Scholarlink

<p>Virtual power plants face dual challenges in coordinating internal distributed resources and engaging in strategic external trading under market uncertainties. This study proposes a two-stage game-theoretic decision framework to address these challenges. In the first stage, distributionally robust optimization aggregates photovoltaic systems, energy storage, and flexible loads to construct an internal cost function that incorporates operational risks. In the second stage, this cost function is embedded within a Stackelberg game to model the VPP’s strategic interaction with the external market, where a deep reinforcement learning algorithm adaptively approximates the equilibrium bidding strategy. Simulation results based on real market data from a Chinese provincial spot market and an actual VPP demonstration project demonstrate the effectiveness of the proposed approach. Over a 28-day test period spanning diverse seasonal conditions, the method increases average daily net profit by 18.7% compared to single-stage optimization and by 12.3% compared to deterministic game models, while improving the 95% Value-at-Risk by 22.5%. The findings confirm that integrating distributionally robust optimization for risk-aware internal aggregation with deep reinforcement learning for adaptive external gaming effectively supports VPPs in achieving risk-aware profit maximization within complex market environments.</p>

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

Research on the demand response trading strategy of virtual power plants under the two-stage game model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月27日

查看全文
  • Qi Yu,
  • Lei Nie,
  • Jinxuan Li,
  • Jianjun Lai,
  • Yushi Ling,
  • Ruibin Liang

摘要

Virtual power plants face dual challenges in coordinating internal distributed resources and engaging in strategic external trading under market uncertainties. This study proposes a two-stage game-theoretic decision framework to address these challenges. In the first stage, distributionally robust optimization aggregates photovoltaic systems, energy storage, and flexible loads to construct an internal cost function that incorporates operational risks. In the second stage, this cost function is embedded within a Stackelberg game to model the VPP’s strategic interaction with the external market, where a deep reinforcement learning algorithm adaptively approximates the equilibrium bidding strategy. Simulation results based on real market data from a Chinese provincial spot market and an actual VPP demonstration project demonstrate the effectiveness of the proposed approach. Over a 28-day test period spanning diverse seasonal conditions, the method increases average daily net profit by 18.7% compared to single-stage optimization and by 12.3% compared to deterministic game models, while improving the 95% Value-at-Risk by 22.5%. The findings confirm that integrating distributionally robust optimization for risk-aware internal aggregation with deep reinforcement learning for adaptive external gaming effectively supports VPPs in achieving risk-aware profit maximization within complex market environments.