基于联邦学习的船舶AIS轨迹谱聚类算法研究

Federated spectral clustering algorithm for ship AIS trajectory

如何实现数据的安全共享,促进多源数据的碰撞、融合是当前学术界和产业界共同面临的重要技术挑战之一,近年来,联邦学习作为应对这一挑战的新技术受到了广泛的关注,已在智慧医疗、智慧城市建设等领域得到应用,但是在充满潜力的轨迹数据挖掘领域却鲜有研究。为了解决这个问题,提出一种安全的、分布式的基于联邦学习的谱聚类算法框架FSC(federated spectral clustering),并应用于船舶AIS(automatic identification system)轨迹数据谱聚类。该算法通过加密样本对齐技术和同态加密技术,在保证用户数据安全的前提下实现了多参与方联合训练机器学习模型。实验部分以合成数据和船舶AIS轨迹数据为样本,通过与其他聚类算法对比,验证算法具有良好的聚类性能;聚类结果能够准确提取水域船舶的主要航线,可为海事监管系统智能化提供技术支撑。

How to realize safety data sharing and promote the integration of multi-source data is one of the important technical challenges faced by academic and industrial circles.In recent years,federated learning has received widespread attention,which is a new technology to deal with this challenge.Federated learning has been applied in fields such as smart healthcare and smart city construction,but there is little research in the field of potential trajectory data mining.To solve this problem,this paper proposed a distributed and secure framework named federated spectral clustering(FSC),and applied it to the spectral clustering of ship AIS trajectory data.In the FSC framework,it used the encrypted sample alignment technology and a homomorphic encryption scheme as building blocks for the clustering algorithm,guaranteeing the security of the data in the process of federal training executed by multi-participants.To illustrate the effect of this algorithm,this paper conducted the experiments on both synthetic datasets and ships AIS trajectory datasets.The comparisons of experiments results with other similar clustering algorithms demonstrate that,besides its security advantage,this algorithm performs well in terms of clustering effect.The results indicate that the FSC can obtain the main route in the marine navigation area,which can provide specialized support for the intelligence of maritime supervision systems.