Dispatch-Embedded Long-Term Tail Risk Assessment and Mitigation via CVaR for Renewable Power Systems

Kai Kang, Feng Liu* · 2026

Read Dispatch-Embedded Long-Term Tail Risk Assessment and Mitigation via CVaR for Renewable Power Systems by Kai Kang, Feng Liu* (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-00926.txt.

Introduction

Kang and Liu propose a long-term framework for judging and easing tail risk in renewable power systems that embeds dispatch plans in full. A typical case method joins Copula models over many time scales to capture long-range swings and links. A change-based model then uses Conditional Value-at-Risk as a robust tail-risk metric. A guided change-based easing scheme then refines long-term dispatch plans, with tests on an altered IEEE-39 bus system using real-world data backing the method.

Catalog record

CREATORKai Kang, Feng Liu*
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26685
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.00926

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Dispatch-Embedded Long-Term Tail Risk Assessment and Mitigation via CVaR for Renewable Power Systems Thanks: This work was supported by the Joint Research Fund in Smart Grid (No.U1966601) under cooperative agreement between the National Natural Science Foundation of China (NSFC). Kai Kang, Feng Liu* Abstract Renewable energy (RE) generation exhibits pronounced seasonality and variability, and neglecting these features can lead to significant underestimation of long-term power system risks in power supply. While long-term dispatch strategies are essential for evaluating and mitigating tail risks, they are often excluded from existing models due to their complexity. This paper proposes a long-term tail risk assessment and mitigation framework for renewable power systems, explicitly embedding dispatch strategies. A representative scenario generation method is designed, combining multi-timescale Copula modeling to capture RE’s long-range variability and correlation. Building on these scenarios, an evolution-based risk assessment model is established, where Conditional Value-at-Risk (CVaR) is employed as a robust metric to quantify tail risks. Finally, a controlled evolution-based risk mitigation scheme is introduced to refine long-term dispatch strategies for mitigating tail risks. Case studies on a modified IEEE-39 bus system incorporating real-world data substantiate the efficacy of the proposed method. I Introduction The environmental and economic benefits of renewable energy (RE) drive its growing integration into modern power systems. However, RE exhibits seasonality and strong dependence on weather conditions. For instance, due to variations in solar radiation, photovoltaic (PV) generation in Germany during winter typically amounts to only 20–40% of that in summer, while in the United Kingdom, wind power shows a strong seasonal profile, with average capacity factors approaching 45% in winter but declining to below 20% in summer [ 1 , 2 ] . Neglecting these long-term characteristics can result in a substantial underestimation of power supply tail risks [ 3 ] , underscoring the urgent need for advanced long-term risk assessment and mitigation methods. Accurately characterizing RE uncertainty is essential for risk assessment. On short timescales, uncertainty can be modeled using forecasting techniques such as ARIMA models, machine learning, or statistical resampling [ 4 , 5 , 6 ] , and expressed by uncertainty sets, probabilistic scenarios, or fuzzy distributions. In contrast, long-term scenario generation must capture seasonal patterns and the possibility of extreme events, which are difficult to predict. Traditional clustering-based methods [ 7 ] provide useful insights but struggle to represent joint extreme events. Monte Carlo simulation [ 8 ] can generate numerous samples but struggles with efficiently capturing tail risks. Copula-based methods [ 9 ] offer improved modeling of joint extreme events, yet face computational challenges in high dimensions. Currently, most risk assessment models focus on short-term horizons, addressing uncertainties in renewable generation, load forecasting errors, and extreme weather events [ 10 , 11 , 12 ] . However, these methods often fail to capture persistent extreme conditions, such as prolonged low-wind periods or cloudy seasons, that significantly influence system risk but remain invisible in short-term analyses. Traditional long-term risk assessments rely on simplified methods such as power duration curves [ 13 ] , which cannot adequately address the increasing RE variability. More recent advances include optimization-based approaches, such as stochastic optimization (SO) [ 14 ] , robust optimization (RO) [ 15 ] , and distributionally robust optimization (DRO) [ 16 ] , typically combined with risk metrics like loss-of-load probability (LOLP) [ 17 ] , expected unserved energy (EUE) [ 18 ] , and other reserve adequacy indices [ 2 ] . These methods often assume idealized dispatch, overlooking realistic long-term operational strategies that crucially affect system risks, especially tail risks, under uncertainties [ 19 ] . Consequently, they may misestimate system risks and provide limited insight for risk mitigation. To address these gaps, this paper proposes a comprehensive framework for dispatch-embedded long-term tail risk assessment and mitigation. First, a representative scenario generation approach is developed, integrating multi-timescale Copula modeling to capture long-range dependencies and variability in RE. Subsequently, a long-term risk assessment model is formulated, embedding dispatch strategies and applying Conditional Value-at-Risk (CVaR) to quantify tail risks. Finally, a controlled evolution-based risk mitigation method is proposed to refine the long-term dispatch strategies. II Renewable Power System Model II-A Outline The renewable power system consists of thermal power generators (TPGs), renewable energy (RE), seasonal energy storage (SES), demand response (DR), load demand (LD), and transmission lines (TLs), as shown in Fig. 1 . Each type of equipment is represented by a set \Omega_{\bullet} , where \bullet corresponds to TPG, RE, SES, DR, LD, or TL, respectively. The time period horizon is denoted by set \Omega_{\mathrm{T}} , with index t and number \left|\Omega_{\mathrm{T}}\right| . Fig. 1: Renewable power system diagram. II-B System Constraints II-B 1 TPG Each TPG g\in\Omega_{\mathrm{TPG}} is subject to: \displaystyle({x_{g,t}^{\mathrm{TPG}}}-{x_{g,t+1}^{\mathrm{TPG}}})(X_{g,t}^{\mathrm{ON}}-T_{g}^{\mathrm{ON}})\geq 0 (1a) \displaystyle({x_{g,t}^{\mathrm{TPG}}}-{x_{g,t+1}^{\mathrm{TPG}}})(X_{g,t}^{\mathrm{OFF}}-T_{g}^{\mathrm{OFF}})\geq 0 (1b) \displaystyle{y_{g,t+1}^{\mathrm{TPG}}}-{z_{g,t+1}^{\mathrm{TPG}}}={x_{g,t+1}^{\mathrm{TPG}}}-{x_{g,t}^{\mathrm{TPG}}} (1c) \displaystyle{x_{g,t}^{\mathrm{TPG}}},{y_{g,t}^{\mathrm{TPG}}},{z_{g,t}^{\mathrm{TPG}}}\in\left\{0,1\right\} (1d) \displaystyle\underline{P}_{g}^{\mathrm{TPG}}{x_{g,t}^{\mathrm{TPG}}}\leq{p_{g,t}^{\mathrm{TPG}}}\leq\overline{P}_{g}^{\mathrm{TPG}}{x_{g,t}^{\mathrm{TPG}}} (1e) \displaystyle{p_{g,t}^{\mathrm{TPG}}}-{p_{g,t+1}^{\mathrm{TPG}}}\leq{R_{g}^{\mathrm{D}}}{x_{g,t+1}^{\mathrm{TPG}}}+\overline{P}_{g}^{\mathrm{TPG}}(1-{x_{g,t+1}^{\mathrm{TPG}}}) (1f) \displaystyle{p_{g,t+1}^{\mathrm{TPG}}}-{p_{g,t}^{\mathrm{TPG}}}\leq{R_{g}^{\mathrm{U}}}{x_{g,t}^{\mathrm{TPG}}}+\overline{P}_{g}^{\mathrm{TPG}}(1-{x_{g,t}^{\mathrm{TPG}}}) (1g) where binary variables x_{g,t}^{\mathrm{TPG}} , y_{g,t}^{\mathrm{TPG}} , and z_{g,t}^{\mathrm{TPG}} indicate commitment, startup, and shutdown states. Variable p_{g,t}^{\mathrm{TPG}} is the dispatched power of TPG g . Constraints ( 1a )-( 1b ) enforce minimum on-time T_{g}^{\mathrm{ON}} and off-time T_{g}^{\mathrm{OFF}} durations, based on the accumulated on and off durations X_{g,t}^{\mathrm{ON}} and X_{g,t}^{\mathrm{OFF}} . The relationship among x_{g,t}^{\mathrm{TPG}} , y_{g,t}^{\mathrm{TPG}} , and z_{g,t}^{\mathrm{TPG}} is given by ( 1c ). TPG generation limits [\underline{P}_{g}^{\mathrm{TPG}},\overline{P}_{g}^{\mathrm{TPG}}] are enforced by ( 1e ). Ramp-down/up limits R_{g}^{\mathrm{D}} and R_{g}^{\mathrm{U}} are captured by ( 1f )-( 1g ). II-B 2 RE Each RE r\in\Omega_{\mathrm{RE}} satisfies: 0\leq p_{r,t}^{\mathrm{RE}}\leq\overline{P}_{r,t}^{\mathrm{RE}} (2) where variable {p_{r,t}^{\mathrm{RE}}} is the utilized RE power, and \overline{P}_{r,t}^{\mathrm{RE}} is the available RE generation. II-B 3 SES Each SES e\in\Omega_{\mathrm{SES}} is governed by: \displaystyle 0\leq p_{\mathrm{CH},e,t}^{\mathrm{SES}}\leq{a_{e,t}^{\mathrm{SES}}}\overline{P}_{e}^{\mathrm{SES}} (3a) \displaystyle 0\leq p_{\mathrm{DC},e,t}^{\mathrm{SES}}\leq(1-{a_{e,t}^{\mathrm{SES}}})\overline{P}_{e}^{\mathrm{SES}} (3b) \displaystyle{a_{e,t}^{\mathrm{SES}}}\in\left\{0,1\right\} (3c) \displaystyle w_{e}^{\mathrm{D}}\overline{E}_{e}^{\mathrm{SES}}\leq E_{e,t}^{\mathrm{SES}}\leq w_{e}^{\mathrm{U}}\overline{E}_{e}^{\mathrm{SES}} (3d) \displaystyle{E_{e,t}^{\mathrm{SES}}}={E_{e,t-1}^{\mathrm{SES}}}+\left(\eta_{e}^{\mathrm{CH}}p_{\mathrm{CH},e,t}^{\mathrm{SES}}-p_{\mathrm{DC},e,t}^{\mathrm{SES}}/\eta_{e}^{\mathrm{DC}}\right)\Delta t (3e) \displaystyle{E_{e,0}^{\mathrm{SES}}}=E_{\mathrm{INT},e}^{\mathrm{SES}} (3f) \displaystyle{E_{e,\left|\Omega_{\mathrm{T}}\right|}^{\mathrm{SES}}}\geq{E_{e,0}^{\mathrm{SES}}} (3g) where variables p_{\mathrm{CH},e,t}^{\mathrm{SES}} and p_{\mathrm{DC},e,t}^{\mathrm{SES}} are charge and discharge power, and \overline{P}_{e}^{\mathrm{SES}} is the power capacity. Binary variable a_{e,t}^{\mathrm{SES}} prevents simultaneous charge and discharge. E_{e,t}^{\mathrm{SES}} is the state of charge (SoC); w_{e}^{\mathrm{D}} and w_{e}^{\mathrm{U}} are upper/lower bound coefficients for energy capacity \overline{E}_{e}^{\mathrm{SES}} . SoC dynamics are defined by ( 3e ), with \eta_{e}^{\mathrm{CH}} and \eta_{e}^{\mathrm{DC}} as charge/discharge efficiencies, and \Delta t as a time period duration. E_{\mathrm{INT},e}^{\mathrm{SES}} is the initial SoC, and the final SoC must not fall below E_{\mathrm{INT},e}^{\mathrm{SES}} , as per ( 3g ). II-B 4 DR Each DR m\in\Omega_{\mathrm{DR}} is subject to: 0\leq p_{m,t}^{\mathrm{DR}}\leq\overline{P}_{m}^{\mathrm{DR}} (4) where variable p_{m,t}^{\mathrm{DR}} is the utilized DR power and \overline{P}_{m}^{\mathrm{DR}} is the DR capacity. II-B 5 Power Supply The system power supply is: \displaystyle\begin{array}[]{c}\sum\limits_{g\in{\Omega_{\mathrm{TPG}}}}{p_{g,t}^{\mathrm{TPG}}}+\sum\limits_{r\in{\Omega_{\mathrm{RE}}}}{p_{r,t}^{\mathrm{RE}}}+\sum\limits_{m\in{\Omega_{\mathrm{DR}}}}p_{m,t}^{\mathrm{DR}}\\ +\sum\limits_{e\in{\Omega_{\mathrm{SES}}}}{\left(p_{\mathrm{DC},e,t}^{\mathrm{SES}}-p_{\mathrm{CH},e,t}^{\mathrm{SES}}\right)}+p_{t}^{\mathrm{B+}}\geq\sum\limits_{d\in{\Omega_{\mathrm{LD}}}}{\overline{P}_{d,t}^{\mathrm{LD}}}\end{array} \displaystyle p_{t}^{\mathrm{B+}}\geq 0 (5c) where \overline{P}_{d,t}^{\mathrm{LD}} is the demand of LD d\in\Omega_{\mathrm{LD}} . Variable p_{t}^{\mathrm{B+}} is the non-negative emergency power support, representing system’s power shortages. II-B 6 TL The power flow for each TL l\in\Omega_{\mathrm{TL}} is limited by: \begin{array}[]{c}\left|{\begin{array}[]{*{3}{c}}{\sum\limits_{g\in{\Omega_{\mathrm{TPG}}}}{{\rm{SF}}_{l,g}^{\mathrm{TPG}}{p_{g,t}^{\mathrm{TPG}}}}+\sum\limits_{r\in{\Omega_{\mathrm{RE}}}}{{\rm{SF}}_{l,r}^{\mathrm{RE}}p_{r,t}^{\mathrm{RE}}}}\\ +\sum\limits_{e\in\Omega_{\mathrm{SES}}}{\rm{SF}}_{l,e}^{\mathrm{SES}}\left(p_{\mathrm{DC},e,t}^{\mathrm{SES}}-p_{\mathrm{CH},e,t}^{\mathrm{SES}}\right)\\ {+\sum\limits_{m\in{\Omega_{\mathrm{DR}}}}{{\rm{SF}}_{l,m}^{\mathrm{DR}}p_{m,t}^{\mathrm{DR}}}-\sum\limits_{d\in{\Omega_{\mathrm{LD}}}}{\rm{SF}}_{l,d}^{\mathrm{LD}}\overline{P}_{d,t}^{\mathrm{LD}}}\end{array}}\right|\leq{\overline{\mathrm{PL}}_{l}}\end{array} (6) where parameters {\rm{SF}}_{l,i}^{\bullet} represents the shift factor of TL l for equipment i\in\Omega_{\bullet} , and {\overline{\mathrm{PL}}_{l}} is the TL capacity. II-C Objective Function In the renewable power system, the cost for period t consists of the TPG cost L_{t}^{\mathrm{TPG}} , ES cost L_{t}^{\mathrm{SES}} ( 7b ), DR cost L_{t}^{\mathrm{DR}} ( 7c ), and emergency supply cost L_{t}^{\mathrm{EME}} ( 7d ): \displaystyle L_{t}^{\mathrm{TPG}}=\sum\nolimits_{g}\left(c_{g}^{\mathrm{SU}}y_{g,t}^{\mathrm{TPG}}+c_{g}^{\mathrm{SD}}z_{g,t}^{\mathrm{TPG}}+c_{g}^{\mathrm{TPG}}p_{g,t}^{\mathrm{TPG}}\Delta t\right) (7a) \displaystyle L_{t}^{\mathrm{SES}}=\sum\nolimits_{e}c_{e}^{{\rm{SES}}}\left(p_{\mathrm{CH},e,t}^{\mathrm{SES}}+p_{\mathrm{DC},e,t}^{\mathrm{SES}}\right)\Delta t (7b) \displaystyle L_{t}^{\mathrm{DR}}=\sum\nolimits_{m}c_{m}^{\mathrm{DR}}p_{m,t}^{\mathrm{DR}}\Delta t (7c) \displaystyle L_{t}^{\mathrm{EME}}={c^{\mathrm{EME}}}p_{t}^{\mathrm{B+}}\Delta t (7d) where c_{g}^{\mathrm{SU}} , c_{g}^{\mathrm{SD}} , c_{g}^{\mathrm{TPG}} , c_{e}^{{\rm{SES}}} , and c_{m}^{{\rm{DR}}} are respective prices for each component, while c^{\mathrm{EME}}

Read the complete text