BLOCKCHAIN-ENABLED PRIVACY-PRESERVING FEDERATED LEARNING WITH DEEP CAPSULE NETWORKS FOR ENERGY-EFFICIENT SPEED CONTROL IN AUTONOMOUS VEHICLES

Rough roads negatively affect the ride quality and energy efficiency of autonomous vehicles (AVs) and demand adaptive speed control. To solve this, this research proposes a Deep Capsule Network integrated with Sparse Kernel K-means Clustering (DCapsNet + SK-K-means) to accurately reconstruct pavement conditions and adapt vehicle speed in real time. The Privacy-Preserving Federated Learning (PPFL) with Hyperledger Fabric blockchain allows federated learning to be trained in a secure and decentralized manner without the transfer of raw sensor data. With the BDD100K dataset, supervised and reinforcement learning are used. According to the obtained results of the experiment, there is a 10.45% increase in ride comfort, 28.23% in energy efficiency, and 98.28% in computational efficiency. The framework attains 93.73% security, 96.46% throughput and 0.83 s operation time at 250 vehicles/km.