Work

Innovating island energy management under uncertainty: a risk-aware closed-loop multi-agent reinforcement learning framework for multi-dimensional resilience

Year: 2026
Type: article
Abstract: The increasing integration of renewable energy into island power systems introduces substantial uncertainty in generation, posing challenges to both the reliability and economic performance of energy management. To address these challenges, this study proposes a fuzzy Conditional Value-at-Risk (CVaR)-based multi-agent reinforcement learning (MARL) framework designed to enable risk-aware operational decision-making in island-based electricity-hydrogen coupled systems (ECSs). Within this framework, uncertainties in wind and photovoltaic power forecasts are characterized using trapezoidal fuzzy numbers combined with Gaussian copula-based correlations. A multiple attribute decision-making–CVaR method that incorporates hybrid weighting and Choquet integral-based aggregation is employed to assess multidimensional risks. A key innovation of the proposed approach is the integration of a CVaR-guided attention mechanism into an enhanced MARL algorithm, which directs learning focus toward high-risk scenarios during training. A case study based on a real-world island-based ECS validates the effectiveness of the proposed framework, demonstrating faster convergence and greater policy robustness compared with conventional MARL and robust optimization methods. These results indicate the framework’s ability to support adaptive, interpretable, and resilient energy scheduling under uncertainty, thereby contributing new methodological insights to intelligent management of integrated energy systems, and a replicable model for innovation in sustainable energy operations.
Language: English
Open Access status: gold
Estimated APC paid: $2,200
OpenAlex ID: W7204200180
FWCI:
Cites: 26
Cited by:
Related to: 10