Source-linked AI summary
Chance Constrained Optimal Power Flow: Risk-Aware Network Control under Uncertainty
Daniel Bienstock, Michael Chertkov, Sean Harnett
TL;DR
Standard OPF is risk-unaware of renewable fluctuations, which can overload transmission lines and compromise grid stability. The paper develops a chance-constrained OPF that converts this uncertainty into a convex optimization problem, yielding safer and cheaper operation under supported conditions.
Problem
Standard OPF does not account for renewable-output uncertainty, allowing wind-driven power-flow changes to overload lines and threaten grid stability.
Method
CC-OPF models uncertain generation through chance constraints and, under Gaussian assumptions, produces a convex deterministic optimization problem, with robust extensions for parameter and model error.
Results
CC-OPF delivers feasible, safer, and cheaper operation in cases where standard OPF would leave many lines overloaded for an unacceptably large fraction of time.
Takeaways & Limitations
CC-OPF provides a computationally efficient generation re-dispatch paradigm that accounts for wind uncertainty while retaining a conic optimization formulation.
Takeaways & Limitations
The nominal formulation may be sensitive to small errors in wind-distribution estimates, while data-robust formulations can become too large and nonconvex for present-day solvers.
Abstract
from arXiv · showhide
When uncontrollable resources fluctuate, Optimum Power Flow (OPF), routinely used by the electric power industry to re-dispatch hourly controllable generation (coal, gas and hydro plants) over control areas of transmission networks, can result in grid instability, and, potentially, cascading outages. This risk arises because OPF dispatch is computed without awareness of major uncertainty, in particular fluctuations in renewable output. As a result, grid operation under OPF with renewable variability can lead to frequent conditions where power line flow ratings are significantly exceeded. Such a condition, which is borne by simulations of real grids, would likely resulting in automatic line tripping to protect lines from thermal stress, a risky and undesirable outcome which compromises stability. Smart grid goals include a commitment to large penetration of highly fluctuating renewables, thus calling to reconsider current practices, in particular the use of standard OPF. Our Chance Constrained (CC) OPF corrects the problem and mitigates dangerous renewable fluctuations with minimal changes in the current operational procedure. Assuming availability of a reliable wind forecast parameterizing the distribution function of the uncertain generation, our CC-OPF satisfies all the constraints with high probability while simultaneously minimizing the cost of economic re-dispatch. CC-OPF allows efficient implementation, e.g. solving a typical instance over the 2746-bus Polish network in 20 seconds on a standard laptop.
1. Formulating Chance-Constrained Optimum Power Flow Models.
The paper formulates chance-constrained OPF for transmission-grid dispatch under renewable uncertainty, focusing on probabilistic line-limit violations while retaining tractable optimization. It motivates replacing risk-unaware standard OPF with constraints that control overload probability and incorporates uncertain injections, generator response, and forecast uncertainty.
- 1.1. Transmission Grids: Controls and Limits.: Transmission systems use coordinated controls across multiple time scales to balance demand and generation while maintaining stable operation.The paper places tertiary-control OPF within this broader hierarchy, alongside automatic frequency controls and longer-horizon unit commitment.
- 1.1. Transmission Grids: Controls and Limits.: Grid stability includes synchrony, adequate voltage, and power flows that remain within established line bounds.The paper focuses on line limits because synchrony and voltage problems are considered rare under normal operating conditions.
- 1.2. OPF – Standard Generation Dispatch (tertiary control).: Standard OPF resets generator outputs over a transmission control area by minimizing generation cost subject to power-flow, line-limit, and generation-bound constraints.The formulation is described as a convex quadratic program using estimated demand and controllable-generator outputs.
- 1.2. OPF – Standard Generation Dispatch (tertiary control).: The standard OPF model commonly uses a linearized DC power-flow approximation because full AC equations introduce nonconvexities.The approximation fixes voltage magnitudes, assumes small phase differences, and ignores thermal losses.
- 1.3. Chance constrained OPF: motivation.: Risk-unaware OPF can produce real-time flows that violate line and generator constraints when actual conditions differ from estimated demand or renewable output.Large renewable deviations can substantially overload lines, increasing the likelihood of tripping and compromising grid stability.
- 1.3. Chance constrained OPF: motivation.: Deterministically preventing overloads can require excessively conservative redispatch and may cause extreme electricity-market price volatility.The paper presents chance constraints as a less conservative alternative to strict deterministic line-limit enforcement.
- 1.4. Using chance constraints.: Chance constraints replace dynamic overload-frequency requirements with a static requirement that each line’s probability of exceeding its limit remain small.This proxy avoids resolving grid dynamics throughout the generator-dispatch window.
- 1.5. Uncertain power sources.: The uncertain-source formulation models renewable injections probabilistically and uses Gaussian assumptions to obtain a computationally tractable optimization formulation.A data-robust version allows Gaussian parameters to be unknown within specified windows, addressing parameter misestimation and model error.
2. Solving the Models.
The paper formulates CC-OPF with affine controls for renewable uncertainty, then develops computational methods for solving the resulting chance-constrained models efficiently. A cutting-plane approach exploits convex subproblems and rapidly converges on large realistic grids, while data uncertainty remains an important robustness concern.
- Model formulation: The affine control introduces α_i alongside standard generator setpoints to compensate renewable fluctuations while preserving power balance.Viability requires generator outputs under the control law and uncertain outputs to exactly match total demand.
- Model formulation: Each stochastic phase angle and line flow is an affine function of the renewable deviations ω_i.This property enables the chance constraints to be expressed through the flow means and variances.
- Model formulation: The chance constraints impose probabilistic limits on line flows and generator outputs, but the resulting formal problem is not directly suited to standard optimization algorithms.The paper therefore develops an efficient solution method for relevant classes of the formulation.
- Model formulation: The CC-OPF objective minimizes expected stochastic generation cost and is convex quadratic in the nominal generation and affine-control variables.The formulation accounts for random wind output while retaining convex quadratic generation costs under the stated cost assumptions.
- Computational solution: 2746 buses, 3514 lines, and 8 wind farms produced a large conic model that challenged an 8-core CPLEX workstation, motivating specialized solution procedures.The reported instance contained 36,625 variables and 38,507 constraints, including 6,242 conic constraints.
- Computational solution: The cutting-plane algorithm solves linearly constrained convex quadratic master problems, whose objective values provide valid lower bounds, and experiments show robust, rapid convergence to an optimum.The method discovers only the conic constraints needed at optimality rather than incorporating the full set initially.
- Computational solution: 25 seconds was the total run-time for a typical Polish-grid convergence instance, where the algorithm quickly reduced infeasibility despite a nearly flat objective.The main computational challenge was finding a feasible solution satisfying the chance constraints, not materially improving the objective.
- Data robustness: Small errors in wind-variance estimates caused only small empirical degradation of the chance constraints in the reported experiments.The paper treats imprecise variance estimates as a threat to the usefulness of the nominal chance-constrained formulation and studies robustness explicitly.
3. Experiments/Results.
The standard extension of OPF uses fixed renewable-output estimates and proportional generator ramping, but chooses these controls without accounting for renewable-output stochasticity. CC-OPF instead selects risk-aware and topology-aware nominal outputs and ramping parameters.
- 3. Experiments/Results.: Standard OPF handles renewable fluctuations using fixed output estimates and proportional ramping of traditional generators.These fixed generator levels and ramping rates are selected without considering the stochastic nature of renewable output.
3.1. Failure of standard OPF.
The experiments show that standard OPF can produce substantial line-overload risk when renewable fluctuations are included, even when its nominal solution is thermally safe. The proposed approach instead uses risk-aware generator reconfiguration and may avoid curtailing available wind power.
- 3.1. Failure of standard OPF.: Standard OPF selects generator controls without considering stochastic renewable output, whereas CC-OPF produces risk-aware, topology-aware controls.The topology awareness reflects network proximity to wind farms.
- 3.1. Failure of standard OPF.: Five IEEE 118-bus lines exceeded their limits 8% or more of the time under standard OPF with 30%-of-mean Gaussian wind fluctuations.The nominal standard-OPF solution was safely within thermal limits before renewable fluctuations were modeled.
- 3.1. Failure of standard OPF.: Six lines in the stressed 2746-bus Polish grid exceeded their limits over 45% of the time, while one line exceeded its limit over 10% of the time.The case used 10 wind sources at 2% penetration after all loads were scaled up by 10%.
- 3.1. Failure of standard OPF.: Reconfiguring standard generators is proposed as an alternative to wind curtailment that could use available wind power and lower operating costs.The paper presents this consequence conditionally: the benefit depends on whether the proposed solution is successful.
3.2. Cost of reliability under high wind penetration.
The experiments compare standard OPF with CC-OPF under high wind penetration, showing that reliability through curtailment can impose substantially higher generation costs. CC-OPF instead maintains feasibility at higher penetration while addressing geographically distributed line overloads.
- The 118-bus experiment shows five lines exceeding their limits 8% or more of the time under standard OPF with 5% average wind penetration.
- CC-OPF costs 264,000 at 30% penetration, while standard OPF costs 1,275,020 at 5% penetration.
- 30% wind penetration is feasible under CC-OPF, whereas standard OPF produces five excessive line overloads even with a 10% buffer.
- Reducing standard-OPF wind penetration from 30% to 5% relieves the lines but more than quadruples cost relative to CC-OPF.
- The generator changes required by CC-OPF are not obviously localized near violated lines, indicating that simple local adjustments may not bypass the model.
3.3. Non-locality.
The paper presents CC-OPF as a centralized approach whose adjustments are not confined to generators near overloaded lines. It therefore contrasts with a strategy of making small local corrections after standard OPF.
- Generator-output differences between standard OPF and CC-OPF are not obviously associated with individual line violations.
3.4. Increasing penetration.
The experiments examine how system loading and wind penetration constrain CC-OPF feasibility. In the studied 39-bus system, 30% penetration remains feasible but lies near a threshold beyond which infrastructure investment is required.
- The 39-bus system is tested as wind penetration increases under line limits scaled by .7 to represent heavy loading.
- Figure 3.4 maps standard-OPF high-probability overload lines and shows generator differences between standard and chance-constrained solutions.
- 30% wind penetration is feasible under CC-OPF with a .02 per-line overload probability, but lies close to the dangerous feasibility threshold.
- Beyond the 30% threshold, higher penetration is impossible in this model without upgrading lines and investing in other hardware.
3.5. Changing locations for wind farms.
Wind-farm placement materially changes the penetration that the studied networks can support. The location experiments therefore treat siting as a critical factor in achieving higher renewable penetration.
- The 39-bus case displays 0.1%, 8%, and 30% average wind-penetration scenarios with a .02 maximum chance of line-limit violation.
- The 39-bus problem becomes infeasible beyond 30% wind penetration.
- In the 30-bus case, one wind-farm placement remains feasible up to 10% penetration, while another supports up to 55%.
- The location experiment concludes that choosing wind-farm locations is critical for achieving high renewable penetration.
- A 9-bus example shows that admissible wind configurations can substantially reverse flow on a particular line.
3.6. Reversal of line flows.
Shifting winds can reverse power flow on a line, while the study evaluates CC-OPF performance under errors in the assumed wind-power distribution.
- 3.6. Reversal of line flows.: Shifting winds cause the flow on an orange line to change direction with a large absolute difference in the 9-bus case.The case uses 25% average penetration from two wind sources.
- 3.6. Reversal of line flows.: The distribution-error experiments compare non-Gaussian true wind distributions and Gaussian distributions with altered mean or standard deviation.The experiments use the BPA grid with 2,209 buses and 2,866 lines.
3.7. Out-of-sample tests.
Out-of-sample Monte Carlo tests examine how CC-OPF behaves under non-Gaussian wind distributions and forecast errors in means and standard deviations.
- 3.7. Out-of-sample tests.: Monte Carlo tests draw from Laplace, logistic, Weibull, Student’s t, and Cauchy distributions after solving under a Gaussian assumption.The Gaussian solution targets a maximum overload probability of about 2.27%.
- 3.7. Out-of-sample tests.: A Weibull distribution with shape parameter 1.2 approximately doubles the desired maximum overload chance, while logistic and Student’s t distributions produce lower-than-desired maxima.The Weibull case is highly asymmetric; the logistic and Student’s t cases suggest conservatism.
- 3.7. Out-of-sample tests.: Approximately 2.27% is the target maximum chance of exceeding a line limit for cases within two standard deviations of the mean.The out-of-sample tests use 10,000 samples on the BPA case.
- 3.7. Out-of-sample tests.: A 25% mean-forecast error results in a maximum 15% overload chance, whereas a 25% standard-deviation error results in less than 6%.The desired safety level is ǫ = 2.27% for each line.
3.8. Scalability.
On the 2,746-bus Polish grid, standard OPF exposes substantial overload risk, whereas CC-OPF sharply reduces that risk with a small cost increase and efficient computation.
- 3.8. Scalability.: Two lines are overloaded half the time and one line one-third of the time under standard OPF with 20% wind penetration.The simulation distributes wind across 18 wind farms.
- 3.8. Scalability.: The CC-OPF reduction in overload probability comes with a cost increase of less than one percent.The figure caption reports this cost comparison alongside the overload-probability reduction.
- 3.8. Scalability.: CC-OPF incorporates renewable-output uncertainty through chance constraints requiring line non-overload except for a small fraction of time.The formulation also accounts for local frequency response by controllable generators.
4. Discussions.
CC-OPF is formulated as a convex deterministic optimization problem and evaluated across power-grid datasets, with extensions proposed for evolving forecasts and nonlinear power flows.
- 4. Discussions.: CC-OPF is a convex conic optimization problem solved efficiently on realistic large-scale instances.The formulation relies on assumptions about exogenous uncertainty and linearized power-flow equations.
- 4. Discussions.: The experiments report feasible CC-OPF solutions where average-forecast standard OPF would leave many lines overloaded for an unacceptably large fraction of time.The summary also reports cheaper operation for CC-OPF in the compared high- and low-penetration settings.
- 4. Discussions.: The study finds that CC-OPF dispatch can differ substantially from naive adjustments targeting generators near overloaded lines.It also examines wind penetration, farm location, and reliability-related cost differences.
- 4. Discussions.: The approach assumes static forecasts, valid power-flow linearization, and line-congestion failures while excluding synchronicity and voltage-variation difficulties.These assumptions are presented as simplifications that permit later generalization.
- 4. Discussions.: Time-varying forecasts and loads can be incorporated by splitting the dispatch interval into sub-intervals and allowing more frequent redispatch.The proposed extension can also include generation ramping constraints.