An energy trajectory-based reinforcement learning framework with meta-adaptation for robust energy management in solar-powered electricity–heat–hydrogen integrated energy system

A robust energy management system (EMS) is critical for integrated energy systems (IESs), particularly those with high renewable energy penetration. The inherent complexities resulting from multi-energy conversion, transmission processes, and numerous uncertainties significantly challenge the economic and reliable operation of IESs. This study proposes a meta-learning-enhanced reinforcement learning framework (MPPO) for optimal scheduling in IESs. Each operational scenario is explicitly formulated as an uncertainty-aware learning environment, allowing the agent to adapt its policy to scenario-specific uncertainty patterns. A trajectory-based reward function (TBRF) is further designed to capture device-level energy flow behaviors, improving coordinated electric–heat–hydrogen dispatch under diverse operating conditions. Meta-learning provides a transferable initialization that accelerates convergence and enhances robustness across scenarios. Case studies show that the proposed MPPO considerably reduces training epochs by approximately 40%, while simultaneously improving policy stability across various scenarios. Annual simulation shows that MPPO decreases peak-period grid reliance from 112.88 MWh to 7.56 MWh, reduces total battery degradation from 15.56% to 6.28% and eliminates the large heat-shortage events. The results demonstrate that MPPO offers a robust and uncertainty-resilient solution for long-term IES optimal scheduling.