A modeled virtual power plant reported higher regulation scores and ancillary-service profit, as well as lower operating cost, in the joint scenario with fast-response requirements. The paper compares four modeled cases, so the figures describe one specified setup and its inputs.
In the joint-VPP comparison, the fast-response case reported an average regulation performance score of 0.964, compared with 0.952 without the requirement. Its minimum score was 0.921, versus 0.857. Operating cost was $59.0 rather than $65.2, a reported 9% reduction, and ancillary-service profit was $106.5 rather than $98.8, a reported 8% increase.
How the model treated rapid changes
Fast-response demand was derived from quantiles of a control-signal ramp-rate variable. The requirement therefore came from selected points in the historical distribution of how quickly the signal changed.
The paper put those requirements into the bidding model as chance constraints, which let the model set a target confidence for meeting a requirement. Regulation used a one-minute sampling interval and a 90% confidence level, while reserve response was required within 10 minutes.
Four cases were compared: independent resources without fast-response requirements, independent resources with them, a joint VPP without them and a joint VPP with them. The portfolio consisted of an energy-storage unit, an EV fleet and a thermostatically controlled load, or TCL.
The resource mix mattered
The results differed by resource. Storage and EV performance was unchanged in the independent-resource comparisons, while the TCL-alone and joint-VPP constraint cases reported higher performance scores and ancillary-service profits.
For the TCL operating alone, the fast-response case reported lower frequency-regulation capacity bids and higher reserve capacity bids. The bidding change was smaller when resources participated together; the paper attributes that difference to fast resources compensating for TCL limitations.
The TCL-alone comparison showed a clear difference between scenarios. The fast-response case reported average and minimum performance scores of 0.960 and 0.912, versus 0.833 and 0.590 without the requirements. Its reported ancillary-service profit was $39.2, marked in the paper as a 17% increase. No confidence intervals, statistical tests or repeated-simulation uncertainty measures were reported.
A small system with mixed data
The modeled storage unit had 0.5 megawatts of power capacity and 1 megawatt-hour of energy capacity. The EV fleet had 30 bidirectional vehicles rated at 7.68 kilowatts each, and the TCL was rated at 1 megawatt. Within the model, storage and EV power could be adjusted between zero and rated power within five seconds, while the TCL ramp rate was restricted to 100 kilowatts per minute.
The simulations used NYISO day-ahead prices from April 13, 2024, and historical PJM RegD control-signal data. The analysis notes that NYISO control-signal data were unavailable, so the simulation combined the two sources rather than using a single set of market inputs.
The optimization problems were solved with Gurobi V11.0.0 and MATLAB R2023b with YALMIP. The paper states that the complete model, code and data were open sourced through a cited repository.
What the numbers do not establish
The authors present the approach as a way to balance ancillary-service response requirements with the different capabilities of distributed energy resources, and as a way to formulate more reasonable VPP operating strategies and enhance performance score, revenue and support for renewable-energy integration.
The evidence is limited to mathematical modeling of one small-scale VPP configuration across four cases. It does not show that a physical VPP would achieve the reported profits or performance scores. No confidence intervals, statistical tests or uncertainty analysis were reported for the joint-VPP comparison.
That leaves open how the method would behave with control-signal data from the target market, different resource mixes, confidence levels, market rules or VPP scales, and whether the reported gains would persist in real-time or hardware-in-the-loop operation.
Metadata identifies the document as arXiv version 1 dated August 26, 2026. The work was supported by the THU-CSPG Digital Power Grid Joint Research Institute Project.
Paper data and sources
Original title: Integrating Fast-response Capability into Virtual Power Plant Operation for Ancillary Services
Authors: Qixing Liu, Ruike Lyu, Zhe Zhai et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: 10.1109/powertech59965.2025.11180557
Original paper · Full text