In the smart grids, the power supplier achieves power distribution and price regulation by the real-time data aggregation of meters’ readings in each region. However, electricity readings of users usually contain sensitive information, such as living habits and lifestyles, which can leak of identity privacy of users. To address this problem, a privacy-preserving data aggregation scheme based on the additive homomorphism of shamir secret sharing was proposed which users only need to send the shares of secret data to the aggregator to complete the data aggregation operation. Meanwhile, our scheme can effectively resist collusion and man-in-the-middle attacks from external adversaries by adding random numbers to the data and binding with users’ hash signatures. The whole process does not require the participation of a trusted authority. In addition, our scheme is fault-tolerant to a certain extent and dynamic: the aggregation operation can still be completed robustly in the case of network or part of aggregators failure, and the meters can free to leave or join the system. The security analysis demonstrates that the proposed scheme satisfies the security requirements of smart grids. Performance experiments indicate superior computational and communication efficiency compared to existing schemes.