Research On Data Mining Technology Of Moving Target Trajectory Based On Federated Learning

With the development of technologies such as mobile internet,satellite positioning and navigation,the trajectory data of moving objects has been increasing explosively in the era of big data.How to realize secure and federal data mining of multi-source trajectory data and break the“data island” is a key and difficult problem in theory and industry.In recent years,federated learning with privacy-preserving is an effective technology to deal with this challenge.Federated learning has been applied in fields such as smart financial,smart healthcare and smart city,but there is little research in the field of promising trajectory data mining.Based on the technical line of federated learning,this paper takes the trajectory data of automatic identification system as an example,studies the secure and federal data mining of multi-source trajectory data,and explores the key technologies of clustering,classification and privacy-preserving in data mining.The main contributions are as follows:1.We propose an algorithm named federated spectral clustering algorithm for ship AIS trajectory.The collection and use of data is the basis for the sustainable development of data mining technology in the era of artificial intelligence.However,the current development of artificial intelligence technology faces two major challenges: one is the data island which is difficult to break due to the competition and privacy-preserving between different departments,as a result,data cannot be securely shared;another is the legal challenge which enhances the security management of data,privacy-preserving has become a global trend.To meet the challenges of artificial intelligence and the need for privacy-preserving,a secure algorithm based on federated learning and spectral clustering is proposed.This algorithm uses encrypted sample alignment and homomorphic encryption technologies over vertically partitioned datasets.Multi-participants can federally train a machine learning model on the premise of data security.Compared with other clustering algorithms,experiments show that,besides its security advantage,this algorithm performed well in terms of clustering effect.2.We give out an outsourced secure C4.5 decision tree algorithm based on BCP homomorphic encryption.By introducing the BCP homomorphic encryption algorithm,we propose an outsourced secure C4.5 decision tree algorithm,which can expand from two parties to multi-participants.Moreover,the experiments show that our algorithm has considerable accuracy compared with the original C4.5 decision tree algorithm while additionally preserving data privacy,and calculation time overhead on participant client decrease greatly.Security analysis indicates that our algorithm will not reduce the training accuracy,and proves the algorithm is safe under the honest-but-curious model.This algorithm can provide privacy-preserving technology for distributed data classification model by multi-participants,and solve the problem of multi-participants securely training decision tree on federated learning.3.We propose an algorithm named federated random forest classification algorithm for ship AIS trajectory.In order to improve the classification effect of AIS trajectory data and realize secure data mining of multi-participants data,we propose an algorithm named federated random forest classification algorithm for ship AIS trajectory by introducing the BCP homomorphic encryption algorithm.Under the federated learning model,the algorithm analyzes the ship's trajectory data and extracts the optimal trajectory features as the input of the model.The algorithm realizes the federal classification of four typical ships,namely fishing boat,passenger ship,cargo ship and oil tanker.The experiments through setting up multiple AIS station participants,from the accuracy and availability show that the algorithm performs well in terms of classification effect.The algorithm can decrease the calculation time overhead on participant client,realize the secure and federal data mining by multi-participants.The algorithm can be applied to ship trajectory identification and ship navigation risk analysis.