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

• Proposes a residential energy hub (REH) integrating electricity, gas, photovoltaics, wind turbines, CHP, and battery storage for multi-carrier energy management. • Develops a surrogate gradient clipping (SGC) learning framework using deep neural networks to obtain an optimal energy scheduling policy. • Implements a two-stage strategy combining reinforcement learning-based policy derivation with dynamic control of energy units. • Simulation results show improved operator profit and reliable, efficient energy management under diverse operating scenarios.