Source-linked AI summary
Design and Stability of Load-Side Primary Frequency Control in Power Systems
Changhong Zhao, Ufuk Topcu, Na Li, Steven Low
TL;DR
The paper addresses the limited analytic treatment of load-side primary frequency control in multimachine networks, including the costs incurred by participating loads. It formulates optimal load control and shows that network dynamics with frequency-based control implement a distributed primal-dual algorithm. The resulting system is globally asymptotically stable and achieves globally optimal load control while rebalancing power and resynchronizing frequencies after disturbances.
Problem
Existing analytic models did not adequately relate load behavior to multimachine network dynamics or account for the disutility of load participation in primary frequency control.
Method
The paper formulates optimal load control and identifies swing dynamics, branch flows, and frequency-based load control as a distributed primal-dual algorithm for its dual.
Results
The resulting trajectories converge to the unique optimal load-control and frequency-deviation vectors, while branch flows are optimal dual variables.
Takeaways & Limitations
Local frequency deviations enable decentralized load decisions that are globally optimal, and the controlled network can rebalance power and resynchronize bus frequencies after disturbances.
Abstract
from arXiv · showhide
We present a systematic method to design ubiquitous continuous fast-acting distributed load control for primary frequency regulation in power networks, by formulating an optimal load control (OLC) problem where the objective is to minimize the aggregate cost of tracking an operating point subject to power balance over the network. We prove that the swing dynamics and the branch power flows, coupled with frequency-based load control, serve as a distributed primal-dual algorithm to solve OLC. We establish the global asymptotic stability of a multimachine network under such type of load-side primary frequency control. These results imply that the local frequency deviations at each bus convey exactly the right information about the global power imbalance for the loads to make individual decisions that turn out to be globally optimal. Simulations confirm that the proposed algorithm can rebalance power and resynchronize bus frequencies after a disturbance with significantly improved transient performance.
I. INTRODUCTION
The paper addresses the limited analytic understanding of distributed load participation in multimachine frequency control by formulating an optimization-based design. It shows that frequency-based load control can achieve globally optimal rebalancing and improved transient performance.
- I. INTRODUCTION: Prior work included simulations and field trials, but lacked sufficient analytic study of load behavior and multimachine network dynamics.Classical generator-focused models also omitted the cost or disutility incurred by participating loads.
- I. INTRODUCTION: The paper formulates optimal load control to design continuous, fast-acting, distributed load participation in primary frequency regulation.The approach accounts for load consumption patterns and network-wide power imbalance.
- I. INTRODUCTION: Swing dynamics, branch flows, and frequency-based load control form a distributed primal-dual algorithm for solving the dual of the optimal load control problem.Frequency deviations and branch-flow deviations serve as optimization variables associated with imbalance and synchronization.
- I. INTRODUCTION: Local frequency deviations provide sufficient information for loads to make decentralized decisions that are globally optimal, without explicit communication.The paper identifies this as a consequence of the network-wide optimization formulation.
- I. INTRODUCTION: Simulations of the IEEE 68-bus test system show that the proposed mechanism can rebalance power and resynchronize bus frequencies with significantly improved transient performance.The results are presented as confirmation that frequency-adaptive loads can perform the corresponding primary-control functions.
II. NETWORK MODEL
The paper models a connected transmission network with generator and load buses, linearized swing or algebraic power-balance dynamics, and dynamic branch flows. The model captures transient frequency differences and network rebalancing at the timescale of seconds.
- II. NETWORK MODEL: The transmission network is represented as a connected directed graph with buses and lossless lines characterized by reactances.Voltage magnitudes are fixed, and reactive power injections and flows are ignored.
- II. NETWORK MODEL: The model assumes small frequency deviations and phase-angle differences and focuses on constant step disturbances at a seconds-scale timescale.It excludes turbine-governor action and secondary frequency control from the modeled response.
- II. NETWORK MODEL: Generator-bus dynamics combine inertia, mechanical power injection, electrical power export, damping, controllable loads, and branch power flows.The electrical export includes local loads and net power exchanged with the rest of the network.
- II. NETWORK MODEL: Loads are separated into frequency-sensitive, frequency-insensitive controllable, and uncontrollable categories.Frequency-sensitive consumption varies linearly with frequency deviation, while controllable loads are the subject of the designed control laws.
- II. NETWORK MODEL: Load buses without generators satisfy algebraic power-balance equations, while branch-flow deviations evolve according to linearized frequency differences.The resulting dynamic model is specified by the generator, load-bus, and branch-flow equations.
- II. NETWORK MODEL: Different buses can have different local frequencies during transients before converging to a common equilibrium frequency.This behavior motivates distinguishing local frequency deviations in the multimachine model.
III. DESIGN AND STABILITY OF PRIMARY FREQUENCY
The paper designs load-side primary frequency control by posing power rebalancing after a disturbance as an optimal load control problem. It derives the feedback controller from that optimization objective rather than specifying it independently.
- III. DESIGN AND STABILITY OF PRIMARY FREQUENCY: A constant disturbance injected at network buses requires controllable loads to adjust consumption while minimizing aggregate disutility.The paper considers state-feedback control but instead derives the controller from the optimal load control formulation.
- III. DESIGN AND STABILITY OF PRIMARY FREQUENCY: The paper first formulates the optimal load control objective and then derives a frequency-based feedback controller as a distributed algorithm solving it.The formulation connects the desired load-control outcome to the network dynamics.
- III. DESIGN AND STABILITY OF PRIMARY FREQUENCY: The design and stability results are proved in the paper’s convergence-analysis section.The section introduces the optimization-based design before the subsequent proofs.
A. Optimal load control
Optimal load control minimizes controllable-load disutility together with a frequency-deviation cost, subject to network-wide power rebalance. Strict convexity and feasibility provide the stated optimization conditions.
- A. Optimal load control: The OLC objective combines controllable-load disutility with a squared frequency-deviation cost weighted by relative damping.The frequency-sensitive load cost is proportional to the squared frequency deviation.
- A. Optimal load control: The optimization minimizes total cost over controllable and frequency-sensitive load changes while balancing generation and load across the entire network.The formulation permits controllable-load bounds on each bus.
- A. Optimal load control: OLC requires only network-wide balance, not generation-load balance at every individual bus.Additional local or control-area constraints may be imposed when desired for economic or regulatory reasons.
- A. Optimal load control: The cost functions are assumed feasible, strictly convex, and twice continuously differentiable.Their selection can reflect physical load characteristics and user comfort levels.
B. Main results
The paper formulates a distributed dual problem and shows that network dynamics with frequency-based load control implement a primal-dual algorithm whose trajectories converge to optimal operating points.
- Distributed optimal load control: The distributed dual problem assigns each bus a local variable ν_j, constrained to equal neighboring variables across every network edge.This replaces the nonseparable scalar dual formulation with coordinated local variables.
- Distributed optimal load control: Solving DOLC recovers the unique optimal controllable-load vector of OLC from the unique dual optimum ν∗.The DOLC objective is strictly concave, supporting uniqueness of the dual optimum.
- Primal-dual interpretation: The Lagrangian assigns ν to primal variables and π to dual variables measuring the cost of unsynchronized bus variables.The resulting partial primal-dual algorithm evolves these variables over time.
- Primal-dual interpretation: Identifying ν with frequency deviations ω and π with branch flows P makes the network dynamics equivalent to the DOLC primal-dual algorithm.The paper also notes consistency between the corresponding physical units.
- Convergence and optimality: Every trajectory from any initial state converges to a limit with unique optimal load control, unique optimal frequency deviations, and optimal branch flows.The convergence theorem applies to controllable loads, frequency-sensitive loads, frequencies, and branch flows.
C. Implications
The proposed load-side control supports decentralized primary frequency regulation by using local frequency deviations to achieve globally optimal load decisions and stable network behavior.
- Load-side primary control: Frequency-adaptive loads can rebalance power and resynchronize frequencies after disturbances, while Theorem 1 establishes global asymptotic stability of the multimachine network.The control explicitly optimizes aggregate disutility across heterogeneous loads.
- Complete decentralization: Local frequency deviations convey the global power imbalance needed for loads to make globally optimal decisions without explicit communication among buses.This provides a completely decentralized solution.
- Equilibrium frequency: At optimality, bus frequencies synchronize to ω∗, which can remain nonzero relative to the pre-disturbance nominal frequency.Isochronous generators or automatic generation control are needed to restore nominal frequency, typically through integral action.
- Frequency and branch flows: Frequency deviations act as Lagrange multipliers measuring power-imbalance cost, while branch-flow deviations measure the cost of frequency asynchronism.These interpretations connect physical network variables to the optimization problems.
IV. CONVERGENCE ANALYSIS
The convergence analysis identifies shared optimal and equilibrium sets, then uses a Lyapunov argument to establish convergence of the network trajectories.
- Proof structure: The proof begins by relating optimal points of DOLC and its dual to equilibrium points of the network dynamics.These sets are nonempty and identical.
- Proof structure: For tree networks, the shared optimal-equilibrium set is a singleton with a unique equilibrium point; mesh networks instead have infinitely many equilibria sharing the same frequency.Mesh equilibria differ in branch flows.
- Lyapunov convergence: A Lyapunov argument shows every trajectory approaches a nonempty compact subset of the shared optimal-equilibrium set.For mesh networks, an additional argument establishes convergence to a point rather than oscillation around that subset.
- Network representation: The analysis focuses on frequency deviations separated into generator- and load-bus components and uses the network incidence matrix partitioned accordingly.The decomposition supports the subsequent convergence analysis.
G CGP
The paper establishes that network equilibria coincide with primal-dual optima of DOLC and analyzes their convergence using a Lyapunov argument. Network topology determines whether the equilibrium is unique or forms a continuum, while trajectories still converge to an optimal equilibrium.
- Optimality and equilibrium: Primal-dual optimal points of DOLC and its dual are exactly the equilibrium points of the network dynamics, with a unique optimal frequency.At least one such point exists, and the frequency component is unique across all optimal points.
- Stability analysis: A Lyapunov function proves that every trajectory approaches a nonempty, compact subset Z+ of the equilibrium set Z*.The derivative of the Lyapunov function is nonpositive along trajectories.
- Network topology: Tree networks have a singleton equilibrium set, whereas mesh networks have uncountably many equilibria sharing the same frequency but differing in branch flows.For trees, the reduced incidence matrix is square and invertible; for meshes, its nontrivial null space produces multiple branch-flow equilibria.
- Stability analysis: The Lyapunov analysis uses a function of generator frequencies and branch flows, with load-bus frequencies handled through the algebraic network equation.This construction also shows convergence to a primal-dual optimal point rather than oscillation around an equilibrium subspace.
- Convergence: Despite multiple equilibria in mesh networks, practical trajectories satisfying the initial branch-flow constraint converge to one unique equilibrium point.The limiting branch flows depend on the trajectory’s initial state, while the frequency converges to the unique optimal frequency.
V. CASE STUDIES
Simulations on the IEEE 68-bus test system evaluate OLC under detailed nonlinear network and load models. OLC improves frequency and voltage regulation, with control cost converging to the theoretical minimum.
- Simulation setup: The simulation uses a detailed IEEE 68-bus model with nonlinear power flows, generator dynamics, stabilizers, and non-zero line resistances.The model includes two-axis subtransient generator models, IEEE DC1 exciters, classical PSS models, AC power flows, and realistic line resistance.
- Simulation setup: Thirty load buses perform OLC, with controllable-load size varied and load control updated every 250 ms.The test system has 35 load buses and 18.23 GW of total real load; three additional 1 pu steps create the disturbance.
- Frequency response: Adding OLC decreases frequency overshoot and settling time and reduces steady-state frequency error, whether PSS is enabled or disabled.With PSS enabled, the total controllable-load size is 1.5 pu in the compared cases.
- Optimality: The OLC cost trajectory converges to the minimum cost for the imposed change in mechanical power.The trajectory is evaluated from controllable and frequency-sensitive load trajectories and compared with the minimum-cost value.
VI. CONCLUSION
The paper designs distributed load-side primary frequency control by formulating OLC and interpreting network dynamics as a distributed optimization algorithm. It proves convergence to a unique optimum and confirms improved rebalancing and synchronization in simulation.
- Conclusion: OLC minimizes aggregate load-control cost subject to power balance across the network.The method targets ubiquitous, continuous, fast-acting, distributed load control for primary frequency regulation.
- Conclusion: Generator swing dynamics, branch power flows, and frequency-based load control form a distributed primal-dual algorithm for solving OLC.The cost trajectory in simulation also converges to the minimum-cost value.
- Conclusion: Despite multiple equilibrium points and nonunique branch power flows, the system converges to a unique optimal point.The conclusion states this result as a proved property of the proposed control framework.
- Conclusion: IEEE 68-bus simulations confirm power rebalancing and bus-frequency resynchronization with significantly improved transient performance.The simulation evidence complements the stability and optimality results.
APPENDIX A SIMULATION SHOWING FEATURE OF MODEL
The appendix validates a modeling assumption: local bus frequencies can differ substantially during transients before converging to a common equilibrium. It also establishes optimization properties supporting uniqueness and duality.
- Model feature: Different buses can maintain distinct local frequencies during transients for a duration comparable to their convergence time to equilibrium.The assumption is examined using the IEEE 68-bus system without OLC.
- Model feature: Frequencies within coherent groups are nearly identical, while frequencies across different groups differ substantially during transients.The 68 buses are divided into four groups for visualization.
- Model feature: The time for different bus frequencies to converge to a common frequency is on the same order as the time to reach equilibrium.This supports modeling bus-specific transient frequencies rather than assuming instantaneous synchronization.
- Optimization properties: The OLC objective has a unique minimizer because it is continuous and strictly convex on a compact convex feasible subset.The objective is lower bounded, and every optimum must lie in that subset.
- Optimization properties: The affine OLC constraint gives zero duality gap and an attained dual optimum, while strict concavity ensures uniqueness of the dual variable.These properties connect the primal optimum to the network's distributed dynamics.
4) Proof of Lemma 4:
The proof characterizes the invariant set of the Lyapunov analysis and shows that its trajectories satisfy the conditions defining the optimal equilibrium. Concavity and monotonicity force synchronized generator and load frequencies at convergence.
- Optimality conditions: The proof therefore identifies the zero-derivative invariant trajectories with the optimal equilibrium conditions.The argument combines the generator-frequency condition with the load-frequency condition.
- Optimality conditions: Strict concavity in generator frequencies forces ωG = ω∗1G whenever the Lyapunov derivative is zero.The remaining condition concerns the load-frequency and branch-flow variables.
- Optimality conditions: Monotonicity of the load-control functions makes the relevant summation nonnegative and zero exactly when load frequencies equal the optimal common value.This establishes equivalence between the invariant-set condition and the optimality condition.
- Invariant-set argument: LaSalle’s invariance principle places the positive limit set inside the equilibrium set and reduces the proof to showing it lies in the optimal set.The limit set is nonempty, compact, and invariant.
- Invariant-set argument: In the invariant set, generator frequencies remain equal to the optimal common frequency, so their derivatives vanish.Invariance and the equilibrium characterization imply ωG(t) = ω∗1G.