Edge Computing Network Privacy Protection Method Based On Federated Learning

Mobile Crowd Sensing(MCS)is a new sensing data collection paradigm in the field of the Internet of Things.It uses various sensors built into users’ smart mobile devices to collect massive data conveniently and quickly,which is widely used in various life scenarios.However,the sensing data collected by users often contains various important privacy information,which will face serious privacy leakage problems.The existing research methods mainly use various privacy protection technologies to protect users’ sensitive information under the traditional mobile crowdsensing network architecture,and most of them study users’ privacy security at the level of original data.Based on this,this paper introduces edge computing into the mobile crowdsensing network to construct a distributed Edge Computing Network(ECN),to separate users and the sensing platform and avoid centralized leakage of sensing data.Under this architecture,two privacy protection methods are mainly studied to further solve the problem of user privacy data leakage.This paper mainly does the following works:(1)Aiming at the privacy security problems such as personal data information leakage when users collect and submit sensing data in the edge computing network,this paper proposes a federated learning privacy protection method based on local differential privacy.Firstly,participants use federated learning to train sensing data locally to obtain local models,avoiding interaction with edge computing nodes and the sensing platform.Then,the trained model parameter values are subjected to noise perturbation using local differential privacy and uploaded to edge computing nodes.Finally,edge computing nodes perform edge aggregation on the noisy model parameters and upload them to the sensing platform to complete the global aggregation operation.(2)Aiming at the privacy leakage problem in the process of sensing data transmission by users and aggregation of edge computing nodes and the sensing platform in the edge computing network,this paper proposes a federated learning privacy protection method based on homomorphic encryption.Firstly,participants use the private key issued by the key generation center to decrypt the model parameters and perform the local training update operation.Then,the local model parameter values are encrypted by the Paillier cryptosystem in homomorphic encryption and uploaded to edge computing nodes.Finally,edge computing nodes and the sensing platform complete the secure edge and global aggregation operations in the ciphertext form.(3)The effectiveness of the two federated learning privacy protection methods is verified by simulation experiments.The experimental results show that the privacy protection method based on local differential privacy can achieve the balance between privacy protection degree,model accuracy,and data availability according to the user’s privacy requirements.The privacy protection method based on homomorphic encryption can obtain better model performance and less time cost while protecting user data privacy.This paper also combines the two methods to propose a privacy protection method based on local differential privacy and homomorphic encryption,and conducts comparative experiments and privacy protection level analysis to verify the privacy and security of the method.