PaperlayerView exact revision on arXiv

Source-linked AI summary

Headroom-Aware Stochastic Adaptive Model Predictive Control for Load Frequency Control in Microgrids

Erfan Mehdipour Abadi, Shuo Yuan, Le Yi Wang, Caisheng Wang, Feng Lin

arXiv:2609.05332v1eess.SY

TL;DR

Fixed-headroom MPC can overestimate deliverable IBR regulation when renewable capacity varies, causing stochastic saturation and degraded LFC, especially with tight reserve margins. The paper develops SHCMPC with headroom-dependent constraints and SAMPC with adaptive penalties; simulations show that both improve regulation, with SHCMPC eliminating saturation and SAMPC substantially mitigating it at lower computational effort.

  • Problem

    Fixed-headroom MPC may command more regulation than stochastic, time-varying renewable headroom can physically deliver, degrading LFC under reduced reserve margins.

  • Method

    SHCMPC imposes predicted headroom as time-varying input constraints, whereas SAMPC incorporates normalized headroom through adaptive input penalties.

  • Results

    SHCMPC eliminates saturation in tested cases, while SAMPC reduces 5% reserve saturation severity by approximately 90% and duration by approximately 71%.

  • Takeaways & Limitations

    Headroom-aware MPC improves regulation under tight reserve margins while trading strict saturation avoidance against lower online computational effort.

Abstract

from arXiv · show

As the penetration of inverter-based resources (IBRs) increases in microgrids, they are increasingly expected to play a greater role in load frequency control (LFC). Model predictive control (MPC) is attractive for LFC because it incorporates system dynamics and operational constraints. However, most MPC-based LFC formulations rely on fixed reserve headroom based on forecasted renewable availability or storage systems. Under short-term renewable intermittency, IBR headroom is stochastic and time-varying, causing optimal control commands to exceed the physically deliverable regulation capability and cause stochastic saturation. This control-actuator mismatch degrades LFC performance. Accordingly, this study develops two headroom-aware strategies for PV-dominated microgrids. First, stochastic headroom constrained MPC (SHCMPC) incorporates headroom predictions through time-varying input constraints to enforce control feasibility. Second, stochastic adaptive MPC (SAMPC) embeds headroom awareness into the MPC objective function by adaptively penalizing control actions based on predicted headroom, reducing reliance on units with limited headroom without hard time-varying constraints. Simulation results show that stochastic saturation degrades conventional MPC-based LFC, particularly under tight reserve margins. Both strategies improve regulation performance. SHCMPC eliminates saturation events, while SAMPC achieves substantial saturation mitigation with lower computational effort, offering a computationally efficient alternative for real-time LFC under stochastic renewable availability.

I. INTRODUCTION

High IBR penetration creates a need for fast LFC while making renewable regulation headroom stochastic and time-varying. The paper develops headroom-aware MPC strategies to address saturation and improve regulation under tight reserve margins.

  • IBRs can provide fast regulation, but reduced inertia makes microgrid frequency dynamics more sensitive to active-power imbalances.
  • Renewable IBR headroom differs from scheduled reserve because actual available capacity varies stochastically with environmental conditions.
  • Fixed-headroom MPC can issue commands exceeding instantaneous deliverable power, producing stochastic saturation and degrading LFC performance as reserve margins shrink.
  • SHCMPC uses predicted headroom as time-varying input constraints to enforce feasibility and mitigate optimization-actuation mismatch.
  • The formulation distinguishes scheduled reserve margin from actual headroom and evaluates their effects using LFC and saturation-severity measures.
  • SAMPC uses normalized headroom in adaptive input penalties to shift control effort away from resources with limited headroom while retaining standard MPC structure.

III. STOCHASTIC HEADROOM-AWARE MICROGRID MODELING

The modeling framework combines microgrid control architecture, aggregate frequency-response dynamics, and Markovian prediction of short-term IBR headroom variations.

  • The framework first describes the control architecture, then derives an aggregate state-space frequency model, and finally models IBR availability as a Markovian stochastic process.

A. Microgrid Control Architecture

The microgrid architecture routes headroom-limited IBR commands through converter dynamics into aggregate frequency-response dynamics, with frequency feedback driving the headroom-aware LFC layer.

  • Each renewable IBR receives an LFC active-power deviation command, and converter response is represented by first-order dynamics.
  • Stochastic availability determines each IBR's headroom, while a saturation block limits the delivered command to feasible levels.
  • Saturated IBR power deviations are aggregated into the microgrid frequency-response model, while load deviation enters as an exogenous disturbance.
  • The resulting frequency deviation is fed back to SHCMPC and SAMPC, which use it and short-term headroom information to compute regulation commands.
  • The continuous model is discretized with zero-order hold and augmented with an integral frequency-error state for MPC prediction.

C. Markovian Stochastic Model for IBR Capacity

IBR capacity is modeled as an infrequently switching finite-state Markov chain to predict short-term stochastic headroom over the MPC horizon. The framework can also use alternative forecasting methods that provide predicted capacity.

  • Real-time IBR capacity varies with irradiance, clouds, temperature, and shading and is temporally correlated over short horizons.
  • Capacity is quantized into discrete levels and modeled as an infrequently switching finite-state Markov chain.
  • Self-transition-dominated probabilities make the capacity process likely to remain in its current state over short horizons.
  • The Markov model provides short-term capacity predictions whose realizations determine stochastic headroom for the headroom-aware MPC formulations.
  • The proposed framework is not restricted to Markovian prediction and can use any forecasting method that supplies predicted available capacity.

IV. HEADROOM-AWARE MPC-BASED LFC

The baseline uses finite-horizon constrained MPC to regulate frequency while penalizing frequency error, accumulated error, and PV-input modulation. Its fixed input limits are based on scheduled reserve margins rather than updated stochastic headroom, so realized saturation can degrade regulation.

  • MPC-based LFC: Finite-horizon MPC predicts frequency evolution and applies only the first optimized control action at each sampling instant.The formulation uses receding-horizon feedback with measured-state updates.
  • Conventional constrained MPC: The conventional baseline fixes input limits offline according to the scheduled reserve margin and does not update them using real-time headroom.This distinguishes the baseline from headroom-aware formulations.
  • MPC-based LFC: The cost penalizes predicted frequency deviation, accumulated frequency error, and PV power modulation.The state weights target frequency deviation and integral error, while the input weight regularizes PV control effort.
  • Conventional constrained MPC: When actual headroom is below the scheduled bound, applied commands can be clipped, degrading LFC performance under reduced reserve margins.The baseline therefore serves as the fixed-headroom reference for evaluating the proposed strategies.

B. Stochastic Headroom Constrained MPC

SHCMPC replaces fixed scheduled-reserve bounds with predicted stochastic headroom bounds over the control horizon. It uses short-term capacity predictions to constrain commands according to future deliverable PV capability.

  • Stochastic headroom constraints: SHCMPC replaces offline fixed input bounds with predicted stochastic headroom bounds at each future control step.The bounds are derived from short-term PV capacity predictions under the Markovian capacity model.
  • Stochastic headroom constraints: Stochastic headroom constraints are enforced over the first N hard control moves, while scheduled reserve bounds remain for later moves.This allocation reflects the greater reliability of short-term headroom predictions.
  • Stochastic headroom constraints: The resulting upper-bound vector combines predicted headroom for near-term moves with scheduled reserve bounds for the remaining horizon.These bounds are stacked across the control horizon for the constrained MPC problem.

U SHC

The SHCMPC formulation replaces the conventional fixed upper input bound with a predicted stochastic headroom bound. This directly constrains optimized commands by predicted deliverable PV capability.

  • U SHC: SHCMPC replaces the fixed upper bound U sch max with the predicted stochastic headroom bound U SHC max(k).The bound is introduced in the optimization problem at time k.
  • U SHC: Constraining commands by predicted deliverable PV headroom reduces the likelihood of stochastic saturation.Unlike scheduled-reserve limits, the bound reflects predicted future availability.

C. Stochastic Adaptive MPC

SAMPC incorporates stochastic headroom through adaptive input penalties rather than time-varying hard constraints. It shifts effort toward units with greater predicted capability while increasing overall conservatism when aggregate headroom is scarce.

  • Adaptive headroom weighting: SAMPC replaces time-varying hard constraints with adaptive input weighting based on predicted stochastic PV headroom.The resulting optimization remains an unconstrained quadratic problem.
  • Adaptive headroom weighting: Units with smaller normalized headroom shares receive larger penalties, shifting regulation effort toward units with greater available headroom.The adaptive factors quantify each unit’s relative share of available regulation capability.
  • Aggregate scarcity adaptation: An aggregate-headroom scarcity factor increases the overall penalty when predicted regulation capability is limited.SAMPC remains close to nominal MPC with sufficient reserve and becomes more conservative as aggregate headroom declines.
  • Adaptive headroom weighting: The adaptive weighting matrix is assembled from scalar penalties and replaces the fixed input-weighting matrix in conventional MPC.The matrices are defined over the control horizon and stacked into a block-diagonal form.
  • Computational trade-off: SAMPC lowers online computation by removing explicit time-varying input constraints, although plant-level saturation still applies if commands exceed realized headroom.This creates a computational trade-off relative to SHCMPC.

V. CASE STUDIES

The case studies evaluate conventional MPC, SHCMPC, and SAMPC for a four-unit PV-dominated microgrid under stochastic PV availability, multiple reserve margins, and a step load disturbance. They also define the simulation settings and metrics used to compare frequency regulation, saturation, and computational burden.

  • System model: The simulated microgrid contains N = 4 PV units participating in load frequency control.The PV-dominated system is represented by an aggregate frequency-response model with PV converter dynamics.
  • Stochastic availability: PV capacity follows a bounded Markovian random-walk model with ∆P = 0.05 MW, pstay = 0.99, and pup = pdown = 0.005.Each capacity trajectory is initialized at its forecasted value and evolves stochastically at each sampling instant.
  • Disturbance: A deterministic 0.2 MW load increase is applied at t = 40 s while PV capacities evolve stochastically under identical load conditions.This setup isolates the effect of headroom uncertainty on LFC performance.
  • Controller settings and metrics: Controllers use Np = 50 and Nc = 5; SHCMPC enforces stochastic headroom constraints over the first Nc = 2 moves, while remaining moves use scheduled reserve bounds.The simulations evaluate accumulated frequency error, large excursions, deviation persistence, saturation duration, and average online solution time.
  • Simulation cases: The study compares ideal conventional MPC, conventional MPC with stochastic saturation, SHCMPC, and SAMPC to isolate optimization–actuation mismatch and evaluate headroom-aware control.The stochastic-saturation case applies conventional MPC commands through actual headroom limits, while SHCMPC and SAMPC are evaluated as proposed alternatives.

B. Effect of Stochastic Saturation on Conventional MPC

Stochastic PV headroom varies across reserve-margin scenarios and units, creating an optimization–actuation mismatch when conventional MPC commands exceed realized capability. The resulting saturation increasingly degrades frequency regulation as reserve margins tighten, especially at 5%.

  • Headroom profiles: Stochastic headroom changes across 20%, 10%, and 5% scheduled reserve margins, with uneven regulation capability among PV units.The 5% case shows especially uneven distribution, so fixed scheduled margins do not capture instantaneous capability.
  • Evaluation setup: Conventional MPC is evaluated with commands applied directly under fixed bounds and with the same commands clipped by realized headroom limits.The difference isolates the optimization–actuation mismatch caused by short-term PV capacity variations.
  • Frequency response: As scheduled reserve margin decreases, conventional MPC produces a deeper frequency nadir and visible post-transient deviations in the 5% case.The results indicate that stochastic saturation mainly affects sustained regulation after the main recovery.
  • Input saturation: Under the 5% reserve margin, conventional MPC repeatedly exceeds realized headroom, especially for PV3 and later PV2, so delivered inputs are clipped.The redistribution is triggered indirectly through frequency feedback rather than planned using predicted headroom, contributing to post-transient fluctuations.
  • Performance impact: Approximately 74% higher IAE and 594% higher ITAE occur under stochastic saturation in the 5% reserve case.The larger ITAE increase indicates degradation associated with persistent frequency deviations.
  • Implication: Neglecting short-term headroom variations produces repeated input saturation and persistent frequency deviations, particularly under tight reserve margins.These findings motivate headroom-aware control strategies that incorporate stochastic PV availability into control decisions.

C. Performance of Headroom-Aware LFC Strategies

Under tight reserve margins, stochastic headroom makes conventional MPC vulnerable to persistent frequency deviations and saturation. SHCMPC removes infeasible commands through hard headroom constraints, while SAMPC reduces deviations and saturation with lower computational burden.

  • At 5% reserve, conventional MPC exhibits persistent frequency fluctuations, whereas SHCMPC and SAMPC substantially reduce them.At 20% reserve, responses are nearly identical; differences emerge at 10% and become significant at 5%.
  • SHCMPC reduces the IAE and ITAE by approximately 42% and 86%, respectively, relative to the stochastic-saturation baseline at 5% reserve.
  • SHCMPC eliminates saturation in all reserve-margin cases, while SAMPC reduces 5% reserve saturation severity by approximately 90% and duration by approximately 71%.For SAMPC, saturation severity decreases from 1667.47 to 171.14 and duration from 526.3s to 150.4s.
  • SAMPC requires approximately 0.20-0.21 ms per solution, compared with 0.62-0.70 ms for SHCMPC.The adaptive-penalty formulation yields roughly a threefold reduction in average solution time while retaining substantial saturation mitigation and competitive LFC performance.
  • The two strategies improve renewable utilization under reduced reserve margins by incorporating short-term headroom information into MPC-based LFC.The results reflect a tradeoff between strict saturation avoidance and online computational efficiency.
Loading 2609.05332v1…