Hierarchical Blockchain-enabled Federated Learning with Reputation Management for Mobile Internet of Vehicles

Federated learning (FL), as a distributed technology, has great application potential in autonomous driving. However, due to the lack of a quality verification mechanism for the client, it faces the problem of malicious users attacking the global model, which reduces the accuracy of FL model. In response to this problem, this paper proposes a hierarchical blockchainenabled FL with reputation management scheme, which improves the efficiency and accuracy of FL while ensuring privacy and security. In this paper, we first propose a hierarchical blockchain framework and design a blockchain consensus algorithm based on parameter proof of quality (PoQ), which can provide FL workers with a secure and efficient data storage environment in a decentralized manner, while making up for the poor scalability of traditional single blockchains. On this basis, we design a reputation management method based on Bayesian theory and multiple subjective logic models to select high-quality local participating users. In particular, the weights of factors such as interactive activity and interactive position are taken into account to improve the accuracy of reputation calculations. Extensive simulations validate the performance of our scheme in improving FL accuracy and efficiency.