发电技术 ›› 2025, Vol. 46 ›› Issue (3): 508-520.DOI: 10.12096/j.2096-4528.pgt.24227

• AI在新型电力系统中的应用 • 上一篇    

基于强化学习的新型电力系统优化策略应用综述

闫正义, 赵康, 王凯   

  1. 青岛大学电气工程学院,山东省 青岛市 266071
  • 收稿日期:2024-10-22 修回日期:2025-01-19 出版日期:2025-06-30 发布日期:2025-06-16
  • 通讯作者: 王凯
  • 作者简介:闫正义(1998),男,硕士研究生,研究方向为锂电池荷电状态估计,zzxx1998111@163.com
    赵康(2000),男,硕士研究生,研究方向为锂电池剩余寿命估计,mailboxofzk@163.com
    王凯(1985),男,博士,教授,研究方向为分布式微电网和储能技术、新能源储能技术、能源互联网等,本文通信作者,wangkai@qdu.edu.cn
  • 基金资助:
    国家自然科学基金项目(12374088);山东省高等学校青年创新技术项目(2022KJ139)

Review of Application on Optimization Strategies for New-Type Power System Based on Reinforcement Learning

Zhengyi YAN, Kang ZHAO, Kai WANG   

  1. College of Electrical Engineering, Qingdao University, Qingdao 266071, Shandong Province, China
  • Received:2024-10-22 Revised:2025-01-19 Published:2025-06-30 Online:2025-06-16
  • Contact: Kai WANG
  • Supported by:
    National Natural Science Foundation of China(12374088);Youth Innovation Technology Project of Higher School in Shandong Province(2022KJ139)

摘要:

目的 随着电力系统向更高程度的智能化和自动化演进,强化学习(reinforcement learning,RL)作为人工智能领域的一项关键技术,在电力领域的智能化发展方向上展现出广阔前景。完善RL在电力领域的应用研究方案,对于深入挖掘其在电力系统运行、控制和优化等方面的潜力至关重要。为此,分析了RL在实际电气应用中的效能表现,并展望了未来可能的研究方向,以期为电力系统智能化进程提供助力。 方法 对RL在各类电气领域的关键应用进行了综述。系统性地介绍了RL的基本原理和标志性算法,详细探讨这些算法如何被应用于新型电力系统领域的实际问题中。对各研究中主流的RL算法进行归类,并对在这些算法中进行的结构化改进进行优缺点分析。 结果 相比于传统算法,RL显著提升了新型电力系统的智能化水平,并在多个应用场景中取得了显著成效,特别是在应对系统复杂性和不确定性方面表现出色。然而,尽管有诸多成功案例,但目前该领域仍存在一些亟待解决的问题,比如计算成本高、训练时间长、泛化能力不足等。 结论 RL为新型电力系统的智能化提供了新的解决方案,然而,要实现大规模应用,还需要克服一系列技术和实践上的挑战。研究成果可为电气工程领域的研究者和实践者提供参考和启示。

关键词: 新型电力系统, 强化学习(RL), 深度强化学习(DRL), 智能电网, 优化策略, 能源管理, 态势感知, 优化调度, 人工智能(AI)

Abstract:

Objectives As power systems evolve toward higher levels of intelligence and automation, reinforcement learning (RL), a key technology in artificial intelligence, shows great potential in the intelligent development of the power sector. Enhancing research methods for RL applications is crucial for fully exploring its potential in power system operation, control, and optimization. Therefore, the performance of RL in practical electrical applications is analyzed, and the possible research directions in the future are prospected, so as to provide assistance for the intelligent transformation of power systems. Methods This study provides a systematic review of RL applications across diverse fields of electrical engineering. It systematically introduces the fundamental principles and landmark algorithms of RL, detailing how these algorithms are applied to address practical problems in new-type power system. The study categorizes mainstream RL algorithms in current research and analyzes the advantages and disadvantages of structural improvements made to these algorithms. Results Compared to traditional algorithms, RL significantly enhances the intelligence level of new-type power system. It achieves remarkable success in various application scenarios, particularly in addressing system complexity and uncertainty. However, despite many successful cases, several urgent issues still exist in this sector, such as high computational costs, long training times, and limited generalization abilities. Conclusions Reinforcement learning provides novel solutions for the intelligent development of new-type power system. However, achieving large-scale application still needs to overcome a series of technical and practical challenges. This study provides references and insights for researchers and practitioners in electrical engineering.

Key words: new-type power system, reinforcement learning (RL), deep reinforcement learning (DRL), intelligent grid, strategy optimization, energy management, situational awareness, optimized scheduling, artificial intelligence (AI)

中图分类号: