传统三维地质建模主要依靠地质、钻孔等资料建立地下空间地质体的三维地质结构，然而钻孔数量少以及地质信息不足，使传统地质建模存在不确定性。重磁资料反演可以获得地下地质体的空间分布，但多解性较强。项目将多点地质统计学引入到重磁三维随机反演中，通过数据样板扫描训练图像获得先验概率，产生大量的等可能的岩相模型。再利用聚类分析方法对初始岩相模型进行模式识别和分类，筛选出空间相似性高的模型。根据这些岩相模型，从地面和钻井岩样物性参数的累积概率分布中逆采样获得岩石物理模型，计算相应的重磁异常。最后用概率扰动反演策略来优化反演过程，减少多点地质统计学随机模拟的次数，提高反演的效率，从而实现多点地质统计学建模与重磁异常随机反演的有效融合。方法以辽东青城子矿集区为例，获得该地区3000米以浅成矿地质体三维地质模型，为深入研究辽东地区中生代金多金属成矿作用提供可靠的地球物理依据。

Traditional three-dimensional geological modeling mainly relies on geological and borehole data to establish the three-dimensional structure of underground bodies. However, the number of boreholes is limited and the geological prior information is insufficient, which makes the traditional geological modeling results uncertain. The gravity and magnetic data inversion can provide the spatial distribution of underground geological bodies, but remains non-unique. By introducing multi-point geostatistics into the three-dimensional stochastic inversion of gravity and magnetic data, we can calculate the prior probability through scanning the training image by the data template, and generate a large number of lithofacies models. Then the cluster analysis is used to identify and classify the initial lithofacies model, and the model with high spatial similarity is screened out. The rock physical properties, namely density and susceptibility, are imported into the selected models with the inverse sampling results from the cumulative probability distribution of the corresponding physical parameters of surface and drilling samples. After the theoretical gravity and magnetic anomalies are calculated, the probabilistic perturbation inversion strategy is carefully conducted to optimize the inversion process, and reduce the number of stochastic simulation of multi-point geostatistics. It can greatly improve the inversion efficiency, and obtain high quality inversion results. The proposed methodology effectively integrates multi-point geostatistics modeling with stochastic inversion of gravity and magnetic anomalies, and presents a reasonable solution to reduce the non-uniqueness and uncertainty in three-dimensional geological modeling. Finally, the proposed inversion strategy will be adopted to deal with the gravity and magnetic data of the Qingchengzi ore deposits in eastern Liaoning Province. The 3-D model of the ore-forming geological body within a depth of 3000 meters obtained from the inversion results will provide a reliable geophysical understanding for further research on the Mesozoic gold and polymetallic mineralization in eastern Liaoning Province.