Abstract: Eclipse attacks, as a significant threat to the blockchain network layer, can isolate the targeted node from the entire network by controlling its network connections, thus affecting its ability to receive block and transaction information. On this basis, attackers can launch double-spending and other attacks, causing substantial damage to the blockchain system. To address this issue, a dynamic detection method for Eclipse attacks at the blockchain network layer was proposed, incorporating deep learning models. The Node Comprehensive Resilience Index (NCRI) was utilized to represent the multidimensional attribute features of nodes, and Graph Attention Networks (GAT) were introduced to dynamically update the node features of the network topology. Convolutional Neural Networks (CNN) were employed to fuse the multidimensional features of nodes, and a Multilayer Perceptron (MLP) was used to predict the overall vulnerability of the network. Experimental results indicated that an accuracy of up to 89.80% was achieved under varying intensities of Eclipse attacks while maintaining stable performance in continuously changing blockchain networks.

Key words: Keywords: blockchain network layer, deep learning, eclipse attack detection, Graph Attention Networks, Multi-Layer Perceptron

摘要： Eclipse攻击作为针对区块链网络层的一种显著威胁，通过控制节点的网络连接，可导致被攻击节点与整个网络隔离，影响其接收区块和交易信息的能力。攻击者可以在此基础上发起双重支付等攻击，对区块链系统造成巨大破坏。针对该问题，结合深度学习模型，提出了一种面向节点分析的区块链网络层Eclipse攻击动态检测方法。该方法利用节点综合韧性指标（NCRI）表达节点的多维属性特征，并引入图注意力神经网络（GAT）动态更新网络拓扑结构的节点特征，使用卷积神经网络（CNN）融合节点的多维特征，最终结合多层感知机（MLP）预测整体网络的脆弱性。实验结果表明，该方法在不同的Eclipse攻击强度下准确率最高可以达到89.80%，同时能够在不断变化的区块链网络中保持稳定的性能。

关键词: 区块链网络层, 深度学习, Eclipse攻击检测, 图注意力神经网络, 多层感知机