DQN-Raft+: A Deep Reinforcement Learning-Optimized Lightweight Consensus Algorithm for Secure Edge Storage in IoT Environments

Abstract

The rapid development of the Internet of Things (IoT) has intensified security and privacy challenges across data generation, transmission, and storage. This study introduces a blockchain-based secure edge storage model tailored for IoT environments and presents a lightweight consensus algorithm, Deep Q-Network (DQN)-Raft+, which incorporates deep reinforcement learning. By combining the decentralized features of edge computing and blockchain, the model enables automated data access control through smart contracts. Furthermore, it optimizes leader node selection in the Raft consensus process using a DQN, formulating the consensus as a Markov Decision Process to enhance responsiveness and privacy protection in dynamic network conditions. Experiments were performed in a simulated environment using TensorFlow 2.6 and a MySQL database. The performance of DQN-Raft+ was compared against traditional consensus algorithms, including Proof of Work, Proof of Stake, Practical Byzantine Fault Tolerance, and Delegated Byzantine Fault Tolerance. Results indicate that DQN-Raft+ significantly reduces block generation delay (175.77 ms) and achieves a high privacy protection score (0.95). It also maintains a low data loss rate of 0.01%, demonstrating enhanced robustness and real-time capability. These findings indicate that DQN-Raft+ effectively strengthens data security and privacy in IoT systems, offering a technically sound and efficient mechanism for secure data exchange. The study provides both a theoretical framework and practical direction for future research in secure IoT deployment.