PoLVR: A Blockchain Consensus Mechanism Based on Deep Learning and Dynamic Reputation Management

Abstract

Consensus algorithms are essential for blockchain networks to achieve agreement on transaction outcomes. However, mainstream algorithms like Proof of Work (PoW) and Proof of Stake (PoS) exhibit significant limitations in security and efficiency, including high energy consumption, wealth centralization, and a lack of effective node behavior evaluation to guard against internal attacks. To address these issues, this paper proposes an intelligent reputation-based consensus mechanism leveraging a Long Short-Term Memory (LSTM) network. This mechanism analyzes multi-dimensional node attributes (e.g., hostname, country, event sequence, timestamp) to model behavioral patterns using the LSTM, enabling accurate reputation quantification and early detection of malicious intent. Furthermore, we design a dynamic reputation scoring system that calculates a composite reputation score by weighting the LSTM’s predicted score against the node’s historical behavior score. This composite score is directly applied to the dynamic election of authoritative nodes and their role assignment within the consensus process. Simulation results demonstrate that, compared to traditional PoW and PoS mechanisms, our approach significantly reduces the attack success rate of malicious nodes attempting to form monopolies, thereby enhancing the fairness of the consensus process and the overall robustness of the system.