Dynamic Virtual Power Plants

• DVPPs are dynamic aggregations of heterogeneous DERs engineered to provide prescribed frequency and voltage responses, distinguishing them from static VPPs.
• They employ advanced control methods such as adaptive participation factors and decentralized model matching to manage device limits and communication challenges.
• DVPP applications include fast frequency control, voltage support, and market-based ancillary services, validated via simulations, hardware-in-the-loop, and field tests.

Searching arXiv for recent and foundational DVPP papers to ground the encyclopedia entry. Dynamic Virtual Power Plants (DVPPs) are coordinated aggregations of heterogeneous distributed energy resources (DERs) that are engineered to exhibit a prescribed dynamic input–output behavior for grid services rather than merely a static aggregate power schedule. In the core power-systems formulation, a DVPP is explicitly dynamic at the local, global, and economic levels: local device dynamics and constraints are incorporated, aggregate ancillary-service behavior is specified and enforced, and internal operation is coordinated under resource variability and market participation requirements (Marinescu et al., 2021). Subsequent work has specialized this idea to grid-following fast frequency control, grid-forming voltage and frequency support, hybrid AC/DC interfaces, direct participation in secondary frequency control, delay-aware participation in inertia and primary frequency response markets, and experimentally validated hardware-in-the-loop implementations (Andrejewski et al., 2023, Häberle et al., 2022, Feng et al., 6 Mar 2025).

1. Definitions and conceptual scope

The defining feature of a DVPP is that the aggregation is specified by a dynamic contract. Rather than treating a portfolio of wind, photovoltaic, storage, hydro, controllable demand, or flexible building resources as a purely commercial bundle, the DVPP literature formulates the aggregate as a single controllable entity whose response to frequency and voltage disturbances is deliberately shaped. This is the sense in which a DVPP differs from conventional VPPs and aggregators that mainly address pooling, bidding, and dispatch on static or slow time scales (Marinescu et al., 2021).

A closely related strand of work uses the term in an operational and intertemporal sense: a DVPP is then a VPP whose dispatch explicitly recognizes the dynamic link between past service allocation and future availability or participation. In that formulation, current allocations affect future resource availability through a latent participation state, and fairness becomes a control lever for long-run flexibility and profitability (Chen et al., 1 Jun 2026). A plausible implication is that the contemporary DVPP literature spans both electrodynamic aggregation and intertemporal flexibility management, with the former emphasizing ancillary-service dynamics and the latter emphasizing state-dependent availability.

The concept emerged from a common system-level motivation. As non-synchronous, inverter-based DERs increase, conventional inertial and governor-based support from synchronous machines declines, while weather-driven variability and device-specific power, energy, and bandwidth limits become more prominent. The central claim across the literature is therefore not that any single DER can emulate a conventional plant across all scales, but that a heterogeneous ensemble can collectively provide fast frequency control, voltage control, inertia-like behavior, damping, and reserve delivery when its internal contributions are coordinated with explicit respect for device limits and response times (Andrejewski et al., 2023, Tong et al., 5 Mar 2026).

2. Aggregate dynamic specification and disaggregation

Most DVPP control designs begin by prescribing an aggregate transfer behavior and then decomposing that behavior into local targets. In the adaptive divide-and-conquer formulation, the aggregate specification is written as a desired transfer matrix

$$\begin{bmatrix}\Delta p_\mathrm{des}(s)\ \Delta q_\mathrm{des}(s)\end{bmatrix} = T_\mathrm{des}(s) \begin{bmatrix}\Delta f(s)\ \Delta v(s)\end{bmatrix},$$

with diagonal channels for frequency-to-active-power and voltage-to-reactive-power response. The local closed-loop device transfers $T_i(s)$ are required to sum to the aggregate specification, and each controllable device is assigned a local reference model through adaptive dynamic participation matrices $M_i(s)$, with $\sum_i M_i(s)\approx I_2$ over the frequency range of interest (Häberle et al., 2021).

This decomposition appears in several technically distinct forms. In grid-following fast frequency control, the aggregate response is described by a desired transfer function from frequency deviation to active power, sometimes represented by a second-order reference model

$$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$

and the DVPP controller computes device-level references $P_{i,\mathrm{ref}}(t)$ so that

$$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$

while respecting device limits and availability variations (Andrejewski et al., 2023). In the Nordic fast-frequency-reserve design, the aggregate objective is cast as a decentralized model-matching problem in which the global open-loop gain $L(s)=\sum_i L_i(s)$ is shaped to match a prescribed target $G(s)F_{\mathrm{FCR}}(s)$, and local controllers are parameterized as

$K_i(s)=c_i(s)\frac{F_{\mathrm{FCR}}(s)}{H_i(s)},$

with dynamic participation factors $T_i(s)$0 chosen so that $T_i(s)$1 over the relevant bandwidth (Björk et al., 2021).

The participation factors themselves are dynamic objects. The adaptive formulations distinguish static participation factors (SPF), dynamic participation factors (DPF), and adaptive dynamic participation factors (ADPF). In the LPV $T_i(s)$2 design, low-pass, band-pass, and high-pass participation profiles are assigned according to device time scales, and low-pass DC gains $T_i(s)$3 are updated in proportion to time-varying active or reactive capability: $T_i(s)$4 This allows weather-driven units such as wind and PV to remain inside the control architecture without assuming fixed capacity (Häberle et al., 2021).

Local realization is then handled by device-level controllers. In grid-following designs, this usually means PLL-based synchronization, dq-frame current control, and outer active/reactive power loops. In the adaptive divide-and-conquer framework, controllable devices are equipped with LPV $T_i(s)$5 state-feedback laws synthesized from vertex LMIs, while transient device limits are embedded through ellipsoidal LMIs. This makes the aggregate specification a supervisory object and the local controller a robust model-matching mechanism (Häberle et al., 2021).

3. Converter paradigms: grid-following, grid-forming, hybrid, and modular DVPPs

The literature distinguishes grid-following, grid-forming, and hybrid DVPPs. Grid-following DVPPs measure grid frequency, typically via a PLL, and modulate active power injection in response to frequency deviation. They support an existing grid reference but do not establish it. Grid-forming DVPPs, by contrast, synthesize voltage and frequency as functions of measured power and can therefore contribute virtual inertia, damping, black-start capability, and operation in weak grids (Andrejewski et al., 2023, Häberle et al., 2022).

DVPP type	Defining control principle	Representative paper
Grid-following	PLL-based synchronization and frequency-to-power modulation	(Andrejewski et al., 2023)
Grid-forming	Voltage-source behavior with virtual inertia, damping, and droop	(Häberle et al., 2022)
Hybrid	Coordinated combination of grid-forming and grid-following DERs	(Häberle et al., 2022)

In the grid-forming formulation, the aggregate specification is inverted relative to the grid-following case. At the point of common coupling, active and reactive power disturbances induce frequency and voltage responses through

$T_i(s)$6

with a representative choice

$T_i(s)$7

The local grid-forming units are then coordinated so that inverse-sum or sum conditions on their local transfer functions reproduce the aggregate target (Häberle et al., 2022).

Weak-grid operation is a central reason for this distinction. Grid-following inverters rely on a PLL and can suffer from degraded damping and destabilizing interactions as grid impedance rises. Grid-forming converters instead regulate terminal voltage and frequency directly; recent multi-timescale designs implement this at the aggregate DVPP level using a virtual synchronous generator law

$T_i(s)$8

supplemented by filtered droop relations for the active-power–frequency and reactive-power–voltage channels (Tong et al., 5 Mar 2026).

The architecture has also been generalized spatially and structurally. For spatially distributed DVPPs in MV grids with non-negligible $T_i(s)$9 ratios, the control design uses rotational active and reactive powers

$M_i(s)$0

so that frequency and voltage coordination remain meaningful when the usual inductive-network decoupling is inaccurate (Häberle et al., 2022). In parallel, modular DVPPs have been introduced as standardized Advanced Grid Interfaces composed of four basic module types—AC-coupled AC-output, DC-coupled AC-output, AC-coupled DC-output, and DC-coupled DC-output—so that diverse AC, DC, and hybrid microgrid configurations can be represented by compatible terminal contracts such as $M_i(s)$1, $M_i(s)$2, $M_i(s)$3, and $M_i(s)$4 (He et al., 2024).

4. Ancillary services, reserve products, and market participation

Frequency support is the dominant application domain. One line of work addresses fast frequency reserves and disturbance containment in low-inertia systems by coordinating slow hydropower FCR with fast wind-based FFR. In the Nordic 5-machine representation, the desired FCR-D response is encoded as

$M_i(s)$5

with $M_i(s)$6 in the reported case. The key result is that wind FFR can compensate hydro’s non-minimum-phase bandwidth limitations, allowing system-operator requirements to be met without battery storage or wind curtailment in the studied low-inertia scenario (Björk et al., 2021).

A second line integrates DVPPs directly into secondary frequency control. In the AGC-based formulation, the area control error is

$M_i(s)$7

and the DVPP receives a secondary-control share exactly like a synchronous generator through a participation factor $M_i(s)$8, with

$M_i(s)$9

Internal DVPP redispatch is then computed from filtered proportional weights

$\sum_i M_i(s)\approx I_2$0

and unit setpoints are assigned as $\sum_i M_i(s)\approx I_2$1. In the two-area benchmark, the DVPP contributes $\sum_i M_i(s)\approx I_2$2 to a $\sum_i M_i(s)\approx I_2$3 secondary-frequency-control requirement when $\sum_i M_i(s)\approx I_2$4, while synchronous generators supply the remaining $\sum_i M_i(s)\approx I_2$5 (Adabi et al., 2022).

More recent work has shifted from control design alone to joint dynamic-market formulations. In the energy–IPFR market framework, a VPP/DVPP bids non-delayed inertia, delayed inertia, and droop factor, and the market clears these products together with energy while respecting RoCoF, nadir, and quasi-steady-state frequency-security constraints. The staged dynamic model explicitly distinguishes immediate inertial response, delayed grid-forming support, and later primary response, and the modified IEEE 30-bus case shows that VPP participation reduces total social cost while improving flexibility in both energy and inertia provision (Feng et al., 6 Mar 2025).

Robust reserve sizing has also been derived analytically. In the robust frequency-regulation formulation, the DVPP exposes aggregate virtual inertia $\sum_i M_i(s)\approx I_2$6 and aggregate virtual damping $\sum_i M_i(s)\approx I_2$7. These are sized against RoCoF, nadir, and steady-state frequency limits under worst-case disturbances, and then allocated among IBRs through an optimization that respects heterogeneous costs and network constraints. On the modified IEEE 39-bus system, a near-minimal feasible pair reported in the case study is $\sum_i M_i(s)\approx I_2$8 and $\sum_i M_i(s)\approx I_2$9, satisfying the stated RoCoF and steady-state limits while reducing overshoot relative to larger-inertia choices (Zhu et al., 2024).

Beyond ancillary-service markets, the term DVPP has also been extended to risk-aware multi-market scheduling. In that setting, the DVPP is a portfolio operator that dynamically aggregates heterogeneous DERs, co-optimizes day-ahead energy, reserve capacity, reserve activation, and imbalance positions, and internalizes dynamic network tariffs through a two-stage stochastic program with CVaR and Benders decomposition. The reported case study on a Swiss low-voltage feeder shows that dynamic tariffs shift withdrawals toward low-tariff hours, while strong tariff signals reduce expected profitability by up to $$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$0 with limited additional flexibility gains (Zapparoli et al., 29 Apr 2026).

5. Internal redispatch, participation dynamics, and long-run availability

Internal redistribution of aggregate obligations is a persistent theme. In direct secondary-frequency-control participation, the DVPP’s few-second redispatch layer is intentionally simple: it relies only on available active-power measurements and a first-order filter with $$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$1, avoiding a more complex optimization while adapting automatically when a wind turbine trips or a unit’s available output changes (Adabi et al., 2022). In adaptive divide-and-conquer control, the same functional role is performed by ADPF or ADPM gains that are updated from active and reactive headroom, either centrally or through consensus filters, so that constrained or unavailable units reduce participation and unconstrained units increase it (Häberle et al., 2021).

A more decentralized formulation has recently been proposed for fast local frequency correction. In that layered architecture, each DVPP node estimates unmeasured active-power imbalance with a Luenberger-type disturbance estimator and computes a local BESS setpoint

$$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$2

When actuator saturation prevents complete compensation, neighboring DVPP nodes exchange only redispatch signals and apply a DAPI law

$$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$3

so that residual mismatch is reallocated proportionally across the communication graph. The 4-bus study uses a communication delay $$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$4, BESS limits of $$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$5, $$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$6, and $$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$7 p.u. at three DVPP nodes, and demonstrates that neighboring nodes cover the shortfall when one node saturates (Ahmad et al., 9 Jul 2025).

The operational literature adds another layer by treating participation itself as dynamic. In the fairness-based model, each consumer’s future availability depends on a latent participation state,

$$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$8

and fairness is imposed as a dispersion cap on per-period allocations. Strict fairness can induce curtailment, but a slack-augmented fairness mechanism preserves aggregate collection while improving future aggregate availability when $$H_d(s) = \frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2},$$9 is increasing and strictly concave (Chen et al., 1 Jun 2026). This suggests that, in some DVPP formulations, internal redispatch is not only a physical feasibility problem but also a participation-design problem.

Closely related work on PEM-enabled VPPs reaches a similar conclusion through energy-state-aware MPC. There the upper-level MPC dispatches VPPs and conventional generators subject to time-varying energy-state constraints, preventing unexpected capacity saturation. In the hardware-in-the-loop study, AGC drives a $P_{i,\mathrm{ref}}(t)$0 battery to saturation after about $P_{i,\mathrm{ref}}(t)$1 minutes following a $P_{i,\mathrm{ref}}(t)$2 load step, whereas the energy-aware MPC begins throttling at about $P_{i,\mathrm{ref}}(t)$3 minutes and sustains service beyond $P_{i,\mathrm{ref}}(t)$4 minutes (Amini et al., 2019). Although this work is framed in terms of VPPs and PEM rather than the power-systems DVPP formalism, it reinforces the broader point that dynamic capability must be modeled as state-dependent, not static.

6. Validation platforms, benchmark systems, and empirical evidence

The validation corpus is unusually diverse. Foundational control papers rely on EMT-level or phasor-domain simulation, while later work extends to PHIL, cyber-physical HIL, and market-oriented case studies.

The clearest experimental validation of the DVPP concept uses a multi-converter power hardware-in-the-loop bench. The setup comprises three $P_{i,\mathrm{ref}}(t)$5 back-to-back converter systems as DER emulators, a $P_{i,\mathrm{ref}}(t)$6 synchronous generator providing a frequency-variable grid, a load unit for short-term disturbances, and a Bachmann industrial PLC that computes power setpoints from measured grid frequency. All converter systems operate in grid-following mode, and experiments are conducted with and without DVPP coordination as well as under SPF, DPF, and ADPF schemes. The reported conclusion is that the DVPP delivers effective grid-following frequency control despite load variations and fluctuating generation capacities, and that ADPFs are superior to SPF and DPF in this setup (Andrejewski et al., 2023).

Simulation-based validation is equally extensive. The IEEE nine-bus system is used repeatedly as a control-design benchmark. In the adaptive divide-and-conquer study, a frequency-only DVPP at bus 1 combines hydro, a $P_{i,\mathrm{ref}}(t)$7 BESS, and a $P_{i,\mathrm{ref}}(t)$8 supercapacitor with a target $P_{i,\mathrm{ref}}(t)$9, where $$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$0 and $$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$1. A separate MIMO case at bus 3 combines wind ($$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$2), PV ($$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$3), and a STATCOM ($$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$4) with $$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$5, $$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$6, $$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$7, and $$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$8. Across these cases, the DVPP tracks the desired active and reactive references, improves frequency nadir and voltage deviation relative to the displaced synchronous generator, and preserves performance under step-like PV capacity loss through online adaptation (Häberle et al., 2021).

Grid-forming and modular formulations add further test systems. Spatially distributed and hybrid DVPPs are validated on the IEEE nine-bus transmission grid coupled to an MV distribution grid, with cases that replace synchronous generators, vary the grid-forming share, and compare classical $$P_{\mathrm{DVPP}}(t)=\sum_{i=1}^{N} P_i(t), \qquad e(t)=P_{\mathrm{ref}}(t)-P_{\mathrm{DVPP}}(t),$$9–$L(s)=\sum_i L_i(s)$0 control to rotational $L(s)=\sum_i L_i(s)$1–$L(s)=\sum_i L_i(s)$2 control under $L(s)=\sum_i L_i(s)$3. The modular grid-forming framework uses detailed converter models with $L(s)=\sum_i L_i(s)$4 switching, $L(s)=\sum_i L_i(s)$5 current limiting, and a $L(s)=\sum_i L_i(s)$6 disturbance in the IEEE 13-bus system; the reported behavior includes accurate tracking of $L(s)=\sum_i L_i(s)$7, $L(s)=\sum_i L_i(s)$8, and $L(s)=\sum_i L_i(s)$9, as well as correct $G(s)F_{\mathrm{FCR}}(s)$0 power sharing in modular sub-DVPPs (Häberle et al., 2022, He et al., 2024).

Market-oriented and system-level validations expand the scale. The Nordic 5-machine study examines disconnection of a $G(s)F_{\mathrm{FCR}}(s)$1 importing DC link under high- and low-inertia scenarios; the modified IEEE 30-bus system is used for joint energy–IPFR market clearing; the modified IEEE 39-bus system is used for robust frequency-regulation sizing under sequential disturbances; the Swiss 97-bus low-voltage feeder supports risk-aware multi-market scheduling with dynamic tariffs; and the Finnish week-long case study evaluates coalition formation and reward allocation for a portfolio containing $G(s)F_{\mathrm{FCR}}(s)$2 wind, $G(s)F_{\mathrm{FCR}}(s)$3 PV, and $G(s)F_{\mathrm{FCR}}(s)$4 BESS (Björk et al., 2021, Feng et al., 6 Mar 2025, Zhu et al., 2024, Zapparoli et al., 29 Apr 2026, Holly-Ponientzietz et al., 30 Jun 2026).

7. Cooperation, limitations, and research directions

A mature DVPP cannot be understood as a control problem alone. Cooperative design and revenue allocation have therefore become explicit research topics. In the coalitional formulation for dynamic ancillary services, the coalition value is computed from optimal bids and ex-post delivery tests, and allocations are selected from the Nucleolus or the Shapley value depending on realized convexity. The reported mechanism satisfies individual rationality, coalitional stability, incentive compatibility, optimality, fairness, and ex-post consistency in the sense stated in the paper, and the Finnish case study reports that the realized cooperative reward exceeded the sum of realized standalone values by $G(s)F_{\mathrm{FCR}}(s)$5 on average over $G(s)F_{\mathrm{FCR}}(s)$6 hours (Holly-Ponientzietz et al., 30 Jun 2026).

Several misconceptions are corrected by this literature. DVPPs are not restricted to grid-following inverter fleets; grid-forming and hybrid designs are central to the field (Häberle et al., 2022). They are not merely static aggregators with faster dispatch; their defining characteristic is an explicit dynamic contract at the aggregate interface (Marinescu et al., 2021). Nor are they necessarily centralized in real time: decentralized model matching, disturbance-estimation-based local compensation, and neighbor-to-neighbor DAPI redistribution all appear as viable architectures (Björk et al., 2021, Ahmad et al., 9 Jul 2025).

The current literature also states clear limitations. Experimental papers note that detailed PHIL interface methods, switching frequencies, sampling rates, latency compensation, and quantitative performance metrics are often omitted from brief summaries, which constrains exact reproducibility (Andrejewski et al., 2023). Control-design papers acknowledge communication-delay modeling, polytope conservatism in LPV synthesis, and simplified saturation handling as open issues (Häberle et al., 2021). Broader conceptual papers identify regulatory adaptation, new stability notions for converter-dominated systems, cross-DSO/TSO coordination, and tractable multi-scale modeling of power-electronic interactions as unresolved challenges (Marinescu et al., 2021).

The dominant future directions follow directly from these gaps. They include voltage-control and grid-forming field demonstrations beyond simulation and PHIL, richer optimization-based disaggregation under device and network constraints, larger ensembles and geographically distributed portfolios, explicit delay-robust and uncertainty-aware control, standardized Advanced Grid Interface contracts for AC/DC hybrid grids, and tighter integration of market design with dynamic deliverability metrics such as RoCoF, nadir, and activation envelopes (He et al., 2024, Zhu et al., 2024, Feng et al., 6 Mar 2025). A plausible implication is that the long-run trajectory of DVPP research is toward a unified framework in which dynamic control, intertemporal resource state, network feasibility, and remuneration design are co-specified rather than treated as separate layers.