PDF 摘要：中文翻译尚未生成，请点击“显示原文依据”查看英文内容。

The large number of Electric Vehicles (EVs) now connected to the Internet-of-Things (IoT) network has created numerous obstacles in securing data sharing, managing decentralised authentication, safeguarding privacy and ensuring trust management. However, traditional centralized solutions can be compromised by data manipula - tion, spoofing, unauthorized user access, and the systems can become vulnerable to single points of failure, potentially reducing the reliability and scalability of intelligent transportation systems. The challenges men - tioned above motivate this paper to present a Blockchain-Assisted Federated Graph Reinforcement Learning (BFGRL) framework for secure and scalable smart EV networks. Within the proposed framework, PUF-ECC is used to create lightweight and tamper-resistant authentication of vehicles, GATs for complex interaction modelling, Federated Learning (FL) for privacy-preserving distributed trust assessments and Proximal Policy Optimization (PPO) for optimizing vehicle validation and blockchain consensus. It includes three parts: decen - tralized identity authentication, secure data sharing with intelligent trust and privacy mechanisms, and efficient transaction processing with blockchain consensus optimization. The proposed approach presented in this docu - ment was benchmarked against the VeReMi vehicular security dataset as well as various simulated scenarios of a blockchain-based EV network, while also comparing against the ones already existing in the literature: block - chain-based authentication, federated learning and trust management methods. The experimental results showed that all of the existing baseline approaches were outperformed in terms of authentication accuracy (99.2%), attack detection rate (98.7%), data integrity score (99.5%), privacy preservation (98.4%), scalability (97.2%) and transaction throughput (5200 TPS), with a statistical difference ( p < 0.05). The proposed framework also resulted in an average computation overhead of 41 ms for authentication and consensus operations, making it feasible for real-time applications. The results indicate that BFGRL is a suitable solution that is privacyprotected, efficient and powerful for the intelligent transportation system and next-generation smart electric vehicle system. Experiments conducted in a controlled simulation environment show that BFGRL can effec - tively improve security, trust management, scalability, and operational efficienssssscy in blockchain-assisted smart EV networks. These results suggest its promising potential for intelligent transportation applications, with further testing with real-world vehicular communication data and field trials being critical research areas. Article in Press Scientific Reports https://doi.org/10.1038/s41598-026-71684-y

全文第 3 页依据：中文翻译尚未生成，请点击“显示原文依据”查看英文内容。

To address these challenges, this study presents a secure and trustworthy Blockchain-Assisted Federated Graph Reinforcement Learning (BFGRL) approach for data sharing and authentication in smart electric vehi - cle (eVehicle) networks with IoT support.

全文第 1 页依据：中文翻译尚未生成，请点击“显示原文依据”查看英文内容。

Within the proposed framework, PUF-ECC is used to create lightweight and tamper-resistant authentication of vehicles, GATs for complex interaction modelling, Federated Learning (FL) for privacy-preserving distributed trust assessments and Proximal Policy Optimization (PPO) for optimizing vehicle validation and blockchain consensus.

全文第 1 页依据：中文翻译尚未生成，请点击“显示原文依据”查看英文内容。

Within the proposed framework, PUF-ECC is used to create lightweight and tamper-resistant authentication of vehicles, GATs for complex interaction modelling, Federated Learning (FL) for privacy-preserving distributed trust assessments and Proximal Policy Optimization (PPO) for optimizing vehicle validation and blockchain consensus.

全文第 1 页依据：中文翻译尚未生成，请点击“显示原文依据”查看英文内容。

The experimental results showed that all of the existing baseline approaches were outperformed in terms of authentication accuracy (99.2%), attack detection rate (98.7%), data integrity score (99.5%), privacy preservation (98.4%), scalability (97.2%) and transaction throughput (5200 TPS), with a statistical difference ( p < 0.05).