Scholarlink

<p>As renewable penetration continues to increase, virtual power plants (VPPs) are required to evolve from energy aggregators into high-confidence capacity providers capable of delivering firm commitments under uncertainty. The stochastic and asymmetric characteristics of wind and photovoltaic (PV) generation challenge conventional deterministic margins and chance-constrained formulations, which often lead either to excessive conservatism or elevated overcommitment risk. To address this limitation, this paper develops an expectile-based modeling framework for credible capacity evaluation of aggregated wind-PV-storage VPPs. The proposed approach explicitly captures the tail behavior of forecast errors and embeds statistical confidence directly into operational decision-making. In contrast to value-at-risk (VaR) and conditional value-at-risk (CVaR) formulations, expectile modeling introduces a continuous and convex risk measure that asymmetrically penalizes downside deviations, thereby enabling smoother and more interpretable risk calibration. The framework consists of two integrated components. First, a stochastic forecasting and uncertainty quantification module combines quantile regression and asymmetric expectile regression to generate site-specific predictive distributions. Second, a risk-adjusted optimization layer co-schedules renewable commitments and battery dispatch subject to expectile-consistent constraints. Through epigraph reformulations and asset-level derating mechanisms, the model preserves convexity and computational scalability without requiring restrictive distributional assumptions. Case studies conducted on a regional VPP portfolio demonstrate that the proposed method achieves tighter and more economically efficient commitment bounds than CVaR-based benchmarks. Specifically, commitment shortfall events are reduced by up to 73% relative to baseline quantile approaches, while maintaining up to 95% of achievable revenue across varying market conditions. The results further reveal that battery dispatch decisions naturally concentrate during periods of heightened statistical exposure, confirming the value of integrating tail-sensitive forecasting with dynamic storage coordination. Overall, the study establishes a unified, risk-calibrated, and practically implementable framework for credible capacity modeling in renewable-dominated power systems, supporting reliable participation of VPPs in capacity and reserve markets.</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

An expectile-based framework for risk-calibrated credible capacity evaluation of virtual power plants under wind and PV forecast uncertainties

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月31日

查看全文
  • Dong Hua,
  • Jihan Zeng,
  • Qinglin Lin,
  • Shunju Cui,
  • Yiqing Liu,
  • Yuzhan Liang

摘要

As renewable penetration continues to increase, virtual power plants (VPPs) are required to evolve from energy aggregators into high-confidence capacity providers capable of delivering firm commitments under uncertainty. The stochastic and asymmetric characteristics of wind and photovoltaic (PV) generation challenge conventional deterministic margins and chance-constrained formulations, which often lead either to excessive conservatism or elevated overcommitment risk. To address this limitation, this paper develops an expectile-based modeling framework for credible capacity evaluation of aggregated wind-PV-storage VPPs. The proposed approach explicitly captures the tail behavior of forecast errors and embeds statistical confidence directly into operational decision-making. In contrast to value-at-risk (VaR) and conditional value-at-risk (CVaR) formulations, expectile modeling introduces a continuous and convex risk measure that asymmetrically penalizes downside deviations, thereby enabling smoother and more interpretable risk calibration. The framework consists of two integrated components. First, a stochastic forecasting and uncertainty quantification module combines quantile regression and asymmetric expectile regression to generate site-specific predictive distributions. Second, a risk-adjusted optimization layer co-schedules renewable commitments and battery dispatch subject to expectile-consistent constraints. Through epigraph reformulations and asset-level derating mechanisms, the model preserves convexity and computational scalability without requiring restrictive distributional assumptions. Case studies conducted on a regional VPP portfolio demonstrate that the proposed method achieves tighter and more economically efficient commitment bounds than CVaR-based benchmarks. Specifically, commitment shortfall events are reduced by up to 73% relative to baseline quantile approaches, while maintaining up to 95% of achievable revenue across varying market conditions. The results further reveal that battery dispatch decisions naturally concentrate during periods of heightened statistical exposure, confirming the value of integrating tail-sensitive forecasting with dynamic storage coordination. Overall, the study establishes a unified, risk-calibrated, and practically implementable framework for credible capacity modeling in renewable-dominated power systems, supporting reliable participation of VPPs in capacity and reserve markets.