A dynamic revenue allocation model for trusted data spaces based on blockchain and federated learning

The integration of blockchain and Federated Learning (FL) offers promising solutions for secure multi-party collaboration. However, existing incentive mechanisms within such integrations often suffer from static reward distribution and an inability to adapt to dynamic data environments, leading to unfair contribution valuation. Therefore, this study proposes a novel Multimodal Federated Learning Framework (MFLF). Based on modal feature significance, its Dynamic Weight Allocation (DWA) mechanism autonomously adjusts computational resources, reducing feature fusion latency from 45.2 ms to 21.7 ms. A gradient adaptive compression algorithm, by analyzing the information entropy distribution of parameters, maintains 93.2% classification accuracy even at a 35:1 compression ratio. The cross-modal complementary architecture employs adversarial samples generated by Digital Twin (DT) technology for pre-training, thereby enhancing the signal-to-noise ratio (SNR) stability. Experimental validation in a simulated 5G-Advanced network environment demonstrates improved end-to-end processing efficiency. In the simulated industrial IoT environment—where vibration, temperature, and pressure sensors operate asynchronously—cross-modal feature misalignment is observed to increase average fusion latency by approximately 45% compared to ideally synchronized conditions. With total communication volume compressed to 2.6 GB, which has a 19% reduction compared to baseline. Noise resilience tests reveal a robustness coefficient ρ of 0.91 under hybrid interference scenarios, which outperforms the state-of-the-art methods by 11.2%. These advancements enable the framework to maintain a false alarm rate below 3.2% (compared to an industry-standard threshold of 8%) in applications such as fault prediction for steel rolling mills, while reducing bandwidth usage by 73%. By establishing a quantitative model correlating noise intensity with performance degradation, this work provides a feasible architecture for distributed intelligent operation and maintenance in the 6G era.