Research On Key Technologies Of Data Aggregation Without Trusted Authority

Emerging technologies such as big data,cloud computing,and artificial intelligence has given full play to the potential value of Internet of Things(IoT)data in many fields such as energy,finance,transportation,and medical care.However,IoT data faces serious security and privacy issues when it is published and utilized.Privacy-preserving data aggregation technology can realize the release of the sum or average value of data in a specific area,while maintaining the statistical characteristics of all data in the area without revealing a single data value.It provides a way of thinking to solve the contradiction between ’availability’ and ’privacy’ in data release,and in a sense realizes the ’available and invisible’ of data.Therefore,privacy-preserving data aggregation technology has become a research hotspot in recent years.However,most of the existing data aggregation solutions need to rely on the help of one or more trusted third-party authorities to achieve privacy protection.Although many entities are assumed to be trusted authorities in reality,this assumption is not reliable.Because operators in trusted authorities are not always trusted,a large number of data security and leakage incidents indicate that data release cannot rely on any trusted authority.At the same time,the existing privacy-preserving data aggregation schemes still have some key problems in reliability,practicability,and efficiency in practical application scenarios of the IoT that need to be solved urgently.Therefore,research on privacy-preserving data aggregation technology that does not rely on trusted authorities has important theoretical significance and broad application prospects.This dissertation studied the key technologies in several parts,such as fault tolerance of privacy-preserving data aggregation,multi-subset data aggregation,data aggregation with optional privacy,and application of privacy-preserving data aggregation in federated learning-based electricity theft detection.The contents and innovations are as follows:1)Fault-tolerant and privacy-preserving data aggregation without trusted authority.In view of the fact that IoT smart devices may suffer from malicious damage or temporary power outages and network failures,most of the existing data aggregation solutions cannot upload their data in time when a smart device fails,all normal devices’ data cannot be collected and aggregated,a data aggregation scheme based on Shamir threshold secret sharing and Lifted EC-ElGamal homomorphic encryption was proposed.Through Lifted EC-ElGamal homomorphic encryption technology,the data of smart devices was privately processed and released,and the control center can aggregate the ciphertexts of all smart devices and decrypt the exact sum of data without obtaining a single data.Through the secret sharing between smart devices during system initialization,the decryption parameters of each smart device were backed up on other smart devices.When a fault occurs,the decryption parameters of the faulty device can be restored through the backup information,fault-tolerant data aggregation without trusted authority was realized.2)Privacy-preserving multi-subset data aggregation without trusted authority.In the existing data aggregation solutions,the control center can only obtain a single sum or average value of all user data,resulting in limited use value in smart grids,which cannot meet the needs of smart grids for more fine-grained multi-dimensional data.A multi-subset data aggregation scheme based on Paillier homomorphic encryption,super-increasing sequences,and key exchange protocol was proposed.In the initialization phase,each user negotiated with any other user based on the key exchange protocol to obtain a security key for encryption.Through Paillier homomorphic encryption and super-increasing sequences,the user’s electricity consumption data was privately processed and the consumption interval was selected,the control center can aggregate the sum data and the total number of users distributed in different power consumption intervals without obtaining individual data.3)An optional privacy-preserving data aggregation scheme without trusted authority.Aiming at the problem that all users or smart devices choose a data collection method with privacy protection to publish their data by default in existing data aggregation schemes,otherwise the data aggregation process cannot be completed,which limits the release of data,an optional privacy-preserving data aggregation scheme was proposed.Users can choose different privacy processing methods to publish their data according to their privacy sensitivity and interests.Using the homomorphic nature of the BGN encryption system,the control center can aggregate the sum of data of all users.In addition,the control center is also able to decrypt the individual data of users who have not opted for privacy encryption.An efficient and secure blind factor update protocol is designed.The data encrypted based on the updated blind factor can be guaranteed not to be decrypted by the control center,so as to realize the privacy protection of users who choose privacy encryption.4)A federated learning-based privacy-preserving electricity theft detection framework.Aiming at the problem of user privacy leakage in the existing electricity theft detection methods based on plaintext data and the overfitting or lack of accuracy of the electricity theft model trained in independent mode,a new privacy-preserving electricity theft detection framework based on federated learning and inner product function encryption was proposed.The inner product function encryption technology is used to publish the user’s electricity consumption data,so as to ensure that the user’s privacy is not leaked,and at the same time,it can realize accurate electricity theft detection based on deep learning.Secondly,multiple detection centers or power departments published the model parameters trained based on the local user’s electricity consumption data set,and the cloud server performed the aggregation and update of the model parameters based on the homomorphism of the inner product function,which significantly improved the model accuracy of each independent detection center or power department.