A reinforcement learning-based reliability assessment and optimization framework
Physica Scripta · 19.06.2026 · 101 (24) · 246013
DOI: 10.1088/1402-4896/ae7ac4 ↗Kurz gefasst
Fachartikel in der Zeitschrift Physica Scripta, online erschienen am 17.06.2026. Verfasst von Adil Waheed und 3 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben.
Abstract (Originalsprache)
Optimal power flow (OPF) in renewable-rich power systems has become increasingly challenging due to stochastic load variations, renewable intermittency, network constraints, and the need for real-time reliable operation. Conventional OPF solvers and existing machine learning approaches often suffer from high computational complexity, limited adaptability under uncertainty, and difficulty in maintaining operational feasibility during dynamic system conditions. To address these limitations, this study proposes a reliability-aware deep reinforcement learning framework for solving the DC OPF problem under renewable uncertainty. The proposed methodology formulates OPF as a sequential decision-making problem, where a reinforcement learning agent learns generator redispatch, renewable curtailment, energy storage coordination, and topology switching actions. The learning process is further shaped by a reliability-aware reward function that directly incorporates reliability indices like expected energy not served (EENS), loss of load probability, and physics-based operational penalties. This reward design forces the agent to prioritize long-term grid adequacy over short-term cost minimization. The experimental setup is based on IEEE 30-, 118-, and 300-bus systems to validate the framework under stochastic operating conditions, Monte Carlo-based uncertainty analysis, and extreme contingency scenarios. Simulation results demonstrate feasibility rates of 98.37%, 93.11%, and 81.63% for IEEE 30-, 118-, and 300-bus systems, respectively, while maintaining low reliability indices and reduced operational violations under uncertainty. In the larger IEEE 300-bus case, the framework further produces a 15% EENS deviation and a constraint violation penalty of 168.3, together with an approximately 26.9 × computational speedup over conventional Monte Carlo simulation-based OPF evaluation. These results confirm that the proposed framework provides reliability-aware decision-making for large-scale renewable-integrated power systems.
Zitieren
APA
Waheed, A., Li, J., Zhao, K., & Ullah, S. (2026). A reinforcement learning-based reliability assessment and optimization framework. Physica Scripta, 101(24), 246013. https://doi.org/10.1088/1402-4896/ae7ac4
Vancouver
Waheed A, Li J, Zhao K, Ullah S. A reinforcement learning-based reliability assessment and optimization framework. Physica Scripta. 2026;101(24):246013. doi:10.1088/1402-4896/ae7ac4
BibTeX
@article{Waheed2026,
author = {Waheed, Adil and Li, Jueyou and Zhao, Kequan and Ullah, Saeed},
title = {A reinforcement learning-based reliability assessment and optimization framework},
journal = {Physica Scripta},
year = {2026},
volume = {101},
number = {24},
pages = {246013},
doi = {10.1088/1402-4896/ae7ac4}
}
EUVIMED