Reputation-Based Regional Federated Learning for Knowledge Trading in Blockchain-Enhanced IoV

The Internet of Vehicles (IoV) aims to perceive, compute, and process environmental data in a collaborative manner. Previous works focus on data sharing between vehicles, but a large amount of data will lead to redundant transmission and network congestion. In addition, security and privacy issues prevent these nodes from participating in the sharing process. Knowledge is extracted from data through machine learning (ML) and shared in the form of small-scale well-trained model parameters, which improves collaborative learning more effectively and relieves network pressure. While traditional ML algorithms are not suitable for distributed IoV with local characteristics. Based on this, this paper first divides the vehicles into multiple regions and proposes a Regional Federated Learning (RFL) framework, in which all regions maintain their own learning models, i.e. knowledge. We design a reputation mechanism to measure the reliability of vehicles participating in RFL. To address the security challenges brought by the untrusted centralized trading market, we propose a blockchain-enhanced knowledge trading framework, in which an authorized market agency coordinates the trading quickly. We model the optimal pricing mechanism as a non-cooperative game, taking into account the competition among all knowledge providers. Numerical simulation shows that the proposed reputation mechanism improves the accuracy of knowledge up to 18%, and the optimal knowledge pricing mechanism effectively increases the utility of market.