A comprehensive review of deep reinforcement learning applications from centralized power generation to modern energy internet frameworks
<p>The energy internet (EI) is evolving toward decentralized, data-rich, and time-critical operation, where legacy optimization often fails to meet complexity, scalability, and real-time constraints. Deep reinforcement learning (DRL) offers a data-driven alternative that couples perception wit...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| مؤلفون آخرون: | , , , |
| منشور في: |
2025
|
| الموضوعات: | |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | <p>The energy internet (EI) is evolving toward decentralized, data-rich, and time-critical operation, where legacy optimization often fails to meet complexity, scalability, and real-time constraints. Deep reinforcement learning (DRL) offers a data-driven alternative that couples perception with sequential decision-making. This review synthesizes evidence from more than 500 peer-reviewed studies published between 2020 and 2026, mapping DRL applications across distributed generation, transmission, distribution, energy storage systems, energy markets, local energy management, grid security, and data privacy. We present a structured taxonomy covering value-based, policy-based, actor-critic, model-based, and advanced multi-agent and multi-objective approaches, and link algorithms to tasks such as dispatch, microgrid coordination, real-time pricing, load balancing, and demand–response. Reported benefits include energy savings, emissions reduction, and improved operational efficiency, although gains are often contingent on data quality, reward design, and benchmark selection. Persistent challenges include dataset scarcity and bias, limited generalization across assets and seasons, computational cost, safety and constraint handling, and the gap between simulation and field deployment. We identify promising directions: hybrid model-free and model-based DRL, offline-to-online learning, transfer and meta-learning for rapid adaptation, integration with federated learning and blockchain for privacy and trust, and progress in interpretability, uncertainty quantification, and formal safety guarantees. The review consolidates a fragmented literature, surfaces unresolved issues, and outlines practical research pathways for building reliable, sustainable, and resilient EI solutions with DRL.</p><h2>Other Information</h2> <p> Published in: Renewable and Sustainable Energy Reviews<br> License: <a href="http://creativecommons.org/licenses/by/4.0/" target="_blank">http://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1016/j.rser.2025.116481" target="_blank">https://dx.doi.org/10.1016/j.rser.2025.116481</a></p> |
|---|