PaperlayerView exact revision on arXiv

Source-linked AI summary

Synchronization and Transient Stability in Power Networks and Non-Uniform Kuramoto Oscillators

Florian Dorfler, Francesco Bullo

arXiv:0910.5673v4math.OCeess.SYmath-phmath.DS

TL;DR

The paper addresses the lack of concise synchronization and transient-stability conditions for network-reduced power systems tied to network structure and state. It uses singular perturbation to connect overdamped swing equations with non-uniform Kuramoto oscillators, then derives algebraic conditions relating synchronization and transient stability to system parameters and initial conditions.

  • Problem

    The paper addresses the open problem of obtaining explicit, concise synchronization conditions based on a power network’s topology, algebraic and spectral properties, parameters, and initial state.

  • Method

    Assuming strongly overdamped generators, the paper applies singular perturbation and combines power-network, Kuramoto-synchronization, transient-stability, and consensus techniques.

  • Results

    The analysis yields purely algebraic conditions under which a network-reduced power-system model synchronizes and achieves transient stability through its non-uniform Kuramoto representation.

  • Takeaways & Limitations

    The conditions relate synchronization and transient stability to underlying system parameters and initial conditions, while applying to non-uniform oscillator networks.

  • Takeaways & Limitations

    The reduction assumes each generator is strongly overdamped, attributed to local excitation control.

Abstract

from arXiv · show

Motivated by recent interest for multi-agent systems and smart power grid architectures, we discuss the synchronization problem for the network-reduced model of a power system with non-trivial transfer conductances. Our key insight is to exploit the relationship between the power network model and a first-order model of coupled oscillators. Assuming overdamped generators (possibly due to local excitation controllers), a singular perturbation analysis shows the equivalence between the classic swing equations and a non-uniform Kuramoto model. Here, non-uniform Kuramoto oscillators are characterized by multiple time constants, non-homogeneous coupling, and non-uniform phase shifts. Extending methods from transient stability, synchronization theory, and consensus protocols, we establish sufficient conditions for synchronization of non-uniform Kuramoto oscillators. These conditions reduce to and improve upon previously-available tests for the standard Kuramoto model. Combining our singular perturbation and Kuramoto analyses, we derive concise and purely algebraic conditions that relate synchronization and transient stability of a power network to the underlying system parameters and initial conditions.

I. INTRODUCTION

The paper connects power-network transient stability with synchronization of non-uniform Kuramoto oscillators, addressing the lack of concise conditions tied to network structure, parameters, and initial state. It derives algebraic synchronization and stability conditions by combining singular perturbation, oscillator synchronization, and consensus methods.

  • Motivation: Existing methods often use simplified models and lack simple formulas for checking whether a power system synchronizes for given parameters and state.The paper identifies explicit, concise conditions based on topological, algebraic, and spectral network properties as an open problem.
  • Motivation: The paper identifies an unbridged relationship among power-network synchronization, Kuramoto oscillators, and consensus protocols.Prior work had recognized similarities, but the gap had not been thoroughly analyzed.
  • Contributions: Singular perturbation reduces transient-stability analysis of swing equations with overdamped generators to synchronization of non-uniform Kuramoto oscillators.The resulting oscillators have multiple time constants, non-homogeneous coupling, and non-uniform phase shifts.
  • Contributions: The paper derives simple, purely algebraic conditions relating power-network synchronization and transient stability directly to network parameters and initial state.The conditions use assumptions that may be less restrictive than those of classic analysis methods.
  • Contributions: The synchronization analysis establishes phase cohesiveness, frequency synchronization, and phase synchronization for non-uniform Kuramoto oscillators.The results also apply beyond complete graphs and, in the uniform case, reduce to or improve upon known Kuramoto conditions.
  • Contributions: The resulting Kuramoto synchronization conditions also suffice for transient stability of the power network.Through topological equivalence, the conditions hold locally even when generators are not overdamped.

A. The Mathematical Model of a Power Network

The network-reduced power-system model represents generators through swing equations coupled by transfer admittances, with transfer conductances producing phase shifts and losses. Classical transient-stability analyses often simplify this model and typically yield numerical rather than concise algebraic synchronization conditions.

  • Power-system model: Each generator is characterized by internal voltage, rotor angle, mechanical input, inertia, and damping, with angles measured relative to a rotating frame.
  • Power-system model: Kron reduction eliminates passive network nodes and produces a reduced transfer-admittance matrix coupling generator dynamics.
  • Power-system model: Transfer-admittance magnitudes determine coupling weights, while phase shifts encode energy losses from transfer conductances; lossless networks have zero phase shifts.
  • Model assumptions: The classical model incorporates higher-order electrical and excitation-control effects into damping and is regarded as adequate during the first swing.
  • Classical analysis: Existing transient-stability methods commonly use relative or reference coordinates, uniform damping, special benchmark cases, or sufficiently small transfer conductances.
  • Classical analysis: For arbitrary lossy networks, prior methods lack quantified conductance bounds and explicit concise conditions relating synchronization to network state, parameters, and topology.

III. THE NON-UNIFORM KURAMOTO MODEL AND MAIN SYNCHRONIZATION RESULT

The paper connects swing dynamics to a non-uniform Kuramoto model through first-order approximation and singular perturbation analysis. It then gives algebraic synchronization conditions whose connectivity, loss, non-uniformity, and phase-cohesiveness terms determine transient-stability guarantees.

  • The non-uniform Kuramoto model: The non-uniform Kuramoto model generalizes classic Kuramoto oscillators with multiple time constants, non-homogeneous symmetric couplings, and non-uniform phase shifts.
  • The non-uniform Kuramoto model: For small inertia-to-damping ratios, the non-uniform Kuramoto model serves as a long-time first-order approximation of the second-order power-network model.
  • The non-uniform Kuramoto model: Both the reduced gradient system and the non-uniform Kuramoto model share equilibria and stability properties, while the latter directly exposes network structure for synchronization analysis.
  • Main synchronization result: The main theorem provides sufficient synchronization conditions, an attraction-region estimate, and an ultimate phase-cohesive set for the Kuramoto and power-network models.
  • Main synchronization result: The critical-to-minimal coupling gap acts as a robustness margin for admissible initial conditions and ultimate phase cohesiveness.
  • Main synchronization result: The scalar condition requires network connectivity to dominate oscillator non-uniformity, transfer losses, and insufficient phase cohesiveness.
  • Main synchronization result: The power-network synchronization condition holds up to an approximation error of order ϵ, and convergence is guaranteed only from almost all initial conditions because saddle points exist.
  • Main synchronization result: For the classic Kuramoto model, the sufficient condition improves previous results and can be tight; refinements use algebraic connectivity and yield synchronization-rate results in lossless networks.

C. Discussion of the Perturbation Assumption

The reduction relies on strongly overdamped generators so that fast frequency damping and slower phase dynamics separate. Under this assumption, singular perturbation and topological-equivalence arguments connect the power-network model to a non-uniform Kuramoto model and transfer synchronization results to local transient stability.

  • C. Discussion of the Perturbation Assumption: Strong overdamping separates fast frequency damping from slower phase dynamics, motivating a singular perturbation reduction.The paper compares this mechanism with overdamped harmonic oscillators and coupled pendula.
  • C. Discussion of the Perturbation Assumption: The perturbation parameter is ϵ = Mmax/Dmin, and simulations indicate that the approximation remains accurate even for ϵ ∈O(1).The discussion attributes this possibility to the topological equivalence among the power-network, first-order, and non-uniform Kuramoto models.
  • C. Discussion of the Perturbation Assumption: With excitation control, damping can increase to Di ∈[10, 35]/(2πf0), making ϵ ∈O(0.1) and the approximation accurate.Without such control, mechanical damping is described as poor, with Di ∈[1, 3]/(2πf0).
  • C. Discussion of the Perturbation Assumption: Topological equivalence makes the synchronization condition locally applicable to the power network independently of ϵ > 0, while ϵ bounds transient approximation errors.The condition guarantees exponential stability of the non-uniform Kuramoto dynamics and local exponential stability of the power network in relative coordinates.
  • C. Discussion of the Perturbation Assumption: For sufficiently small ϵ, the singular perturbation solution exists uniquely and its phase and frequency approximation errors are O(ϵ) uniformly in time.The approximation compares δ(t, ϵ) with the reduced trajectory and ˙θ(t, ϵ) with slow and fast reduced components.

V. SYNCHRONIZATION OF NON-UNIFORM KURAMOTO OSCILLATORS

The non-uniform Kuramoto model generalizes standard oscillators through multiple time constants, non-homogeneous coupling, and phase shifts. These features create lossy and asymmetric interactions that complicate synchronization analysis.

  • V. SYNCHRONIZATION OF NON-UNIFORM KURAMOTO OSCILLATORS: Multiple time constants, non-homogeneous coupling, and non-uniform phase shifts distinguish the model from the standard Kuramoto system.Dividing the dynamics by Di exposes the roles of time constants and phase shifts.
  • V. SYNCHRONIZATION OF NON-UNIFORM KURAMOTO OSCILLATORS: Phase shifts introduce lossy coupling through (Pij/Di) sin(ϕij) × cos(θi −θj), which can inhibit synchronization.The loss term is identified as one of the principal analytical difficulties.
  • V. SYNCHRONIZATION OF NON-UNIFORM KURAMOTO OSCILLATORS: Different time constants produce asymmetric coupling because an oscillator pair is weighted by Pij/Di in one direction and Pij/Dj in the other.The section studies this generality even though the power-network-derived matrix P is complete and symmetric.

A. Frequency Synchronization of Phase-Cohesive Oscillators

Under positively invariant phase cohesiveness, the non-uniform Kuramoto model achieves exponential frequency synchronization when its coupling graph has a globally reachable node. With zero phase shifts and symmetric coupling, the limiting frequency has a weighted-average formula and an explicit rate.

  • A. Frequency Synchronization of Phase-Cohesive Oscillators: A globally reachable node in the graph induced by P suffices for exponential frequency synchronization of non-uniform Kuramoto oscillators.The proof treats frequency dynamics as a time-varying consensus protocol with non-degenerate weights.
  • A. Frequency Synchronization of Phase-Cohesive Oscillators: Phase cohesiveness is assumed through positive invariance of bounded phase differences ¯∆(γ), ensuring the synchronization result applies for every θ(0) ∈¯∆(γ).The admissible range includes γ ∈[0, π/2−ϕmax[.
  • A. Frequency Synchronization of Phase-Cohesive Oscillators: Under zero phase shifts and symmetric coupling, the synchronized frequency is ˙θ∞= Ω:= P i ωi/ P i Di.The same specialization provides an explicit exponential synchronization rate.
  • A. Frequency Synchronization of Phase-Cohesive Oscillators: The rate λfe combines algebraic connectivity, the slowest time constant, phase cohesiveness, and damping-vector alignment.Its factors are λ2(L(Pij)), 1/Dmax, cos(γ), and cos(∠(D1, 1))2.
  • A. Frequency Synchronization of Phase-Cohesive Oscillators: For nonzero phase shifts and symmetric coupling, the synchronization frequency can be smaller than the zero-shift value Ω.The paper identifies a phase-shift effect on the limiting frequency through small-signal analysis.
  • A. Frequency Synchronization of Phase-Cohesive Oscillators: If natural frequencies are non-identical or do not converge exponentially to identical values, frequency synchronization cannot be achieved under the stated proof.Smoothly time-varying natural frequencies also add a ˙ω(t) term to the frequency consensus dynamics.

B. Phase Cohesiveness

The paper develops sufficient synchronization conditions for non-uniform Kuramoto oscillators by enforcing phase cohesiveness through contraction and positive-invariance arguments. For classic Kuramoto oscillators, the resulting bounds improve prior sufficient tests and can become necessary and sufficient under limited frequency information.

  • Synchronization condition I: The contraction approach shows that phase cohesiveness is preserved when the coupling dominates oscillator non-uniformity and adverse lossy-coupling effects.The relevant phase-shift effects arise through non-symmetric coupling and the phase shifts ϕij.
  • Synchronization condition I: Theorem V.3 establishes sufficient synchronization conditions for non-uniform Kuramoto oscillators with complete coupling based on positive invariance of phase-cohesive sets.The analysis keeps all phases within a rotating arc whose maximal length does not increase.
  • Reduction to classic Kuramoto oscillators: For classic Kuramoto oscillators, condition (27) gives synchronization when K > Kcritical and initial phases lie in ∆(γmax), followed by ultimate cohesiveness in ¯∆(γmin).Here γmax and γmin are determined by sin(γmin) = sin(γmax) = Kcritical/K.
  • Reduction to classic Kuramoto oscillators: Condition (27) is reported as the tightest explicit sufficient coupling-gain condition presented for the classic Kuramoto model.The bound is close to the necessary condition K > Kcritical n/(2(n−1)).
  • Reduction to classic Kuramoto oscillators: The bound improves several prior sufficient conditions and is necessary and sufficient when the natural frequencies are known only to lie in [ωmin, ωmax].It is also necessary and sufficient for a bimodal distribution with frequencies in {ωmax, ωmin}.
  • Synchronization condition I: Theorem V.3 extends beyond complete graphs because its conditions require only connectivity of the graph induced by P = P^T.The theorem’s stated result is based on a worst-case bound.

2. Unfortunately, in the case of non-uniform rates

For non-uniform rates, the paper uses a weighted Lyapunov function and algebraic connectivity to derive a second synchronization condition. The result bounds initial and ultimate phase cohesiveness while incorporating frequency non-uniformity, phase shifts, coupling connectivity, and time constants.

  • Synchronization condition II: Theorem V.5 derives synchronization condition II for a connected coupling graph using a Lyapunov analysis of the weighted phase-difference function W.The derivative of W is bounded using algebraic connectivity of the lossless coupling.
  • Synchronization condition II: The condition compares a critical non-uniformity term with the algebraic connectivity of the lossless coupling and the maximum phase shift.The physical interpretation includes frequency non-uniformity, lossless-coupling connectivity, phase cohesiveness, and non-uniform time constants.
  • Synchronization condition II: For admissible initial phase differences, trajectories remain in a positively invariant set and ultimately enter a smaller phase-cohesive region.The initial and ultimate regions are characterized by γmax and γmin determined through the stated equations.
  • Synchronization condition II: The theorem also establishes exponential frequency synchronization to a common frequency within the initial frequency range.When ϕmax = 0, the limiting frequency is Ω and the exponential rate is no worse than λfe.
  • Synchronization condition II: The resulting gap provides a practical stability result determining initial and ultimate phase cohesiveness and can extend to non-reduced power-network models.The paper identifies this as a practical stability result rather than an exact global characterization.

C. Phase Synchronization

Under a globally reachable graph, zero phase shifts, and identical normalized natural frequencies, the paper proves exponential phase synchronization for the non-uniform Kuramoto model. Symmetric coupling further identifies the limiting trajectory as a weighted mean angle.

  • Phase synchronization: For identical natural frequencies and zero phase shifts, the practical stability results imply phase synchronization as γmin decreases to zero.This connects the phase-synchronization result with the earlier practical stability theorems.
  • Phase synchronization: Theorem V.10 assumes a globally reachable graph, ϕmax = 0, and identical normalized natural frequencies ωi/Di = ¯ω.These assumptions define the setting for the phase-synchronization result.
  • Phase synchronization: The phases synchronize exponentially to a trajectory contained in the initial angular interval and rotating with ¯ω.The limiting trajectory is θ∞(t) ∈ [θmin(0), θmax(0)] + ¯ωt.
  • Phase synchronization: With symmetric coupling, the phases synchronize exponentially to a weighted mean angle.The paper contrasts this with the general globally reachable case, where only the limiting trajectory’s interval is specified.
  • Phase synchronization: The proof reformulates the phase dynamics as a time-varying consensus protocol with strictly positive, state-dependent coupling weights.For symmetric coupling, the weights are wij(t) = Pij sinc(θi(t)−θj(t)) with multiple rates Di.

VI. SIMULATION RESULTS

Simulations of a ten-generator power network compare second-order power-network dynamics with the corresponding first-order non-uniform Kuramoto model under disturbances and different damping regimes. The trajectories synchronize, while singular-perturbation errors converge quickly.

  • Simulation setup: The simulation uses a ten-generator power network with clustered initial angles, random initial frequencies, heterogeneous parameters, and a transient high-frequency disturbance.The disturbance is applied at generator n−1, and the parameters are chosen within the stated uniform ranges.
  • Figure 3(a): For the weakly damped case with ϵ = 0.58 s, the Kuramoto angles synchronize very fast from non-synchronized initial conditions within 0.05 s.The disturbance around t = 2 s does not severely affect the synchronization dynamics.
  • Figure 3(a): The quasi-steady-state frequencies show the same synchronization finding, although the disturbance acts directly on the first-order frequency representation.The disturbance is shown at angle n−1 as the yellow curve.
  • Figure 3(a): The underdamped power-network trajectories synchronize with second-order dynamics, and the disturbance affects them less than the first-order Kuramoto dynamics.After initial and mid-simulation transients, δ(t)−¯δ(t) and θ(t)−h(¯δ(t)) become small and converge.
  • Figure 3(b): In the strongly damped case with ϵ = 0.18 s, singular-perturbation errors remain smaller during transients and converge faster than in the weakly damped case.The Kuramoto and quasi-steady-state dynamics have slower time constants than the strongly damped power-network dynamics.

VII. CONCLUSIONS

The paper connects transient stability in power networks to synchronization of non-uniform Kuramoto oscillators through singular perturbation analysis. It derives algebraic conditions based on network parameters and initial phase differences, while acknowledging practical limitations.

  • VII. CONCLUSIONS: The resulting sufficient conditions are purely algebraic and relate transient stability to network parameters and initial phase differences.They are intended as concise tests rather than numerical estimates of attraction regions or critical clearing times.
  • VII. CONCLUSIONS: The analysis assumes each generator is highly overdamped, possibly because of local excitation control.This assumption underpins the reduction from swing-equation dynamics to the first-order oscillator model.
  • VII. CONCLUSIONS: The approach combines techniques from transient stability, Kuramoto synchronization, and consensus protocols to study generalized coupled oscillators.The paper presents the relationship among these areas as a mathematical and methodological connection.
  • VII. CONCLUSIONS: A singular perturbation analysis reduces transient stability of an overdamped, network-reduced power model to synchronization analysis of non-uniform Kuramoto oscillators.The oscillator model incorporates heterogeneous time constants, coupling, and phase shifts.
  • VII. CONCLUSIONS: The authors state that the conditions are not yet competitive with sophisticated numerical algorithms used by the power systems community.They identify tighter conditions, more accurate attraction-region characterization, and more realistic models as needed for real systems.

VIII. APPENDIX: ALTERNATIVE SYNCHRONIZATION CONDITIONS

The appendix develops alternative bounds for synchronization conditions and the associated attraction and phase-cohesion estimates. It compares tighter pairwise analyses with simpler, more conservative theorem statements.

  • VIII. APPENDIX: ALTERNATIVE SYNCHRONIZATION CONDITIONS: Alternative bounding methods modify the synchronization condition, the region-of-attraction estimate, and the ultimate phase-cohesive set.The appendix examines how proof choices affect this triplet of results.
  • VIII. APPENDIX: ALTERNATIVE SYNCHRONIZATION CONDITIONS: Pairwise bounding yields n(n −1)/2 synchronization conditions, trading tighter bounds for increased complexity.The worst multiplicative gap across pairs determines the attraction-region and phase-cohesion estimates.
  • VIII. APPENDIX: ALTERNATIVE SYNCHRONIZATION CONDITIONS: For complete graphs, Theorems VIII.1 and VIII.2 give synchronization conditions based on critical coupling and phase-arc bounds.The conditions use γmin and γmax defined through equations associated with the bounding inequalities.
  • VIII. APPENDIX: ALTERNATIVE SYNCHRONIZATION CONDITIONS: Concavity-based arguments tighten the bounding of shifted sine-function sums and preserve non-increasing Lyapunov behavior over specified arc lengths.The resulting arc-length interval is bounded by γmin and γmax satisfying the relevant inequality.
  • VIII. APPENDIX: ALTERNATIVE SYNCHRONIZATION CONDITIONS: The appendix also presents a simpler theorem statement, but explicitly characterizes the resulting synchronization condition as very conservative.This alternative avoids some pairwise complexity while weakening the bound.

C. Adding and Subtracting the Lossless Coupling

This section simplifies the synchronization analysis by adding and subtracting lossless coupling terms. The resulting scalar condition guarantees non-increasing angle-arc length under a critical-coupling requirement.

  • C. Adding and Subtracting the Lossless Coupling: Adding and subtracting the lossless coupling produces a simple scalar synchronization condition for the non-uniform Kuramoto model.The condition is derived by directly bounding the derivative of the Lyapunov function.
  • C. Adding and Subtracting the Lossless Coupling: The scalar condition applies when the graph is complete and every pair’s minimal coupling exceeds a critical value.The theorem states the requirement pairwise over all oscillator indices.
  • C. Adding and Subtracting the Lossless Coupling: If the critical-coupling inequality holds, the length of the arc formed by the oscillator angles is non-increasing within Δ(γ).This provides the section’s direct dynamical consequence for phase cohesion.
  • C. Adding and Subtracting the Lossless Coupling: Under the condition, γmin is given by arcsin(cos(ϕmax)Pcritical/Pmin), while γmax equals π/2 −ϕmax.These values define the relevant phase-cohesion bounds.
Loading 0910.5673v4…