Digital twin-driven multi-dimensional optimization of methanol-diesel hybrid power systems: emission control and fuel efficiency enhancement via HSAC algorithm

To address the dynamic coupling between traffic emissions and energy efficiency in methanol-diesel hybrid power systems (MDHPS), this study proposes a digital twin (DT)-driven multi-dimensional collaborative optimization framework. Key innovations include: 1) A spatiotemporal emission model integrating neighborhood dynamic and methanol permeation effects, with a dynamic weight adaptive update mechanism, achieving a prediction error <3.7% (34% lower than fixed-weight strategies) verified in Beijing, Shanghai, and Guangzhou road networks; 2) An improved deep reinforcement learning HSAC algorithm (embedded with graph attention mechanism) to enhance decision efficiency; 3) A verifiable differential privacy federated learning framework (privacy budget ɛ=1.5) ensuring data security and model accuracy. In SUMO-GT-Power co-simulations, compared with traditional model predictive control (MPC), the HSAC algorithm raises emission compliance rate to 99.7% (vs. 82.1%), reduces NOₓ emissions by 46.4% and computational latency by 55.8%, and improves fuel efficiency by 12.8% (all p<0.001). The federated learning framework maintains <0.3% accuracy loss when ɛ=1.0, while cutting data leakage risk by 62%. This work provides theoretical and technical support for low-carbon operation of urban transportation-energy coupling systems.