参数进行了优化,探索了该算法对滑坡易发性机器学习模型的优化效果。以湘中地区4县(安化县、新华县、桃江县和桃源县)滑坡易发性评价为例说明该算法的可行性与适用性。基于滑坡历史编录,确定研究区内1 017个滑坡点,并选定15个滑坡影响因子,以此构建滑坡易发性模型的训练集和测试集。利用贝叶斯优化算法对4种机器学习模型的主要超参数进行了优化,依据优化后的超参数建立了4种优化模型,并使用AUC值等指标来比较其预测能力。结果表明：经超参数优化后的4种机器学习模型预测性能均有所提高,且基于贝叶斯优化的随机森林模型表现最好。     

Comparison of different sampling strategies for debris flow susceptibility mapping: A case study using the centroids of the scarp area,flowing area and accumulation area of debris flow watersheds

The quality of debris flow susceptibility mapping varies with sampling strategies. This paper aims at comparing three sampling strategies and determining the optimal one to sample the debris flow watersheds. The three sampling strategies studied were the centroid of the scarp area(COSA), the centroid of the flowing area(COFA), and the centroid of the accumulation area(COAA) of debris flow watersheds. An inventory consisting of 150 debris flow watersheds and 12 conditioning factors were prepared for research. Firstly, the information gain ratio(IGR) method was used to analyze the predictive ability of the conditioning factors. Subsequently, 12 conditioning factors were involved in the modeling of artificial neural network(ANN), random forest(RF) and support vector machine(SVM). Then, the receiver operating characteristic curves(ROC) and the area under curves(AUC) were used to evaluate the model performance. Finally, a scoring system was used to score the quality of the debris flow susceptibility maps. Samples obtained from the accumulation area have the strongest predictive ability and can make the models achieve the best performance. The AUC values corresponding to the best model performance on the validation dataset were 0.861, 0.804 and 0.856 for SVM, ANN and RF respectively. The sampling strategy of the centroid of the scarp area is optimal with the highest quality of debris flow susceptibility maps having scores of 373470, 393241 and 362485 for SVM, ANN and RF respectively.     

Application of Dempster-Shafer theory in mineral resource potential mapping

The Dempster-Shafer theory has been successfully applied to mineral resource potential mapping in G1S environmental. In this applied form, basic probability assignment and combined basic probability assignment are applied to measuring map pattern and map pattern combination, respectively; and the environment composed of the only two singleton sets (deposit set and non-deposit set), is used for expressing the entire map area. For a subarea in which the certain map pattern combination exists, the combined basic probability assignment corresponding to the map pattern combination existing in this subarea, expresses the belief of inferring the subarea belonging to the deposit set from the evidence that the corresponding map pattern combination existing in the subarea. Thus, it may be served as a statistical index measuring the relative mineral resource potentials of the subarea. And it may be determined like 1) dividing the map area into a series of small equal-sized grid cells and then select the training sample set composed of the well-known grid cells or the entire grid cells; 2) estimating the basic probability assignments corresponding to each map pattern from the training sample set; 3) determining the map pattern combination existing in each cell, and then appling the Dempster's Rule of Combination to integrating the all basic probability assignments corresponding to the map patterns existing in the cell into the combined basic probability assignment. Mineral resource potential mapping with the Dempster-Shafer theory is demonstrated on a case study to select mineral resource targets. The experimental results manifest that the model can be compared with the weights of evidence model in the effectiveness of mineral resource target selection.     

Application of Dempster-Shafer theory in mineral resource potential mapping

The Dempster-Shafer theory has been successfully applied to mineral resource potential mapping in GIS environmental. In this applied form, basic probability assignment and combined basic probability assignment are applied to measuring map pattern and map pattern combination, respectively; and the environment composed of the only two singleton sets (deposit set and non-deposit set), is used for expressing the entire map area. For a subarea in which the certain map pattern combination exists, the combined basic probability assignment corresponding to the map pattern combination existing in this subarea, expresses the belief of inferring the subarea belonging to the deposit set from the evidence that the corresponding map pattern combination existing in the subarea. Thus, it may be served as a statistical index measuring the relative mineral resource potentials of the subarea. And it may be determined like 1) dividing the map area into a series of small equal-sized grid cells and then select the training sample set composed of the well-known grid cells or the entire grid cells; 2) estimating the basic probability assignments corresponding to each map pattern fromthe training sample set; 3) determining the map pattern combination existing in each cell, and then appling the Dempster's Rule of Combination to integrating the all basic probability assignments corresponding to the map patterns existing in the cell into the combined basic probability assignment. Mineral resource potential mapping with the Dempster-Shafer theory is demonstrated on a case study to select mineral resource targets. The experimental results manifest that the model can be compared with the weights of evidence model in the effectiveness of mineral resource target selection.     

模糊层次分析法在西南印度洋中脊46°~52°E多金属硫化物远景区预测中的应用

海底多金属硫化物作为热液活动的产物,具有较好的成矿远景和开发潜力,已成为各国海洋矿产勘探的重点。模糊层次分析法将模糊数学和专家知识相结合,是一种典型的知识驱动的矿产资源定量预测方法。利用该方法对西南印度洋中脊46°~52°E研究区进行了数据处理,结合专家经验计算9个证据层的权重值,最后由伽马算子进行综合,生成最终的海底硫化物矿床远景区预测图。通过预测-面积（P-A）图分析,确定模糊层次分析法的最优γ值为0.9。用C-A分形的方法确定了不同成矿概率水平对应的阈值,在此基础上评估了方法的预测性能。结果表明,在基于模糊层次分析法的预测模型中,ROC曲线下的面积为0.887,预测的热液点和地震点占已知的90.5%,表明该方法能够有效地预测研究区的有利成矿区域,为今后的海底硫化物勘探提供依据。      

Rapid susceptibility mapping of co-seismic landslides triggered by the 2013 Lushan Earthquake using the regression model developed for the 2008 Wenchuan Earthquake

The primary objective of landslide susceptibility mapping is the prediction of potential landslides in landslide-prone areas.The predictive power of a landslide susceptibility mapping model could be tested in an adjacent area of similar geoenvironmental conditions to find out the reliability.Both the 2008 Wenchuan Earthquake and the 2013 Lushan Earthquake occurred in the Longmen Mountain seismic zone,with similar topographical and geological conditions.The two earthquakes are both featured by thrust fault and similar seismic mechanism.This paper adopted the susceptibility mapping model of co-seismic landslides triggered by Wenchuan earthquake to predict the spatial distribution of landslides induced by Lushan earthquake.Six influencing parameters were taken into consideration: distance from the seismic fault,slope gradient,lithology,distance from drainage,elevation and Peak Ground Acceleration(PGA).The preliminary results suggested that the zones with high susceptibility of coseismic landslides were mainly distributed in the mountainous areas of Lushan,Baoxing and Tianquan counties.The co-seismic landslide susceptibility map was completed in two days after the quake and sent to the field investigators to provide guidance for rescue and relief work.The predictive power of the susceptibility map was validated by ROC curve analysis method using 2037 co-seismic landslides in the epicenter area.The AUC value of 0.710 indicated that the susceptibility model derived from Wenchuan Earthquake landslides showed good accuracy in predicting the landslides triggered by Lushan earthquake.     

遗传神经网络在地面沉降趋势预测中的应用

受地下流体资源和固体矿物开采、地面动静荷载、地下空间开发等人为因素的影响,区域性地面沉降问题越来越突出;构建区域地面沉降预测预报模型,及时发现潜在隐患并拟定合理的防治措施对保障经济社会与生态环境可持续发展方面具有重要意义。该文基于遗传算法的神经网络在地面沉降趋势预测中应用,对几种算法从训练速度、拟合水平以及预测能力方面作了比较分析。     

支持向量机与Newmark模型结合的地震滑坡易发性评估研究

Newmark位移模型是研究地震滑坡易发性的经典模型,机器学习方法支持向量机模型也越来越多的应用到滑坡易发性评估研究。本文将Newmark位移模型与支持向量机模型相结合,建立基于物理机理的地震滑坡易发性评估模型并应用于2008年汶川地震重灾区汶川县。从震后遥感影像目视解译出汶川县1900处地震诱发滑坡,并将其随机划分为70%的训练数据集和30%的验证数据集。选择地形起伏度、坡度、地形曲率、与构造断裂带距离、与水系距离、与道路距离6个因子与Newmark位移值共同作为地震滑坡易发性影响因素。利用ROC曲线和模型不确定性等指标对模型结果进行评估,并与二元统计模型频率比和多元统计模型Logistic回归的结果进行对比。结果表明：与频率比和Logistic回归模型相比,支持向量机模型的正确率最高,训练集和验证集ROC曲线下的面积分别为0.876和0.851。将模型应用于绘制汶川县地震滑坡易发性图,结果显示滑坡易发性图与实际的滑坡点位分布一致性较高,有80.4%的滑坡位于极高和高易发区。这说明支持向量机与Newmark位移方法结合建立的地震滑坡易发性评估模型有较高的预测价值,可以为滑坡风险评估和管理提供依据。     

基于深度学习的钨钼找矿靶区预测方法研究

随着矿产勘查工作由浅部矿向深部隐伏矿、由易识别矿向难识别矿发展,找矿难度日益增大,地质专家越来越重视新理论、新方法、新技术的应用。深度学习作为人工智能的前沿领域/技术,对于实现矿产资源预测“智能化预测评价”具有得天独厚的优势。本文以陕西省镇安县西部钨钼矿集区单元素化探异常原始数据为基础,提出了基于深度学习的钨钼矿产评价方法。该方法以归一化地球化学数据作为模型训练数据,通过深度学习中深度自编码网络方法实现异常值提取进而识别重点成矿有利地段,实现矿产资源找矿远景区定性预测。研究结果表明,在对957条单元素化探异常原始数据分类且做好模型标签后,整个过程在计算机的“黑盒子”中自动完成学习和预测,相较于传统预测研究方法,本文方法具有自动化程度高和客观性强的特征。此外,本文利用已知矿点构建训练数据集,采用随机森林方法对预测区进行矿产资源找矿靶区预测圈定,为进一步缩小找矿靶区范围提供科学依据。     

大石桥地区辽河岩群变泥质岩石中几种重要的变质反应及其地质意义

在大石桥地区进行１：５００００地质填图过程中，经过大量的岩石薄片观察，在江河岩群变泥质岩石中，首次发现了红柱石转变为蓝晶石和十字石转变为蓝青石的变质反应．详细的岩相学、岩石化学及矿物化学的研究表明，变泥质岩中先后发生的这两种变质反应，代表了本区变质作用演化的两个重要阶段．揭示了本区在伸展体制下地壳拉伸、减薄和收缩体制下的褶叠、推覆及之后的抗返、抬升的地球动力学过程．     

浅谈三轮区划与地理信息系统

文章简单介绍了三轮区划的工作任务和技术要求 ,以及在三轮区划工作中运用地理信息系统实现多学科矿产综合预测的工作方法。     

Research and application of mineral resources assessment by weights of evidence model based on SIG

Geological data are usually of the characteristics of multi-source, large amount and multi-scale. The construction of Spatial Information Grid overcomes the shortages of personal computers when dealing with geological data. The authors introduce the definition, architecture and flow of mineral resources assessment by weights of evidence model based on Spatial Information Grid (SIG). Meanwhile, a case study on the prediction of copper mineral occurrence in the Middle-Lower Yangtze metallogenic belt is given. The results show that mineral resources assessement based on SIG is an effective new method which provides a way of sharing and integrating distributed geospatial information and improves the efficiency greatly.     

印度尼西亚苏门答腊岛浅成低温热液型金(银)矿产资源总量预测

苏门答腊岛位于印度尼西亚西缘，是环太平洋成矿带的重要组成部分，拥有丰富的铜、金、铅、锌等金属矿产资源，对其资源总量进行评价可以为国家"一带一路"战略决策提供基础数据。在对该区进行野外调查和1:100万低密度地球化学调查的基础上，分析了小比例尺尺度下浅成低温热液型金（银）矿床的地质特征、关键控矿因素和找矿标志，构建了该类金（银）矿床的预测模型；认为该区NW向苏门答腊大断裂及其次级断裂与新生代岛弧岩浆岩具有密切的时空耦合关系，该断裂系统和新生代岩浆岩共同控制浅成低温热液型金（银）矿床的产出；已知金（银）矿床点和1:100万低密度地球化学异常可以作为快速圈定研究区有利找矿地段的标志；并据此圈定成矿远景区5个，其中A级成矿远景区3处，B级成矿远景区2处；然后，利用通用密度模型预测了不同概率下的各远景区未知矿床数，并运用蒙特卡罗模拟预测了苏门答腊岛浅成低温热液型金（银）矿产在不同概率下的金（银）资源量。      

Research and application of mineral resources assessment by weights of evidence model based on SIG

Geological data are usually of the characteristics of multi-source, large amount and multi-scale. The construction of Spatial Information Grid overcomes the shortages of personal computers when dealing with geological data. The authors introduce the definition, architecture and flow of mineral resources assessment by weights of evidence model based on Spatial Information Grid （SIG）. Meanwhile, a case study on the prediction of copper mineral occurrence in the Middle-Lower Yangtze metallogenic belt is given. The results show that mineral resources assessement based on SIG is an effective new method which provides a way of sharing and integrating distributed geospatial information and improves the efficiency greatly.     

基于Item2Vec负采样优化的专题地图产品个性化推荐方法研究

建立适用于专题地图产品检索的用户偏好推荐模型是提高专题地图质量的有效方式之一,在专题地图产品推荐场景中,存在严重的内容冷启动和评论数据稀疏问题,现有的推荐算法无法为特定类用户推荐不同特征的专题地图产品,导致用户从专题地图中获取偏好信息受到限制。因此,本文构建基于负采样的连续词袋模型和基于Word2Vec的Item2Vec相结合的用户偏好推荐方法,用于专题地图产品推荐。① 计算用户行为日志文件中交互行为数据的隐性评分,以代替专题地图推荐信息流场景中稀疏的用户评论数据;② 基于负采样的连续词袋模型提取目标专题地图的前后地图序列感知特征信息,通过控制正负样本比例为1:2,提升目标专题地图潜在评分的预测精度;③ 通过Item2Vec将带有用户行为特征信息的专题地图映射到向量空间,计算用户对专题地图的相似度矩阵,根据用户偏好程度完成推荐。在构建的专题地图评分实验数据集Thematic CMaps和4个公开验证数据集MovieLens上的测试结果表明:与LFM、Personal Rank、Content Based和SVD 4种传统推荐算法相比,本文所提方法可有效提高潜在评分的预测精度,推荐性能最高达到27.85%;与以霍夫曼采样方式的Item2Vec基础方法和YouTubeNet 2种神经网络推荐算法相比,评分预测精度有一定提高,且推荐性能不断提升,最高达到2.97%和5.78%。以经典算法奇异值分解（SVD）为例,将MovieLens-20M数据集切分后,在数据量不断增大的数据子集中,本文所用方法的评分预测精度和性能均优于SVD方法。     

基于犯罪空间分异和建成环境的公共场所侵财犯罪热点预测

机器学习是当前犯罪热点预测的主流方法,随机森林算法因需要的数据量较小、有较好的预测能力和预测精确度、且有较高的可理解度,更是被广泛应用,代表地理环境和建成环境的多源数据也被广泛用于模型改进的尝试实践中,但这些实践都只考虑研究区整体的预测精度变化情况,并未区分不同区域犯罪热点预测结果的差异及其原因。因此,本文以公共场所侵财犯罪为例,根据历史犯罪分布情况及过往犯罪热点分布规律,将研究区分为稳定高发热点网格、较高发热点网格、偶发热点网格及非热点网格这4类,并依据社会失序理论、日常活动理论和犯罪模式理论,选取城中村范围、路网密度及POI（餐饮、娱乐、商场3类设施）密度这3个具有代表性的协变量加入到随机森林预测模型中,探讨预测结果精度的变化情况。根据2017年26个双周的犯罪热点预测实验的预测结果,得到以下结论：加入协变量后,研究区整体、稳定高发热点网格及较高发热点网格的预测精度都有不同程度的提高,分区模型的精度显著高于整体模型的精度,说明考虑空间分异对提高模型精度起重要作用。     

确定性系数与随机森林模型在云南芒市滑坡易发性评价中的应用

编制科学的滑坡易发性分区图,可以有效降低灾害带来的损失。以云南省芒市为研究区,利用确定性系数模型（certainty factor,简称CF）方法计算各个因子的敏感值,作为随机森林（random forests,简称RF）的分类数据,选取合适的训练数据和最优化的模型参数进行模型预测,从而对研究区进行滑坡易发性评价分区。采用频率比方法将连续性因子离散化,从而通过确定性系数计算因子不同区间的滑坡易发性,同时利用CF先验模型,对研究区负样本进行选取。通过计算袋外误差得到最优化的RF参数,随后利用RF模型对研究区模型进行训练及预测。绘制ROC曲线和三维遥感影像对预测模型结果分别进行定量和定性评价,结果表明,所得到的模型精度为91%,优于随机抽样得到的结果。最后,采用平均基尼不纯度减少和平均准确度下降两种计算方法计算、评价了研究区各个因子的重要性。基于以上对研究区进行的滑坡易发性评价结果,可以为该区灾害风险评估和管理提供依据。