An integrated security approach for vehicular networks in smart cities

Fog computing, which is an extension of cloud computing is one of the cornerstone for Internet of Things, that witnessed rapid growth because of its ability to enhance several difficult problems such as network congestion, latency, and the lack of regional autonomy. However, privacy concerns and the resulting inefficiency are causing the performance of fog computing to suffer. While suffering from poisoning attacks, the vast majority of current works do not take into consideration of proper balance between them. Specifically, we present blockchain-enabled secure federated learning in vehicle network (BSFLVN) system model for traffic flow prediction in urban computing to overcome the aforementioned difficulties and narrow the gap. When vehicle devices trade local learning updates, BSFLVN enables them to be exchanged with a blockchain-based global learning model, which is confirmed by miners. To improve accuracy of classification we proposed federated learning-based gated recurrent unit for local model update using FL-GRNN algorithm and global model update using FL-AGG algorithm. The BSFLVN, which is built on top of this, allows autonomous deep learning-based GRU to take place without the need for a centralized authority to maintain the global model and coordinate by using the PoW consensus mechanism of blockchain. For preserving privacy and security of local and global model updates we employ the LDP mechanism. For the analysis of the latency performance of BSFLVN, as well as the derivation of the best block production rate we consider communication, consensus delays, and computing cost. The results of a thorough examination demonstrate that BSFLVN outperforms its competitors in terms of privacy protection, efficiency, and resistance to poisoning attacks, among other areas. Various deadline time iteration process is simulated for result evaluation using Fashion MINIST data in which over 93% of accuracy is obtained.