Virtual Power Plant

Concept Updated 2026-09-17

NC DR doesn't use the term "virtual power plant" at all — its regulatory equivalent is the Service Providing Group (SPG), Controllable Units aggregated across multiple connection points into one market-participating entity, which is the legal shape a VPP actually has to take to participate.

The activity exists at real scale in Sweden (CheckWatt alone runs 15,000+ sites), but the term itself barely gets used — only a couple of small consumer-facing actors say "virtual power plant" at all, unlike Germany where Statkraft operates VPPs explicitly branded as such, suggesting the concept is ahead of the vocabulary here.

VPP's legal shape under NC DR — Service Providing Group (SPG)Other active Swedish VPP-like aggregators — Flower, Ingrid Capacity, Capalo AITerm usage in Sweden — rare even among consumer-facing actors (Greenely, Tibber)

A virtual power plant (VPP) is a cloud-based system that aggregates and coordinates distributed energy resources — batteries, solar panels, EV chargers, heat pumps, backup generators, flexible industrial loads — to operate as a single dispatchable power plant in electricity markets. VPPs are the technological realization of Aggregation.

How it works

A VPP combines three layers:

Physical resources — distributed assets behind customer meters (residential batteries, commercial HVAC, industrial processes, EV fleets)
Digital platform — communication, monitoring, optimization, and market interface software
Market participation — the VPP bids into Balancing Markets, Flexibility Markets, wholesale markets, or provides services under contracts

The VPP operator (an aggregator) uses forecasting and optimization algorithms to decide in real time which resources to activate, balancing grid needs, market prices, and individual customer constraints (e.g., “my EV must be charged by 7 AM”).

Which AI method fits which VPP function is the subject of a 2026 academic review (global literature, not Sweden-specific): hybrid deep learning (Transformer + graph neural network) for forecasting; reinforcement learning, especially hybrid RL/model predictive control, for scheduling and dispatch; cooperative and risk-aware learning (safe RL, game theory) for market bidding; multi-agent RL, federated learning and graph neural networks for aggregation/coordination at scale; and deep RL plus digital twins for ancillary services and fault resilience. The review frames the emerging architectural consensus as hierarchical cloud-edge: cloud for long-term planning and training, edge for real-time inference — commercial platforms are cited achieving under 100ms end-to-end scheduling latency this way. (Source - Li et al AI-Driven Virtual Power Plants Review (2026))

VPPs and flexibility

In the Flexibility taxonomy, a VPP is a delivery mechanism rather than a flexibility type:

It can provide Demand Response (reducing consumption)
It can dispatch Energy Storage (charging/discharging batteries)
It can manage distributed generation (curtailing or increasing solar/CHP output)
It can combine all three in a single optimized portfolio

This makes VPPs particularly powerful for value stacking — the same portfolio serving multiple markets and products. The Network Code on Demand Response‘s Table of Equivalences formalizes this by allowing prequalification for one product to count toward others. (Source - NC DR Proposal (ENTSO-E and EU DSO Entity, 2024))

EU context

The Clean Energy Package does not use the term “virtual power plant” but establishes the legal foundations that enable them:

Independent aggregators can operate without supplier consent (Directive Art. 13)
Aggregated demand response and storage participate on equal footing with generation (Regulation Art. 3(j))
The NC DR’s Service Providing Group (SPG) concept is the regulatory equivalent of a VPP — CUs aggregated across multiple connection points into a single market-participating entity

Swedish VPPs in practice

CheckWatt AB is the leading Swedish home BESS VPP, operating across Sweden and Finland as of early 2026 (Source - CheckWatt Website (2025-2026), Source - CheckWatt Website (2025-2026)):

15,000+ customer sites; ~100 MW FCR-D capacity (summer 2024, = 1/5 of Swedish FCR-D market); 10,000 sites connected by summer 2024
Active markets: Sweden (primary), Finland (via partner Solarvoima); delivering FCR-D, FCR-N, mFRR (from May 2025), FFR (larger sites); aFRR not yet active despite barrier removal Jan 2025
Revenue performance: 2.5× vs basic price arbitrage in SE3; 4.0× in Finland (H1 2025, 10 kW/10 kWh battery)
Two operating modes: CheckWatt Optimized (ancillary services + local flex + self-consumption, recommended) and CheckWatt Savings (self-consumption only)
Customer connection via CM10 hardware gateway; EnergyInBalance customer portal; B2B2C installer-network distribution
Fee structure: €5/month fixed + 20% performance fee (10% CheckWatt + 10% installer/support)
Simultaneous participation in TSO ancillary services (FCR-D/N, mFRR), DSO local flex (Effekthandel Väst, E.ON Switch), and behind-meter optimization from the same portfolio

Other active Swedish/Nordic aggregators running VPP-like portfolios include Flower Infrastructure Technologies (home batteries), Ingrid Capacity (a Stockholm-based battery developer/operator, closer to an IPP model than a portfolio aggregator), and Capalo AI (a Helsinki-headquartered AI-driven VPP operator active across the Nordics/Baltics, illustrating that Swedish BSP agreements aren’t exclusively held by Swedish-founded firms). These together constitute the current practical VPP landscape in Sweden. See Aggregation for full profiles, the aggregator-as-infrastructure analysis, and the BSP structural barriers that constrain VPP scale.

“Virtual power plant” as a term is not yet widely used in Sweden, even where the underlying activity exists. A 2026 LTH thesis interview study notes the concept is discussed extensively in international literature and has scaled examples abroad (Statkraft operates working VPPs in Germany), but in Sweden only a handful of consumer-facing actors (Greenely, Tibber) use the term at all, and neither at meaningful scale. (Source - Digitala Monster — AI och Svenska Energisystemet (2026))

Data gaps

Revenue models and market participation breakdown by Nordic VPPs (what share of revenue from TSO vs DSO vs wholesale) — CheckWatt and Flower have public disclosures (see Aggregation); see Aggregation › Data gaps for Ingrid Capacity’s and Capalo AI’s specific market participation, tracked there rather than duplicated here
Reinforcement-learning-based real-time control of renewable/DER-heavy power systems — Energiforsk 2022:894 (Maskininlärningsbaserad realtidsstyrning av förnyelsebara och säkra elkraftsystem, Hagmar & Le, Nov 2022) is already sitting unread in raw/ (deliberately held back during the 2026-09-13 metering-data ingest as out of scope) and is a plausible source for how VPP-style dispatch algorithms are evolving beyond the forecasting/scheduling methods this page and Source - Li et al AI-Driven Virtual Power Plants Review (2026) already cover

Sources

Linked from 20