Blockchain-Based Decentralized Edge Intelligence Collaboration and Adaptive Incentive Mechanism Research

Edge IIoT learning suffers from non-IID data, stragglers, and untrusted coordination that raise latency, energy, and variance. Our objective is to enable reliable, privacy-preserving, and low-latency collaborative learning for edge IIoT under these conditions. We present BEACON, which couples federated learning with a CTDE-MADDPG adaptive incentive on a permissioned blockchain, uses lightweight robust aggregation (FedProx/Krum/Trimmed-Mean), and anchors only hashes on-chain for audit while keeping model payloads off-chain. The system runs over heterogeneous LoRa/Wi-Fi links and supports optional differential-privacy noise and Byzantine-tolerant modes. On a Raspberry Pi testbed and 10–100-node simulations, BEACON improves task completion and accuracy while lowering latency and CPU/energy relative to a static-incentive approach; the blockchain overhead remains modest at a 1-s block time. We also report sensitivity to reward weights and non-IID levels and outline deployment limits with practical mitigations. Compared with a static-incentive baseline, BEACON increases task completion by 19.7% and throughput by 15.4%, while reducing latency by 24.8% and CPU usage by 22.6% under identical budgets; improvements are statistically significant.