Energy and Latency Tradeoff for STAR-IRS-Assisted Vehicle Road Cooperative System in Carbon Intelligent IIoT Leveraging Quantized Federated Reinforcement Learning

The rapid expansion of the Industrial Internet of Things (IIoT) in vehicular networks has significantly increased the demand for energy-efficient and low-latency communication to support intelligent transportation systems. However, the associated carbon footprint poses major challenges to sustainable development. To resolve these problems, we propose an Energy-Efficient and Latency-Minimizing Simultaneously Transmitting and Reflecting-Intelligent Reflecting Surface (STAR-IRS)-Assisted Vehicle-Road Cooperative System (VRCS) within a Carbon-Aware IIoT environment, leveraging Quantized Federated Reinforcement Learning (Q-FRL). The STAR-IRS dynamically enhances signal strength and energy efficiency by adjusting transmission and reflection coefficients, ensuring robust connectivity in complex vehicular environments. Q-FRL helps adjust STAR-IRS settings in real time while reducing computational complexity through quantized decision-making. Adaptive quantized DDPG improves energy efficiency, whereas adaptive quantized DDQN minimizes latency under carbon-aware constraints. Simulation results substantiate the performance of the proposed framework, demonstrating superior energy efficiency, lower latency, and reduced carbon emissions. Specifically, the adaptive quantized DDPG (AQ-DDPG) reduces energy consumption by 31.57%, while gradient quantized DDPG (GQ-DDPG) and fixed (4-bit) DDPG improve it by 16.84% and 6.31%, respectively, compared to fixed (2-bit) DDPG. Furthermore, AQ-DDPG reduces vehicle carbon emissions by 15% and 10% compared to FQ-DDPG and GQ-DDPG, respectively.