PaperlayerView exact revision on arXiv

Source-linked AI summary

Fast Local Voltage Control under Limited Reactive Power: Optimality and Stability Analysis

Hao Zhu, Hao Jan Liu

arXiv:1510.07013v1math.OC

TL;DR

The paper addresses voltage regulation when DER variability and communication imperfections challenge centralized or decentralized control, while local droop methods may underuse limited VAR resources. It formulates a gradient-projection-based local VAR framework with bus-level limits, derives stability and stepsize conditions, and tests the approach on realistic three-phase systems. The analysis establishes convergence to a surrogate centralized problem under suitable parameters and generalizes earlier droop and delayed-droop designs.

  • Problem

    Rapid DER-driven voltage fluctuations and unreliable communication challenge real-time optimization-based voltage control, while stability-oriented droop designs can underutilize VAR resources.

  • Method

    The paper linearizes distribution power flow, formulates box-constrained VAR optimization, and applies gradient projection whose iterations become local updates using instantaneous bus-voltage measurements.

  • Results

    The proposed local schemes converge to a surrogate centralized VAR-control optimum under proper parameter choices and support stability analysis for parameter tuning.

  • Takeaways & Limitations

    Local VAR control can explicitly account for bus-level reactive-power limits without requiring real-time communication, while extending droop and delayed-droop control analysis.

Abstract

from arXiv · show

High penetration of distributed energy resources presents several challenges and opportunities for voltage regulation in power distribution systems. A local reactive power (VAR) control framework will be developed that can fast respond to voltage mismatch and address the robustness issues of (de-)centralized approaches against communication delays and noises. Using local bus voltage measurements, the proposed gradient-projection based schemes explicitly account for the VAR limit of every bus, and are proven convergent to a surrogate centralized problem with proper parameter choices. This optimality result quantifies the capability of local VAR control without requiring any real-time communications. The proposed framework and analysis generalize earlier results on the droop VAR control design, which may suffer from under-utilization of VAR resources in order to ensure stability. Numerical tests have demonstrated the validity of our analytical results and the effectiveness of proposed approaches implemented on realistic three-phase systems.

I. INTRODUCTION

The paper develops local VAR control to address voltage fluctuations and communication-related robustness challenges while explicitly respecting bus-level VAR limits. Its gradient-projection framework yields decentralized updates, stability guidance, and broader applicability than earlier droop-based designs.

  • Motivation: High DER penetration can create rapid voltage fluctuations from variable PV generation and abrupt electric-vehicle charging loads.These changes occur faster than existing on-load tap-changer and capacitor-bank controls can respond to.
  • Motivation: Communication-dependent optimization methods face practical delays and noise that can undermine real-time optimality and stability.The paper notes that high-quality measurement and control communication is not yet broadly available in distribution systems.
  • Motivation: Local VAR control can respond quickly using bus-voltage measurements, but stability constraints may increase OLTC tap changes and restrict VAR utilization.Earlier droop control requires sufficiently small slopes for stability, imposing a high penalty on inverter VAR output.
  • Contributions: The proposed gradient-projection framework models VAR control as a box-constrained quadratic optimization problem and decouples into local voltage-based updates.The framework explicitly represents each bus’s VAR limits and can attain a surrogate centralized optimum with suitable parameters.
  • Contributions: Inverse-Hessian-based diagonal stepsize scaling improves system conditioning and convergence speed without sacrificing the optimality condition.The stepsize design is motivated by acceleration from Newton’s method.
  • Analysis and evaluation: The paper establishes linearized distribution-flow modeling and stability conditions based on network topology and line admittance, then evaluates the methods on realistic three-phase feeders.The modeling assumptions include negligible losses and relatively flat voltage profiles, with approximation errors reported for voltage deviations.

III. GRAPH-REPRESENTATION BASED VAR CONTROL

The paper reformulates LinDistFlow with graph-based matrices, yielding a convex, limit-constrained VAR control problem and a locally implementable surrogate. The formulation establishes positive definiteness, accommodates broader network and modeling conditions, and motivates voltage-mismatch weighting and VAR penalties.

  • Graph-based LinDistFlow: The graph incidence matrix reformulation produces invertible matrices R and X and the linear voltage relation V = Xqg + ¯V.For tree-topology networks, the reduced incidence matrix M has full rank and is invertible; R and X encode resistance and reactance effects.
  • Graph-based LinDistFlow: R and X are positive definite because their quadratic forms use strictly positive line resistances and reactances.The proof establishes zT Rz > 0 for every nonzero z and applies the same argument to X.
  • Model scope and interpretation: The graph formulation extends to meshed networks and supports models with lossy lines, non-flat voltage profiles, and three-phase unbalanced systems.The paper states that the model and ensuing algorithms also hold for general distribution networks such as ring-topology systems, with numerical validation on realistic cases.
  • VAR optimization formulation: The resulting VAR problem is a box-constrained quadratic program with optional per-bus quadratic penalties for reactive-power supply.The penalty coefficient cj is nonnegative, and setting C = 0 recovers the no-penalty case.
  • VAR optimization formulation: Weighting voltage mismatch by the positive definite matrix B preserves convexity and supports a surrogate centralized problem for local control design.With unlimited VAR capability and C = 0, the weighted and unweighted objectives share the solution q† = q⋆ = X−1(µ−¯V).

IV. GRADIENT-PROJECTION METHOD

The gradient-projection method solves the box-constrained VAR optimization problem while naturally decoupling into feasible local updates based on each bus’s voltage and VAR input. Under constant reference voltage, these local updates attain the surrogate problem’s optimum, with stepsizes selectable offline and asynchronous updates supported.

  • IV. GRADIENT-PROJECTION METHOD: Gradient projection solves the box-constrained VAR problem by projecting gradient updates onto each bus’s feasible VAR interval.Coordinate-wise thresholding makes every iterate feasible when initialized within the box.
  • IV. GRADIENT-PROJECTION METHOD: Each bus’s gradient entry depends only on its local voltage magnitude and VAR input, eliminating the need for full-vector information.The gradient is expressed as V(t) − µ + Cq(t), whose j-th entry uses V_j(t) and q_j(t).
  • IV. GRADIENT-PROJECTION METHOD: The proposed local VAR update is essentially equivalent to the centralized gradient-projection solver for the surrogate problem.The projection at bus j is a local computation over its VAR interval.
  • IV. GRADIENT-PROJECTION METHOD: Under constant reference voltage, the fixed point of the iterative or local update achieves the optimum of the VAR control problem.This optimality applies to the constrained surrogate problem.
  • IV. GRADIENT-PROJECTION METHOD: Stepsize parameters can be determined offline from system topology and line admittance, while local updates can operate asynchronously under minimal centralized coordination.The paper identifies asynchronous operation as useful for plug-and-play microgrid functionality.
  • IV. GRADIENT-PROJECTION METHOD: Performance can degrade when limited VAR resources make the surrogate error-norm objective differ from the original objective.With unlimited, or limited but abundant, VAR resources, the surrogate solution is respectively identical or expected to closely approximate the desired optimum.

V. STABLE LOCAL VAR CONTROL

The stability analysis studies scaled iteration errors and gives a contraction-based condition for the local VAR update. This condition handles projection under limited VAR resources and coincides with classical Jacobian-eigenvalue conditions under the linearized model.

  • V. STABLE LOCAL VAR CONTROL: Stability is analyzed by diagonally scaling the iterates and measuring contraction of the scaled error relative to the optimum.The scaling uses D^-1/2, with projection treated as a nonexpansive mapping.
  • V. STABLE LOCAL VAR CONTROL: The contraction proof uses the positive definiteness of H and bounds the iteration’s scaling coefficient through the eigenvalues of I − H.The resulting scaled error norm converges to zero.
  • V. STABLE LOCAL VAR CONTROL: Under the linearized model, the proposed stability condition coincides with classical dynamic-system conditions based on Jacobian eigenvalues.The contraction-based proof additionally handles the projection operator under limited VAR resources.

A. Droop VAR Control

The paper relates its local gradient-projection framework to linear droop VAR control. Droop stability can require reduced sensitivity to voltage mismatch, especially when the voltage-sensitivity matrix is poorly conditioned.

  • A. Droop VAR Control: The GP-based local controller generalizes droop control by setting d_j = 1/c_j and α(t) = 1.Droop scales inverter VAR output according to instantaneous local voltage mismatch.
  • A. Droop VAR Control: Larger c_j represents a higher VAR-supply cost and produces a droop slope with smaller magnitude, reducing sensitivity to local voltage mismatch.The linear droop curve has negative slope −c_j^-1 and no deadband.
  • A. Droop VAR Control: Droop stability under D = C^-1 is equivalent to requiring C^-1 − X to be positive definite.This is the droop specialization of the general stability condition.
  • A. Droop VAR Control: Large eigenvalues of X require a very small droop slope, which reduces voltage-mismatch sensitivity and can slow algorithmic convergence.As networks grow or feeder lines lengthen, X is expected to become more poorly conditioned and droop control more likely to be unstable.

B. Scaled VAR Control

The paper considers alternative diagonal scaling to address droop stability concerns while preserving constrained optimality. Hessian-based scaling provides flexibility in VAR-penalty selection and is intended to improve conditioning and convergence speed.

  • B. Scaled VAR Control: Newton-style inverse-Hessian scaling is problematic with projections because its constrained fixed point generally need not be optimal.The design therefore sets α(t) = 1 while choosing a suitable diagonal matrix D.
  • B. Scaled VAR Control: Compared with droop control, the proposed scaled design can stabilize system dynamics for any matrix C.This gives greater flexibility in selecting the VAR-supply penalty.
  • B. Scaled VAR Control: Hessian-based diagonal scaling improves matrix conditioning and is expected to speed convergence without sacrificing the optimality condition.The paper identifies this improvement as a subject for numerical demonstration.

C. Delayed VAR Control

The delayed VAR scheme interprets gradient-projection updates as weighted averaging, with smaller α(t) improving stability but slowing convergence. Its stability condition depends explicitly on α(t), while adaptive rules can preserve GP convergence.

  • Delayed update: The delayed control update is a weighted average between the previous iterate and the projected gradient-projection result.This general GP form coincides with the delayed droop VAR design.
  • Delayed update: The delayed droop design addresses instability issues, but its earlier stability condition does not provide graph-based stepsize bounds like Proposition 3.The earlier condition requires complete distribution-system case information.
  • Convergence: GP convergence requires constant D and an adaptively selected α(t), such as through limited minimization or Armijo rules ensuring sufficient objective decrease.Numerical work also supports small constant choices of α(t).
  • Stability: With inactive projection, the delayed update has effective Jacobian α(t)H, so smaller α(t) makes stability more likely.This analysis applies when abundant VAR resources keep the projection inactive.
  • Stability: The sufficient stability condition is λ_Hmax < 2/α(t), and stability is increasingly likely as α(t) approaches zero.The analysis connects delayed VAR control with improved droop-control stability.

VI. NUMERICAL TESTS

The numerical section evaluates proposed local control methods on single- and three-phase feeders under static and dynamic loading and generation scenarios. Tests use actual OpenDSS power flows and measured bus voltage magnitudes for updates and comparisons.

  • Test design: Numerical tests assess local control methods on single- and three-phase feeder systems under static and dynamic loading and generation scenarios.Desired voltage magnitude is set to unit per unit at every bus, with V0 fixed at 1.
  • Implementation: All tests use OpenDSS to solve actual power flows rather than the approximate solution from the LinDistFlow-based model.This evaluates control behavior using the simulator’s actual power-flow solution.
  • Implementation: Actual bus voltage magnitudes, rather than LinDistFlow voltage estimates, drive VAR updates and numerical performance comparisons.The same actual-voltage basis is used for both control and evaluation.

A. Single-Phase 16-Bus Radial Feeder

The study first evaluates a 12 kV, 16-bus radial feeder with uniform line impedance, constant loads, and abundant bounded VAR resources. Proposed schemes use a lower VAR penalty than droop control.

  • Test system: The test system is a 12 kV radial feeder with 16 buses, represented by N = 15, and identical line impedance of (0.466+j0.733)Ω.The feeder provides the single-phase static test setting.
  • Operating conditions: Each bus has a constant load of (100 + j50)kVA and abundant VAR resources constrained to qg_j ∈ [−100, 100]kVA.The VAR limits apply uniformly across buses.
  • Control settings: The proposed scaled and delayed schemes use c_j = 0.2, whereas delayed droop control uses c_j = 0.5 with a no-deadband linear droop curve.Voltage limits are [0.95, 1.05] and VAR limits are [−100, 100]kVA.

1) Static scenario:

In the static 16-bus test, scaled and delayed local control converge rapidly, while droop oscillates unless delayed. Stepsize experiments expose a stability–convergence trade-off for both scaled and delayed schemes.

  • Static scenario: The centralized reference minimizes the stated objective with c_j = 0.2, while the unweighted benchmark replaces B with λ̄I.The plotted metric is the iterative voltage mismatch error norm ∥V−1∥.
  • Static scenario: Droop control fails to converge and oscillates between two operating points under the selected static-test settings.Using a larger penalty coefficient associated with a steeper droop slope can lead to instability.
  • Static scenario: 0.055 is the best voltage mismatch performance for local strategies under the surrogate VAR objective, versus 0.031 for the benchmark.The scaled and delayed methods converge very fast to the centralized solution.
  • Stepsize effects: Larger ϵ accelerates scaled-control convergence but can produce oscillations, revealing a trade-off between stability and convergence rate.The tested behavior is shown for different ϵ values.
  • Stepsize effects: For delayed control with ϵ = 0.3, smaller constant α(t) slows convergence, and the best delayed setting does not appear faster than scaled control.The tested α(t) values range from 0.01 to 0.9.
  • Computation: All local control schemes require around 1–2 microseconds per node per iteration in the reported MATLAB implementation.The timing result supports the fast-computation feature discussed in Remark 4.

2) Dynamic tests:

Dynamic tests use minute-resolution residential load and solar profiles to evaluate local VAR control during changing operating conditions. The proposed scaled scheme improves voltage mismatch performance over no VAR support in these scenarios.

  • Dynamic scenarios: The tests model PV-generation drops from cloud coverage and sudden load increases during appliance start-up.Load and PV profiles are updated at minute resolution from a real residential dataset.
  • Dynamic scenarios: Each bus represents 18 residential homes with 3 kW peak solar panels and inverter apparent-power limits 5% above peak capacity.Every home uses the same load and PV generation profiles.
  • Voltage behavior: Without VAR support, peak voltage occurs at noon, while evening demand produces under-voltage below 0.95 p.u. at the feeder end.Noon combines minimal load with peak solar generation; evening combines increasing demand with declining PV generation.
  • Control comparison: Additional VAR control outperforms no VAR support for daily voltage mismatch, with scaled and delayed droop schemes updated every 5 seconds.Load and PV generation remain constant within each minute during these updates.

B. Modified 16-Bus Meshed Network

The paper evaluates local VAR control on modified meshed and IEEE 123-bus three-phase feeders. Both local methods improve voltage support over no VAR support, while performance differences depend on PV penetration and operating conditions.

  • B. Modified 16-Bus Meshed Network: The modified 16-bus radial feeder adds lines connecting nodes 12–14 and 13–15 to test applicability in a meshed network.All other settings follow the static scenario, and voltage mismatch is compared across control schemes.
  • C. IEEE 123-Bus Test Feeder: IEEE 123-bus tests remove four three-phase voltage regulators and assign residential load and PV profiles to the feeder’s load buses.The feeder layout and PV locations are shown in Fig. 10.
  • C. IEEE 123-Bus Test Feeder: Both scaled and delayed droop control improve three-phase voltage support over no VAR support in the IEEE 123-bus case.The scaled method uses constant c_j = 0.01, while delayed droop uses time-varying slopes based on inverter VAR limits.
  • C. IEEE 123-Bus Test Feeder: The scaled scheme almost halves voltage mismatch during evening violations relative to no VAR support in the 123-bus case.Its advantage is more pronounced at higher voltage violation.
  • Framework validation: The framework formulates box-constrained voltage-mismatch minimization and solves it using gradient-projection control with local updates.The analytical results are corroborated by numerical tests on single- and three-phase systems using exact AC power flow.
  • VII. CONCLUSIONS AND FUTURE WORK: Future work will examine asynchronous control updates caused by insufficient coordination among buses.The paper also identifies interactions with other voltage-regulation devices as an ongoing research topic.
Loading 1510.07013v1…