Building a secure data joint analysis platform for multiple institutions, while protecting data privacy and data security is an urgent problem to be solved. Federated learning provides a feasible solution to the aforementioned problem. However, conventional federated learning frameworks require a central server to collect and aggregate models from all clients. Blockchain-based decentralized Federated Learning system brings huge communication, resulting in low efficiency of the entire system. To address the above issues, we integrate blockchain and hierarchical federated learning to improve system scalability while ensuring security. We propose a secure hierarchical federated learning framework based on FISCO group mechanism. First, we build a multi-layer and multi-group blockchain platform, and introduce multiple regulatory nodes with functions including verification, regulation, incentive allocation, and model aggregation. We also propose a malicious node filtering mechanism and incentive mechanism based on cosine similarity. The experimental results demonstrate the effectiveness and security of the framework.