Blockchain-enabled verifiable federated learning with multi-dimensional incentive and dynamic consensus for privacy-sensitive domains

Federated Learning (FL) has emerged as a core privacy-preserving paradigm to address data silos in privacy-sensitive domains. However, its practical deployment in heterogeneous cluster computing environments remains constrained by inherent challenges, including the trade-off between gradient privacy and credibility, irrational incentive distribution mechanisms, and inefficient integration of blockchain and FL. Current solutions typically adopt simplistic technology stacking, resulting in substantial verification overhead and limited scalability in large-scale distributed settings. To address these issues, this paper presents BlockFL-MD 2 C, a blockchain-enabled verifiable FL framework that integrates a lightweight verifiable aggregation protocol, a heterogeneity-aware multi-dimensional incentive model, and an FL-adaptive dynamic consensus mechanism. Extensive experiments conducted on MNIST, CheXpert, and Credit Card Fraud datasets validate that the proposed framework effectively mitigates verification overhead and on-chain latency, while achieving superior model accuracy and accelerated convergence. Furthermore, BlockFL-MD 2 C exhibits strong robustness against malicious nodes, excellent adaptability to large-scale heterogeneous clusters, and broad applicability in healthcare and financial scenarios. This work provides a feasible and scalable solution for the practical deployment of FL in privacy-sensitive cluster computing environments.