| With the rapid development of Internet of Things and intelligent sensor equipment,revolutionary changes have taken place in many fields,including electronic commerce,health care,environmental monitoring,transportation and energy.These low-cost universal sensors produce and collect a large amount of information and data,that provides a good platform for the development of data technology such as machine learning and data mining.At the same time,IP visualization equipment with GPS positioning function can digitize the geographical location of people and objects.People can use location-aware applications for map navigation,location sharing of social software,real-time traffic information query.Although location information provides consumers with high-quality personalized services,these real-time record of users' location and mobile trajectory data will be analysised by malicious attacker on the Internet cloud,which poses a great threat to personal privacy.Therefore,it is necessary to design new solutions and algorithms to deal with the challenges of location big data privacy leakage based on the existing research results of location privacy protection.Differential Privacy Protection(DPP)is a new paradigm independent of prior knowledge of opponents.It can protect sensitive data by adding random noise to make the data slightly distorted inside and keep the external statistical characteristics unchanged.As an important tool for data analysis and processing,cluster analysis can extract valuable knowledge and rules from a large number of data without direct association.According to the structural characteristics of location big data,this paper proposes a differential privacy clustering method for large location data,combining the advantages of differential privacy and clustering analysis.The main tasks are as follows:(1)Comparing and analyzing the popular location privacy protection technology in recent years,the location differential privacy protection model is developed with their performance advantages and disadvantages.In which,the pre-processing method of mixed location large data is given,and the differential privacy budget is reasonably allocated.(2)In this paper,a dimension reduction clustering algorithm RD-means for location big data is proposed,which introduced the concept of synchronous trajectory distance to divide the clustering cluster.Then the distribution of data points in the clustering cluster is used to configure the objective function of feature weight parameters.According to the nearest search strategy,the optimal center point is found and the overlapping data is replaced between clusters.As a result,the dimension and redundancy of location data are greatly reduced.(3)In the environment of precondition,a differential privacy clustering algorithm DPKD for location big data is designed.In algorithm,k random elements are selected and Laplace noise is added to offset the center point,so that it can satisfy the differential privacy protection mechanism.And the clustering results are also efficient.(4)In view of the uneven distribution of location data and non-location information records,an improved differential privacy clustering algorithm Op-DPKD is proposed.With the original clustering center as the reference object,a new element point is selected as the clustering center for comparison.The relative error comparison is introduced to traverse the optimal clustering center and subtract it.Finally,the distance error caused by the sensitivity of the random center to the initial position is reduced,and the better stability and clustering effect are obtained. |
