The rapid development of blockchain technology has led to a constant increase in its financial and technological value. However, this has also led to malicious attacks. Distributed denial-of-service attacks pose a considerable threat to blockchain technology out of many attacks due to its effectiveness and distributed nature. To protect the blockchain from DDoS attacks, researchers have proposed a large number of defensive schemes. However, these schemes are not well-suited for use in practical situations. In this work, we propose a DDoS attack detection scheme based on centralized federated learning, where multiple participating nodes locally train models and upload them to a central node for aggregation. Additionally, we propose a more suitable method for blockchain scenarios, using decentralized federated learning technology, where multiple nodes exchange models in a peer-to-peer manner to complete model training without a central server. We simulate DDoS attacks in blockchain and generate a large dataset by combining it with traditional network layer DDoS attack data to evaluate the effectiveness of our schemes. The experimental results show that the proposed schemes perform well in classification accuracy, demonstrating that our techniques can detect DDoS attacks effectively.

中文翻译：

---

用于区块链 DDoS 攻击检测的联合学习架构


区块链技术的快速发展导致其金融和技术价值不断增加。但是，这也导致了恶意攻击。由于其有效性和分布式特性，分布式拒绝服务攻击在许多攻击中对区块链技术构成了相当大的威胁。为了保护区块链免受 DDoS 攻击，研究人员提出了大量的防御方案。但是，这些方案并不适合在实际情况下使用。在这项工作中，我们提出了一种基于集中式联邦学习的 DDoS 攻击检测方案，其中多个参与节点在本地训练模型并将其上传到中心节点进行聚合。此外，我们提出了一种更适合区块链场景的方法，使用去中心化的联邦学习技术，其中多个节点以点对点的方式交换模型，在没有中央服务器的情况下完成模型训练。我们在区块链中模拟 DDoS 攻击，并通过将其与传统网络层 DDoS 攻击数据相结合来生成大型数据集，以评估我们计划的有效性。实验结果表明，所提出的方案在分类准确性方面表现良好，表明我们的技术可以有效地检测 DDoS 攻击。