| Mobile edge crowdsensing,as a new network sensing mode that perfectly fits the large-scale mobile sensing task scenario mode under the background of "everything connected" in the Internet of things.It is an important part of building a smart city in the future.The core of smart city is smart city people.Sensing user data would drive market changes with great value orientation.However,how to dynamically measure the degree of privacy protection required for sensing user data,how to build an efficient and personalized privacy protection scheme for sensing data,and how to effectively protect the privacy leakage in the mode have not been effectively solved.In view of the above problems,this dissertation focuses on the privacy protection theory of user-oriented sensing data from three aspects: dynamic privacy measurement,personalized privacy protection and multi-phase fine-grained analysis of privacy risk based on machine learning and game theory.The main research works as follows:(1)Dynamic privacy measurement and sensing data uploading.In mobile edge crowdsensing,the privacy content of the data collected by users performing sensing tasks is dynamic and not intuitive,and the data uploading also lacks privacy risk warning value.A dynamic privacy measurement model is proposed.In this model,the privacy perception data is obtained by matrix digitization,and combined with privacy attribute preference and timeliness factors to achieve weight superposition.Based on the weighted matrix,the personalized privacy threshold is calculated reasonably,and the differential privacy is processed.A privacy measurement model evaluation mechanism is designed to help evaluate the privacy quantification effect,privacy protection degree and data utility of the model.According to the given example,a data utility of approximately 0.7 can be achieved,and the degree of privacy protection can be significantly improved as the noise level increases.In addition,combined with reinforcement learning,a dynamic privacy sensing data uploading scheme is proposed to help sensing users maximize data payoff within their personalized privacy threshold and find a balance between data payoff and privacy threats.(2)Personalized privacy protection of sensing data.Aiming at the potential privacy leakage problem caused by the loss of control of sensing users after uploading data,and features in mobile edge crowdsensing network,based on the random forest classifier and K-anonymity algorithm,this dissertation proposes a classification anonymity algorithm for spatiotemporal sensing data,analyzes and trains big data based on artificial intelligence algorithm,and then constructs a classification anonymity model combined with cryptography method.Experimental results show that the accuracy,AUC value and recall rate of the proposed model can reach 80%,75%,60% and above respectively,which has good model performance and improves the efficiency of sensing data privacy processing.(3)Privacy risk analysis of the whole life cycle of sensing data.In view of various external factors in the mobile edge crowdsensing network environment,such as adversaries with different attack capabilities,potential dishonest behaviors of edge nodes and cloud service providers,etc.Considering the multi-phase privacy leakage risk of sensing data in the uploading and subsequent trading process,based on game theory,this dissertation studies the multi-phase data privacy leakage analysis model for three-party game,analyzes the privacy leakage under single-stage game and repeated game in practical application scenarios,and the constraint conditions and norms of multi-party game players' choice of honesty behavior are obtained. |
PDF Full Text Request