Scholarlink

To cope with the complex dynamic characteristics brought about by the sharp increase in penetration of various power electronic devices, online dynamic security analysis in modern power systems has put forward higher requirements for the accuracy of simulation models. Traditional methods for load model parameter identification (PI) struggle with model complexity, computational pressure, and low generalization, which can hinder their effectiveness in online applications. To overcome these challenges, this paper proposes a novel online load model PI method utilizing the Deep Deterministic Policy Gradient (DDPG) algorithm. Initially, a composite load model incorporating various distributed energy resources (DERs) is developed as the model basis. Subsequently, a DDPG-based framework for online load model PI is designed, featuring meticulously crafted interface functions to improve both convergence and generalization capabilities. Case studies on the EPRI-36 buses test system demonstrate the proposed method's efficiency and accuracy compared to traditional heuristic optimization methods.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Online Load Model Parameter Identification with Deep Reinforcement Learning: A DDPG-Based Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zihao Li,
  • Xingwei Liu,
  • Yupeng Huang,
  • Jianxiong Hu,
  • Yi Tang

摘要

To cope with the complex dynamic characteristics brought about by the sharp increase in penetration of various power electronic devices, online dynamic security analysis in modern power systems has put forward higher requirements for the accuracy of simulation models. Traditional methods for load model parameter identification (PI) struggle with model complexity, computational pressure, and low generalization, which can hinder their effectiveness in online applications. To overcome these challenges, this paper proposes a novel online load model PI method utilizing the Deep Deterministic Policy Gradient (DDPG) algorithm. Initially, a composite load model incorporating various distributed energy resources (DERs) is developed as the model basis. Subsequently, a DDPG-based framework for online load model PI is designed, featuring meticulously crafted interface functions to improve both convergence and generalization capabilities. Case studies on the EPRI-36 buses test system demonstrate the proposed method's efficiency and accuracy compared to traditional heuristic optimization methods.