Research on Dynamic Execution Optimization of Smart Contracts Based on GRPO Algorithm

Abstract

Aiming at the core pain points such as low execution efficiency, high resource consumption, and insufficient dynamic adaptability caused by the "deploy-and-freeze" characteristic of traditional blockchain smart contracts, this paper proposes a dynamic execution optimization scheme for smart contracts based on the Group Relative Policy Optimization (GRPO) algorithm [1]. Specifically, the "group sampling + relative advantage" mechanism, which is the core of the GRPO algorithm, is implemented through two key modules: for the group sampling module, the algorithm first divides the smart contract execution state space into multiple sub-scenarios based on key feature dimensions such as transaction type, data volume, and network congestion degree, then randomly selects 3–5 candidate execution actions from each sub-scenario and forms a candidate action group by fusing actions from different sub-scenarios; for the relative advantage calculation module, instead of adopting the absolute advantage evaluation method of the traditional Proximal Policy Optimization (PPO) algorithm [2], it introduces a relative advantage function that takes the average execution effect of the candidate action group as the reference benchmark, quantifies the advantage of each candidate action relative to other actions in the group through indicators such as Gas cost saving rate, execution delay reduction rate, and task completion rate, and weights the relative advantage values to determine the optimal execution action. With the GRPO reinforcement learning algorithm as the core driving force, the scheme constructs a three-layer collaborative architecture consisting of an off-chain Artificial Intelligence (AI) decision-making layer, an oracle data layer, and an on-chain contract execution layer. Through real-time state perception, group sampling action generation, and scenario-based reward function design, it realizes the dynamic adaptive adjustment of smart contract execution strategies. Experimental results show that in typical application scenarios such as Decentralized Finance (DeFi) lending and supply chain finance, compared with traditional static contracts and optimization schemes based on the mainstream PPO algorithm, this scheme can reduce the average Gas fee by 22%~25%, lower the non-performing loan rate from 3.2% to 1.1%, and control the execution response delay within 500ms, significantly improving the execution efficiency, resource utilization, and dynamic adaptation capability of smart contracts. This research provides a new technical path for solving the problem of dynamic execution of smart contracts and has important theoretical and practical significance for promoting the efficient and trusted operation of the Web3 ecosystem.