Security-Constrained Economic Dispatch (SCED)

• SCED is an optimization framework designed to minimize generation cost while satisfying power balance and security constraints, ensuring reliable grid operation.
• Decomposition methods like Lagrangian and augmented Lagrangian relaxation accelerate convergence and balance data exchange with privacy needs.
• Recent extensions incorporate stochastic, chance-constrained, and risk-sensitive formulations to manage uncertainty, cyber threats, and renewable integration.

Security-Constrained Economic Dispatch (SCED) is a foundational optimization framework in power system operations, tasked with minimizing generation cost while maintaining reliability through adherence to power balance and security constraints, including network and contingency-based limits. SCED is central to real-time market clearing and system reliability assurance and is formulated to address both the economic and operational requirements of large-scale electricity networks.

1. Mathematical Formulation of SCED

SCED is typically posed as a constrained optimization to determine optimal real-power generation dispatch under engineering and security constraints:

• Objective: Minimize total generation cost. For quadratic generator cost $C_i(P_{g_i})=a_i + b_i P_{g_i} + c_i P_{g_i}^2$, the centralized SCED can be written as:

$$\min_{\{P_{g_i},\theta_i\}}\;\;\sum_{i\in G}\Big(a_i + b_iP_{g_i} + c_iP_{g_i}^2\Big)$$

where $G$ indexes the controllable generators and $\theta_i$ are bus voltage angles (Amini et al., 2015).

• Power Balance: For each bus $i$,

$$P_{g_i} - P_{L_i} - \sum_{j:(i,j)\in L_{sys}}F_{ij} = 0$$

where $F_{ij} = B_{ij}(\theta_i - \theta_j)$ is the DC flow on line $(i,j)$, $L_{sys}$ denotes the set of system lines (Amini et al., 2015).

• Branch Flow and Generator Constraints:

$$P_{g_i}^{\min} \leq P_{g_i} \leq P_{g_i}^{\max} \qquad \forall i\in G$$

$$-F_{ij}^{\max}\leq F_{ij}\leq F_{ij}^{\max} \qquad \forall (i,j)\in L_{sys}$$

Further, in N-1 SCED, post-contingency versions of all flow constraints are enforced for each credible contingency $k$:

$$-F_{\ell}^{(k)}\leq \sum_{i\in G}\text{PTDF}_{\ell,i}^{(k)}(g_i - d_i) \leq F_{\ell}^{(k)} \quad \forall \ell, \forall k$$

where $\text{PTDF}_{\ell,i}^{(k)}$ are contingent shift factors (Chen et al., 2021).

2. Decomposition Approaches: Lagrangian and Augmented Lagrangian Relaxation

Distributed solution of SCED becomes vital for large-scale and multi-area systems. Amini et al. (Amini et al., 2015) detail two key decomposition strategies:

• Lagrangian Relaxation (LR): The network is partitioned (e.g., at tie-lines) with the inter-area (coupling) constraints dualized. Each area solves a local subproblem parameterized by multipliers $(\lambda, \mu)$ for tie-line equality and inequality coupling constraints. Multipliers are updated via subgradient iterations:

$$\lambda^{(v+1)} = \lambda^{(v)} + k^{(v)} s^{(v)}, \qquad \mu^{(v+1)} = \max\{0, \mu^{(v)} + k^{(v)} t^{(v)}\}$$

with $k^{(v)} = 1/(a+bv)$ and $s^{(v)}, t^{(v)}$ the primal coupling violations.

• Augmented Lagrangian Relaxation (ALR): Adds quadratic penalties for coupling violations:

$$\mathcal{L}_{\text{AL}}(X,\lambda) = \sum_{i\in G} C_i(P_{g_i}) + \lambda^\mathsf{T}g_{\text{int}}(X) + \tfrac{\rho}{2}\|g_{\text{int}}(X)\|^2$$

This accelerates convergence by penalizing disagreement at boundaries, at the cost of additional data exchange (e.g., net-injection sharing).

• Novel Tie-Line Modeling: Instead of coupling via flows, the tie-line constraint $|F_{\text{tie}}|\leq F^{\max}_{\text{tie}}$ is decomposed into equivalent local net-injection constraints at tie buses, allowing complete decomposition into local (area-specific) constraints.

Performance Comparison on a 6-bus System (Amini et al., 2015)

Method	Iterations	CPU Time (s)	Final Mismatch (pu)	Shared Data
LR	225	78.99	$1.34\times10^{-2}$	Tie-line angle
ALR	51	18.77	$3.30\times10^{-3}$	Angle + net boundary injection

Convergence is substantially faster with ALR due to the rich data exchange. There is a trade-off between information privacy (favoring LR) and solution speed/accuracy (favoring ALR).

3. Security Modeling: Contingencies and Vulnerabilities

Modern SCED implements preventive or corrective security via additional post-contingency constraints. The economic cost of such security is quantified by the Price of Security (PoS), defined as the ratio of SCED to unconstrained economic dispatch cost (Hajiesmaili et al., 2017):

$$\text{PoS} = \frac{c^*_{\rm sced}}{c^*_{\rm ed}} \ge 1$$

PoS is maximized in scenarios where cheap generation is abundant and demand is unevenly distributed, forcing expensive local supply under security constraints. Analytical and empirical results show SCED can incur significant surcharges over pure economic dispatch, especially in topologies with constrained corridors or imbalanced generation/demand.

SCED's resilience can be undermined by cyber-physical attacks, including load-redistribution (LR) attacks and false data injection targeting both load and (increasingly) distributed renewable generation, such as solar (Verma et al., 16 Sep 2025, Kaviani et al., 2020). Enhanced models now include explicit manipulation of renewable injection data as an attack vector, with analysis showing dramatic increases in post-attack costs and system vulnerabilities, especially in high-solar scenarios.

4. Stochastic, Chance-Constrained, and Risk-Sensitive Extensions

Conventional SCED is deterministic; advanced approaches internalize uncertainty via:

• Scenario-based Stochastic Look-Ahead Dispatch (SLAD): Two-stage, multi-period, scenario-based models optimize expected system cost over a distribution of renewable and load scenarios, using accelerated Benders decomposition for tractable real-time solution on industry-scale networks (Zhao et al., 2023). SLAD provides quantifiable cost and reliability gains over deterministic formulations.
• Chance-Constrained SCED: Enforces probabilistic (risk-based) satisfaction of operational constraints under renewable forecasting errors (Zhang et al., 2024). Recent results show that convex-hull compression of scenario sets preserves the solution exactly, dramatically reducing constraint counts and computational overhead, and introducing risk metrics (solution and compression risk) to quantify safety under reduced scenario sets.
• Risk-Sensitive SCED (R-SCED): Incorporates tail-risk aversion via Conditional Value-at-Risk (CVaR) penalties on load-shedding in post-contingency states (Madavan et al., 19 Feb 2025). The CVaR-parameter $\alpha$ tunes the trade-off between expected cost and robust resilience against contingencies. Benders decomposition enables tractable solution even for large numbers of contingencies. Market pricing based on total marginal cost (including contingencies, S-LMP) ensures revenue adequacy, compared to pure energy-only models (N-LMP), which may run deficits.

5. Computational Techniques and Real-Time Feasibility

The high dimensionality and complexity of SCED necessitate efficient algorithmic strategies:

• Formulation Structure: PTDF-based models are prevalent but suffer from high-density constraint matrices, slowing barrier methods. Sparsity-preserving formulations in voltage angles or the B–$\theta$ representation are strongly favored for SCED and OPF quadratic programs, offering up to 50× speed-up over dense PTDF and enabling real-time solution at industry scale (Bao et al., 2023).
• Machine Learning Proxies: To enable high-speed risk assessment, ML-based proxies (e.g., classification-then-regression deep neural nets) can emulate SCED solutions at millisecond scale with relative errors below 0.6%, supporting real-time scenario screening for system operators (Chen et al., 2021).
• Distributed/Decomposed Methods: Augmented Lagrangian schemes (ALR) and distributed ADMM-type approaches allow efficient, privacy-preserving solution of SCED across network regions and integrated transmission–distribution systems, with convergence to near-optimal dispatch (Amini et al., 2015, Tian et al., 2023).
• Reinforcement Learning with Safety Layers: SCED has been extended with safe RL controllers, where actions are projected into the secure action space using set-based reachability and zonotope representations to enforce operational and contingency constraints in microgrid applications (Eichelbeck et al., 2022).

6. Extensions: Operability, Flexibility, and Advanced Security

• Cardinality-Minimization SCED: Introduces a framework to directly penalize the cardinality (number) of transmission lines operating in emergency or stressed zones, exploiting the brief use of high-emergency ratings to minimize both cost and exposure to sustained high-risk operation, and ameliorating prolonged scarcity pricing (Troxell et al., 2021).
• Adversarial-Scenario Sampling: Traditional adversarial optimization for SCED/OPF is susceptible to over-optimistic robustness due to local optima in the inner maximization. Adversarial-sampling methods using Sequential Monte Carlo (SMC) and gradient-based Markov chain Monte Carlo (MALA) produce diverse, high-risk contingency sets, eliminating false negatives and delivering probabilistic risk quantification (Dawson et al., 2023).
• Coordinated Frequency-Constrained Dispatch: In integrated transmission–distribution systems with considerable inverter-based resources, frequency security is enforced jointly with network constraints. Joint chance-constrained stochastic economic dispatch co-optimizes energy, reserve, and frequency security indices (RoCoF, nadir, steady-state) via a distributed, two-layer ADMM framework (Tian et al., 2023).

7. Practical Implications and Directions

SCED remains indispensable for both market operation and system security. Recent developments enable its extension to uncertainty, cyber-physical threats, large-scale systems, and integrated networks. Research confirms that advanced formulations and distributed algorithms can now meet real-time computational requirements on industry-scale grids under increasing renewable penetrations (Zhao et al., 2023, Tian et al., 2023, Zhang et al., 2024).

Critical trade-offs exist between solution fidelity, convergence rate, information privacy, and operational risk. Parameterization (e.g., CVaR level, compression risk, penalty weights) should be tuned to system priorities—cost, reliability, robustness to attack, or rapid response. Market design, including pricing for risk and contingency cost recovery, must co-evolve with SCED practice to ensure both economic efficiency and revenue adequacy (Madavan et al., 19 Feb 2025).

Continued integration of advanced optimization, machine learning, and stochastic control methodologies will further enhance SCED’s role as the cornerstone of secure and economical power system operation.