Efficient low-carbon construction pathways for energy ecosystems in sustainable cities through deep reinforcement learning management with green hydrogen diversified utilization under trust safe peer-to-peer trading and social welfare

Achieving sustainable and energy-efficient cities requires management frameworks that move beyond isolated optimization of single energy carriers and address the operational complexity of integrated energy ecosystems. This paper proposes a knowledge-network-assisted deep reinforcement learning (DRL) framework for coordinated management of multi-energy clusters, including photovoltaic generation, microturbines, gas boilers, heat pumps, electrolyzers, hydrogen-fueled combined heat and power units, and coupled hydrogen and thermal storage systems. The core methodology formulates system operation as a constrained Markov decision process, in which a genetic algorithm-derived knowledge network is embedded as a prior within the DRL agent to stabilize learning and substantially accelerate convergence under uncertainty. Diversified green hydrogen utilization is incorporated to enhance operational flexibility and resilience, while trust-safe peer-to-peer energy trading mechanisms enable transparent and secure local exchanges. In parallel, social welfare indicators are employed to align technical optimization with affordability, equity of access, and user comfort. Simulation studies based on a six-month dataset with hourly resolution demonstrate that the proposed framework reduces operating costs by up to 27% and achieves nearly 58% faster convergence compared with baseline DRL approaches, while maintaining indoor thermal comfort within acceptable bounds for more than 95% of the scheduling horizon. The results confirm that embedding structured knowledge into DRL, together with hydrogen diversification and welfare-aware management, provides an effective pathway toward low-carbon construction and sustainable urban energy ecosystems.