Federated Learning (FL) enables model training across multiple devices without requiring data to be sent to a central server, thereby safeguarding data privacy. However, FL still faces challenges such as low efficiency, susceptibility to malicious attacks, and risk of a single point of failure. Therefore, we propose a blockchain-based privacy-preserving asynchronous federated learning framework to ensure the required security and efficiency. The blockchain guarantees the integrity of model data, preventing tampering, while also resolving issues related to single points of failure and unreliable aggregation. Asynchronous learning accelerates global aggregation, and differential privacy enhances the framework’s robustness. Additionally, the entropy weighting method is employed to objectively evaluate the credibility of miners, effectively preventing malicious behavior. Numerous experiments demonstrate that, compared to current methods, our proposed framework possesses higher efficiency and performance while reliably ensuring system accuracy.