Omni-domain energy-efficient decision-making for large-scale heterogeneous platoons with dual-level graph reinforcement learning.

Abstract

Large-scale heterogeneous platoons markedly improve transport efficiency and lower operating costs. However, their deployment is constrained by the super-linear rise in system complexity, the difficulty of smooth cooperative decision-making, and the absence of omni-domain energy optimisation. To address these issues, we propose a Dual-level Graph Reinforcement Learning (DGRL) framework that decomposes interactions into inter-platoon and intra-platoon levels, thereby curbing computational overhead. A multi-head graph-attention mechanism captures non-linear spatiotemporal dependencies. Moreover, we construct, for the first time, an Omni-domain energy-consumption evaluation pipeline encompassing vehicle-side, road-side, and cloud-side components, thus overcoming the limited scalability and sub-optimal global performance of existing approaches. Experiments show that, while ensuring safety, DGRL increases traffic throughput by 6.9% and substantially reduces computational road. Net energy consumption per platoon decreases by 7%, demonstrating that vehicle-side savings fully offset the modest increases in road-side and cloud-side energy use. These findings lay a solid foundation for the practical deployment of large-scale heterogeneous platoons.