Blockchain sharding scheme based on generative AI and DRL: Applied to building internet of things

Abstract

When applying blockchain sharding technology in the building Internet of Things (IoT) domain to enhance the throughput performance of the blockchain, cross-shard transactions triggered by device collaborative tasks have increasingly become a prominent issue. Existing solutions base their shard division on historical transaction moments, using the outcomes for future transaction processing. However, since the historical interaction characteristics do not accurately reflect the interaction details within specific fine-grained time periods, this leads to poor system performance. Additionally, the parameter configuration in blockchain sharding systems is mostly based on arbitrary or default settings, which also results in unstable system performance. To address these two challenges, this paper proposes a blockchain sharding scheme called AI-Shard. Firstly, the system includes a module, G-AI, that utilizes generative AI to predict future node interaction relationships, enabling more proactive and adaptive shard division based on the predicted interaction matrix. Secondly, the system integrates a reinforcement learning module, DL-AI, specifically tailored for configuring parameters of the blockchain sharding system, such as the number of shards, block size, and block interval, to automatically optimize them, aiming to further enhance the system's throughput. Experimental results show that AI-Shard can reduce the proportion of cross-shard transactions and improve the system's throughput.