Reputation-Based Blockchain Sharding Scheme for Urban Rail Transit System: A DRL Approach

With the rapid development of Urban Rail Transit (URT), there is an increasing demand for secure and efficient data transmission and computation. Blockchain provides a promising approach to meet these needs. However, inherent limitations in blockchain scalability significantly impede its performance, particularly in large-scale URT network environments with high real-time demands and mobility. To address these issues, we propose a novel blockchain sharding scheme based on an evolutionary deep reinforcement learning algorithm, which integrates Soft Actor-Critic (SAC) algorithm with genetic algorithm, enabling better policy diversity. First, we develop a dynamic hyperparameter adjustment mechanism that adjusts the number of blocks, the size of the blocks and the interval between blocks according to environmental conditions. Second, we introduce a comprehensive evaluation mechanism encompassing throughput, latency, decentralization, and security metrics to fully evaluate blockchain performance. Third, we formulate the hyperparameter configuration as an optimization problem designed to maximize the system performance and employ a SAC algorithm to solve this problem. Finally, given the number of shards, we consider the trust level and available resources for shard partitioning by employing a genetic algorithm. Simulation results indicate that our algorithm effectively balances security and efficiency trade-offs, yielding significant improvements in overall blockchain system performance for URT applications.