发电技术 ›› 2025, Vol. 46 ›› Issue (3): 496-507.DOI: 10.12096/j.2096-4528.pgt.24074

• AI在新型电力系统中的应用 • 上一篇    

基于人工智能的直流微电网大信号稳定性评估方法研究

刘宿城1,2, 栾李1,2, 李龙1,2, 洪涛1,2, 刘晓东1,2   

  1. 1.安徽工业大学电气与信息工程学院,安徽省 马鞍山市 243000
    2.电力电子与运动控制安徽省重点实验室(安徽工业大学),安徽省 马鞍山市 243000
  • 收稿日期:2024-12-22 修回日期:2025-02-05 出版日期:2025-06-30 发布日期:2025-06-16
  • 作者简介:刘宿城(1981),男,博士,教授,研究方向为直流微电网与电力电子系统建模、稳定性分析与控制,liusucheng@126.com
    栾李(1998),男,硕士研究生,研究方向为直流微电网建模与大信号稳定性分析, luanli510@126.com
    李龙(1999),男,硕士研究生,研究方向为直流微电网大信号稳定性分析,lilonghaha2020@163.com
    洪涛(1999),男,硕士研究生,研究方向为直流微电网建模与小信号稳定性分析,hongtao@ahut.edu.cn
    刘晓东(1971),男,博士,教授,研究方向为DC-DC开关变换器和电机设计,liuxiaodong@ahut.edu.cn
  • 基金资助:
    国家自然科学基金面上项目(52277169)

Research on Large-Signal Stability Assessment Methods of DC Microgrids Based on Artificial Intelligence

Sucheng LIU1,2, Li LUAN1,2, Long LI1,2, Tao HONG1,2, Xiaodong LIU1,2   

  1. 1.School of Electrical and Information Engineering, Anhui University of Technology, Maanshan 243000, Anhui Province, China
    2.Key Lab of Power Electronics and Motion Control (Anhui University of Technology), Maanshan 243000, Anhui Province, China
  • Received:2024-12-22 Revised:2025-02-05 Published:2025-06-30 Online:2025-06-16
  • Supported by:
    National Natural Science Foundation of China(52277169)

摘要:

目的 直流微电网因低惯性和恒功率负载特性易引发大信号稳定性问题,传统基于模型的方法计算复杂、求解困难。针对上述问题,探讨了直流微电网的大信号稳定性智能化分析方法。 方法 选取常见的人工智能技术分类器对直流微电网的稳定性进行分析,并比较分析深度学习、支持向量机和决策树3类(涵盖6种)常见的人工智能技术在具体直流微电网实例中的大信号稳定性评估的应用。 结果 基于具体实例的对比分析表明,在直流微电网大信号稳定性评估中,长短期记忆网络在综合性能(准确性、实时性、适应性)上优于其他方法。 结论 长短期记忆网络分类器与直流微电网的状态空间方程具有高匹配性,比传统的机器学习分类器更适用于直流微电网大信号稳定性的分析。同时,确保分类器的工作性能需选择适当的参数值。

关键词: 直流微电网, 分布式能源, 可再生能源, 储能, 人工智能(AI), 机器学习, 深度学习, 长短期记忆网络

Abstract:

Objectives DC microgrids are prone to the issue of large-signal stability due to the low inertia and constant power load characteristics. Traditional model-based methods involve complex calculations and are difficult to solve. To address these issues, this study investigates intelligent analysis methods for large-signal stability of DC microgrids. Methods Common artificial intelligence (AI) classifiers are selected to analyze the stability of DC microgrids. A comparative analysis is conducted on three types of common AI technologies (covering six methods)-deep learning, support vector machine, and decision trees-for large-signal stability assessment in a specific DC microgrid case study. Results Comparative analysis based on specific examples shows that in the large-signal stability assessment of DC microgrids, long short-term memory (LSTM) networks outperform other methods in terms of overall performance (accuracy, real-time capability, and adaptability). Conclusions The LSTM network classifier shows high compatibility with the state-space equations of DC microgrids, making it more suitable than traditional machine learning classifiers for large-signal stability analysis of DC microgrids. Meanwhile, ensuring the performance of the classifier requires appropriate selection of parameter values.

Key words: DC microgrids, distributed energy resources, renewable energy, energy storage, artificial intelligence(AI), machine learning, deep learning, long short-term memory network

中图分类号: