PaperlayerView exact revision on arXiv

Source-linked AI summary

Distributed reactive power feedback control for voltage regulation and loss minimization

Saverio Bolognani, Guido Cavraro, Ruggero Carli, Sandro Zampieri

arXiv:1303.7173v2math.OC

TL;DR

The paper addresses reactive-power compensation for minimizing distribution losses and regulating voltages when grid information and measurements are incomplete. It uses communicating microgenerator agents with voltage sensing and reactive-power actuation in a duality-based feedback control law. Analytical results establish convergence for synchronous and asynchronous algorithms, while simulations illustrate performance and robustness.

  • Problem

    The problem is to minimize distribution losses and regulate voltages despite unmonitored loads, unpredictable generators, and partially known grid parameters or topology.

  • Method

    Microgenerator agents measure voltages, communicate with neighbors, and adjust reactive-power injections using a duality-based feedback control law for constrained ORPF.

  • Results

    Convergence to the optimal primal solution is proved for the synchronous algorithm, and analytical convergence results are also provided for the asynchronous version within restricted constraint forms.

  • Takeaways & Limitations

    The feedback strategy infers unmeasured disturbances from voltage measurements and achieves practically the same power-loss performance as a centralized solver in simulation.

Abstract

from arXiv · show

We consider the problem of exploiting the microgenerators dispersed in the power distribution network in order to provide distributed reactive power compensation for power losses minimization and voltage regulation. In the proposed strategy, microgenerators are smart agents that can measure their phasorial voltage, share these data with the other agents on a cyber layer, and adjust the amount of reactive power injected into the grid, according to a feedback control law that descends from duality-based methods applied to the optimal reactive power flow problem. Convergence to the configuration of minimum losses and feasible voltages is proved analytically for both a synchronous and an asynchronous version of the algorithm, where agents update their state independently one from the other. Simulations are provided in order to illustrate the performance and the robustness of the algorithm, and the innovative feedback nature of such strategy is discussed.

I. INTRODUCTION

The paper targets distributed reactive-power compensation for loss minimization and voltage regulation under limited monitoring, uncertain loads, generators, parameters, and topology. It models the grid as a cyber-physical system and derives synchronous and asynchronous feedback algorithms for constrained ORPF.

  • Reactive-power compensation is formulated to minimize distribution losses while keeping voltage magnitudes and generator reactive-power injections within prescribed intervals.
  • Centralized ORPF methods assume accurate grid models, complete bus monitoring, and advance knowledge of load and generator behavior.
  • Unmonitored loads, unpredictable renewable generation, and partially known parameters or topology motivate distributed, plug-and-play control.
  • Existing local strategies lack coordination, while other distributed methods may still require measurements at every grid bus.
  • Convergence analysis assumes homogeneous line X/R ratios, whose variability is investigated through simulations for robustness.
  • The proposed cyber-physical model represents power infrastructure physically and equips dispersed intelligent agents with sensing, communication, computation, and actuation capabilities.

B. Cyber layer

The cyber layer represents microgenerators and the PCC as communicating agents with sensing capabilities; microgenerator agents additionally control reactive-power setpoints. Its neighbor structure supports distributed computation and corresponds to the sparsity pattern of matrix G.

  • Agent capabilities: Microgenerator and PCC nodes each correspond to cyber-layer agents equipped with computational and phasor-measurement capabilities.Microgenerator agents can measure voltage amplitude and angle, while microgenerator agents also actuate reactive-power injection.
  • Communication architecture: Agents communicate through a communication channel, potentially using the power lines through power-line communication.
  • Neighbor definition: Each agent knows and communicates with its cyber-layer neighbors, whose connecting path contains no other agent.The architecture can be constructed distributively and reconfigured when new agents connect to the grid.
  • Grid model: The approximate voltage–power relation is used to derive the distributed control strategy for the optimal reactive power flow problem.The approximation is accurate when nominal voltage is large and inverter or load currents are relatively small.
  • Matrix structure: Matrix G is symmetric and has nonzero entry G_hk only when k is a neighbor of h in the cyber layer.Its elements can be estimated by agents using local knowledge of grid parameters.

V. OPTIMAL REACTIVE POWER FLOW PROBLEM

The paper formulates optimal reactive power flow as choosing microgenerator reactive-power setpoints to reduce distribution losses while maintaining voltage and injection constraints. It then targets a distributed controller using only neighbor communication and measured PCC voltage, despite unmeasured network disturbances.

  • Problem formulation: The optimization chooses microgenerator reactive-power setpoints q_h to minimize distribution losses while keeping voltage magnitudes and injections within prescribed intervals.The formulation can be extended to heterogeneous generator-specific reactive-power bounds.
  • Objective: Distribution losses are represented as a quadratic function of voltage drops on the power lines.
  • Problem formulation: The nonlinear grid equations implicitly determine voltages u from the reactive-power decision variables q_G.
  • Control formulation: The control design treats q_G as inputs, PCC voltage as measured output, and p_L, q_L, and p_G as unmeasured disturbances.
  • Distributed control: The paper seeks a distributed ORPF algorithm in which each microgenerator communicates only with agents in its cyber-layer neighbor set N(h).

VI. A SYNCHRONOUS ALGORITHM BASED ON DUAL

The synchronous controller applies dual-ascent updates to voltage and power constraints, then computes distributed reactive-power setpoints from local and neighboring information. Its feedback loop approximates Lagrangian minimization and uses real-time plant measurements to approach the optimal configuration.

  • VI. A SYNCHRONOUS ALGORITHM BASED ON DUAL: Dual decomposition converts the ORPF formulation into a distributed feedback control law for reactive-power compensation.The method uses an approximate explicit power-flow solution to derive dual-ascent updates implementable by grid agents.
  • VI. A SYNCHRONOUS ALGORITHM BASED ON DUAL: Each agent updates dual variables from local voltage and power-constraint violations using projected gradient ascent.Voltage multipliers use measured squared voltage, while power multipliers use reactive-power bounds and positive-orthant projection.
  • VI. A SYNCHRONOUS ALGORITHM BASED ON DUAL: Reactive-power updates use each agent’s new voltage multipliers and neighboring power multipliers, requiring communication only with cyber-layer neighbors.Agents synchronously measure voltage, exchange updated multiplier values, and update injected reactive power.
  • VI. A SYNCHRONOUS ALGORITHM BASED ON DUAL: The reactive-power update minimizes the Lagrangian over primal variables up to a term that vanishes for large nominal voltage U_N.This establishes the connection between the implementable update and the primal step of the dual method.
  • VI. A SYNCHRONOUS ALGORITHM BASED ON DUAL: The controller requires actuation at every iteration so subsequent voltage measurements reflect the plant’s new state and reveal hidden load-demand information.The paper identifies this real-time cyber–physical interaction as the fundamental feedback feature of the approach.

VII. ASYNCHRONOUS ALGORITHM

The asynchronous algorithm removes system-wide timing coordination by allowing each microgenerator agent to update independently using information gathered from its neighbors.

  • VII. ASYNCHRONOUS ALGORITHM: Agents independently update their reactive power and auxiliary dual variables based on neighboring information.The asynchronous state includes q_h, λ_max,h, λ_min,h, µ_max,h, and µ_min,h.
  • VII. ASYNCHRONOUS ALGORITHM: Independent timers trigger agents without coordination, with identically distributed exponential waiting times.The point-of-common-coupling agent is excluded from the individual-timer assumption.
  • VII. ASYNCHRONOUS ALGORITHM: Each triggered agent measures its voltage, gathers neighboring voltage and multiplier values, updates auxiliary variables, and then updates injected reactive power.The update sequence preserves the synchronous algorithm’s local measurement-and-actuation structure.

VIII. CONVERGENCE ANALYSIS

The convergence analysis approximates the closed-loop power-flow dynamics by neglecting infinitesimal terms, yielding a strictly convex quadratic ORPF problem with linear inequality constraints.

  • VIII. CONVERGENCE ANALYSIS: The compact dual-ascent and primal-update equations are obtained by substituting approximate voltage expressions into the ORPF formulation.The derivation rewrites voltage-bound terms using the network matrices and load variables.
  • VIII. CONVERGENCE ANALYSIS: The analysis neglects infinitesimal terms, making voltages and squared voltage magnitudes affine functions of reactive-power decision variables.This approximation transforms the ORPF formulation into a tractable quadratic problem.
  • VIII. CONVERGENCE ANALYSIS: The resulting optimization problem is strictly convex and has linear inequality constraints, so strong duality holds.The synchronous and asynchronous analyses are then treated separately.

A. Syncronous case

The synchronous and asynchronous dynamics are shown to converge to optimal primal reactive-power solutions under their stated formulations and assumptions. For asynchronous control, the theoretical results cover only voltage-only or power-only constraints, while simulations examine the general case.

  • A. Syncronous case: The synchronous trajectory q(t) converges to the optimal primal solution q* for the formulated quadratic problem.The result is stated for the general convergence dynamic system and specialized constraint cases.
  • B. Asynchronous case: Theoretical asynchronous results are limited to voltage-only or power-only constraints, although simulations test the general constraint formulation.At asynchronous equilibrium, Uzawa’s necessary optimality conditions are satisfied.
  • B. Asynchronous case: Almost-sure convergence is established for asynchronous control with voltage-only constraints under exponentially distributed agent timers.Only the triggered agent’s components of the dual variables and reactive-power vector are updated at each iteration.
  • B. Asynchronous case: Almost-sure convergence is likewise established for asynchronous control with power-only constraints under the timer assumption.The corresponding result uses the power-constraint multipliers and the asynchronous reactive-power update.

IX. SIMULATIONS

Simulations on the IEEE 37 testbed show that the distributed feedback algorithm nearly matches centralized loss-minimization performance despite unmonitored loads. The strategy remains robust to larger X/R variations, while abrupt load changes can cause slight transient voltage violations.

  • Testbed and setup: The IEEE 37 testbed represents an actual 4.8kV California distribution-network portion with five deployed microgenerators and X/R ratios from 0.5 to 0.67.The network includes nearly 2 MW of active demand and 1 MVAR of reactive demand, with line lengths from 25 to almost 600 meters.
  • Testbed and setup: Both synchronous and asynchronous algorithms were simulated using a nonlinear exact grid solver, while the approximate model served only algorithm design and convergence analysis.The voltage-magnitude lower bound was set to 4700 V.
  • Testbed and setup: The simulations used time-varying loads to represent both slowly varying aggregate demand and fast-changing or intermittent industrial demands.
  • Results: The proposed algorithm achieves practically the same power-distribution-loss performance as a centralized solver with real-time access to grid parameters and load data.The distributed agents use only microgenerator voltage measurements and neighbor communication, without access to unmonitored load demands.
  • Results: Slight transient voltage-threshold violations occur after abrupt load changes, with their extent depending on execution speed relative to load variation and the primal update step.Reactive-power constraints were enforced by saturating inverter set-point references.
  • Robustness and scope: X/R variation from 0.36 to 2.6 had minimal effect on the closed-loop behavior despite the analytical homogeneity assumption.The simulations attribute this robustness to the feedback nature of the control strategy.

APPENDIX A G-PARAMETERS

The appendix defines grid-topology-dependent g-parameters and relates them to the gains G_hk used in the distributed algorithm. It also records convergence results for synchronous and asynchronous updates under the stated assumptions.

  • G-parameters: The g-parameters g_hk are defined for pairs of generator nodes and depend only on the grid’s electrical topology.They are nonzero exactly when k is a neighbor of h.
  • G-parameters: When paths from h to its neighbors are disjoint and unique, G_hk equals the inverse impedance magnitude of the path connecting h and k.Figure 9 illustrates how the gains are formed from these path relationships.
  • G-parameters: Lemma 5 establishes that the topology-defined parameters g_hk coincide with the matrix gains G_hk used in the control formulation.The appendix proves this correspondence through circuit-theoretic identities and matrix relations.
  • Convergence analysis: The update minimizes the Lagrangian with respect to the primal variables up to a term that vanishes for large U_N.This connects the distributed primal update to the dual-based convergence argument.
  • Convergence analysis: The synchronous dynamics converge to the optimal primal solution q* for the general problem and for cases with only voltage or only power constraints.The proofs use the dual formulation, projected gradient ascent, and the stated spectral conditions.
  • Convergence analysis: Under Assumption 2, the asynchronous evolution converges almost surely to q* for voltage-constrained and power-constrained formulations.The proof tracks dual-variable errors and shows convergence of the primal variables under the corresponding contraction condition.
Loading 1303.7173v2…