Multi-layer surrogate gradient clipping framework for hierarchical scheduling of multi-carrier residential hubs

The rapid increase in residential energy demand for heating, cooling, and electricity necessitates the deployment of multi-carrier energy systems (MCES) to achieve efficient energy production from both economic and environmental perspectives. In this context, combined heat and power (CHP) units, renewable energy sources, battery storage, and demand response (DR) programs play a crucial role. This paper presents a residential energy hub (REH) integrating multiple energy carriers—including electricity, gas, photovoltaics, wind turbines, and battery storage units—and develops an advanced scheduling framework based on surrogate gradient clipping (SGC) learning for the optimal operation of this complex system. The proposed energy management strategy operates in two stages: (i) the SGC framework employs deep neural networks (DNNs) as function approximators to derive the optimal policy through agent-environment interactions while incorporating gradient clipping for stable learning, and (ii) the derived policy is applied to dynamically adjust energy units, maximizing a reward tailored to system objectives. Simulation results under diverse scenarios demonstrate that the proposed approach effectively enhances the operator’s profit while ensuring smooth and reliable energy management.