及时检测出隐藏在大规模账户转账网络数据中的洗钱活动，对于防止非法资金流动和维护金融系统稳定性至关重要。面向洗钱账户检测的方法针对每个账户独立进行分析，无法有效检测大量账户协同起来借助多跳长路径转账实施的洗钱活动。因此，面向多账户洗钱子网络检测的方法成为主流。然而，该类方法目前大多只考虑星状结构或K部图结构，只能识别星状或链式等固定转账模式的洗钱活动。针对以上问题，本论文将多账户洗钱子网络检测问题转化为模式图上的子图搜索问题，主要工作如下： 

（1）提出一种洗钱活动的模式子图表达方法。首先，通过分析洗钱活动特征描述了一种洗钱活动模式子图，该子图可以覆盖不同模式的洗钱活动。其次，提取出洗钱活动数据集中所有的洗钱活动子图，通过对比洗钱活动子图与模式子图之间的结构相似性和属性相似性来判断模式子图对洗钱活动子图的覆盖情况。最后，洗钱活动合成数据集上的实验结果表明，本章提出的洗钱活动模式子图表达方法能够对洗钱活动子图进行准确表达与覆盖，验证了方法有效性。

（2）提出一种洗钱活动的模式子图搜索方法。首先，在层数和层下节点数量的约束下构建出多个模式子图，同时根据每个子图的整体属性特征分别计算得分。其次，根据得分更新全局信息素并指导蚂蚁探索新子图，通过不断迭代子图评分、信息素更新、重新构建子图这个过程，找到参与洗钱活动的子图。最后，洗钱活动合成数据集上的实验结果表明，本文提出的洗钱活动模式子图搜索方法在精确率、召回率和F1-score三个指标的结果分别为89.65%、94.13%和91.87%，验证了本文方法的有效性。 

（3）在上述基础上，设计并开发了一个洗钱活动检测系统。该系统包括账户转账网络数据管理功能、洗钱活动模式子图识别功能、洗钱活动模式子图搜索功能和系统交互功能，并将检测结果进行可视化展示。

Timely detection of money laundering hidden in data from large-scale account transfer networks is critical to preventing illicit financial flows and maintaining the stability of the financial system. Methods for the detection of money laundering accounts analyze each account independently and are not effective in detecting money laundering activities carried out by a large number of accounts with the help of multi-jump route transfers. Therefore, the method of multi-account money laundering sub-network detection has become the mainstream. However, most of these methods only consider the star structure or K-part diagram structure, and can only identify money laundering activities with fixed transfer modes such as star or chain. In view of the above problems, this paper transforms the problem of multi-account suspicious money laundering sub-network detection into a sub-graph search problem on the model graph. The main work is as follows:

（1）A pattern subgraph description method for money laundering activities is proposed.

Firstly, by analyzing the characteristics of money laundering activities, a subgraph of money laundering activity patterns was described, which can cover different patterns of money laundering activities.Secondly, all the money laundering activity subgraphs in the money laundering activity dataset are extracted. The coverage of the money laundering activity subgraph by the pattern subgraph is determined by comparing the structural similarity and attribute similarity between the money laundering activity subgraph and the pattern subgraph.Finally, the experimental results on the synthetic data set of money laundering activities show that the money laundering pattern subgraph expression method proposed in this chapter can accurately express and cover the money laundering subgraph, and verify the effectiveness of the method.

（2）A pattern subgraph search method for money laundering activities is proposed in this paper. Firstly, multiple pattern subgraphs are constructed under the constraints of the number of layers and the number of nodes under the layers. Meanwhile, the scores are calculated respectively according to the overall attribute characteristics of each subgraph.Secondly, update the global pheromone based on the score and guide the ants to explore new subgraphs. Through the process of continuously iterating subgraph scores, pheromone updates, and reconstructing subgraphs, find the subgraphs involved in money laundering activities.Finally, experimental results on the synthetic money laundering dataset show that the proposed pattern subgraph search method achieves a precision of 89.65%, a recall of 94.13%, and an F1-score of 91.87%, thereby validating the effectiveness of the proposed method.

（3）Based on the above, a money laundering detection system is designed and developed. The system includes account transfer network data management function, money laundering activity pattern sub-graph recognition function, money laundering activity pattern sub-graph search function and system interaction function, and the detection results are displayed visually.