Federated Learning (FL) faces challenges in privacy protection, aggregation verification, and dynamic user management. Existing methods like differential privacy and homomorphic encryption have limitations such as collusion vulnerability or performance loss. This paper proposes a dynamically adjustable verifiable secure FL aggregation scheme using Paillier homomorphic encryption with single-masking and secret sharing to resist attacks and keep results encrypted, combined with non-linear homomorphic hashing signatures for verifiability. Experiments show it outperforms state-of-the-art methods in security and scalability.