Skip to content

Qian Zhang

Ph.D. Candidate, Electrical EngineeringHarvard John A. Paulson School of Engineering and Applied Sciences

Deliverable Flexibility, live

Back to Research

Dispatch a ten-generator fleet through a 4-hour, 600 MW ramp event and see whether the system prepared enough flexibility to avoid load shedding.

Dispatch

Net load and output MW

Net load Output Load shed
Net load, output served and load shed, minutes 0 to 240 Cheapest-first: load is shed from minute 130 to minute 175, at most 20.2 MW at minute 160.

Net load 1,870 MW · shed 0 MW · 60-min margin −5.1 MW · first warning

Ramp time left per generator

Cheapest-first
  1. G4 $15/MWh 0 min
  2. G6 $19.5/MWh 0 min
  3. G5 $20/MWh 0 min
  4. G2 $30/MWh 0 min
  5. G7 $48/MWh 0 min
  6. G10 $50/MWh 112 min
  7. G3 $55/MWh 52 min
  8. G9 $57/MWh 260 min
  9. G8 $60/MWh 50 min
  10. G1 $185/MWh 15 min

Ramp screen MW

At this minute

Ramp capability against requirement, by minutes ahead Cheapest-first, minute 85: the fleet is short for 60–85 minutes ahead.
Capability Requirement Short 60–85 min ahead

Over the day

The 60-minute ramp margin, minutes 0 to 180 Cheapest-first: the 60-minute margin's lowest point over the day is −20.2 MW.
60-min margin Below zero

Scoreboard same ramp, every method

Load shed and cost of each dispatch on this ramp; the run drawn above is marked current
Dispatch Load shed, MWh Production, $ Total at $10,000/MWh shed First shed
Cheapest-first 9.862 221,774.96 320,395.79 min 130
Look-ahead, 10 minmulti-interval 7.567 221,676.39 297,344.75 min 135
Ramp product, 10 minreserve ahead 9.862 221,774.96 320,395.79 min 130
Duration-aware, 40–100 minour method 0 223,071.66 223,071.66 none
Perfect foresighta bound 0 218,806.71 218,806.71 none

Cheapest-first: 9.862 MWh shed from minute 130, 45 min after the first warning.

About this case sources, method notes, caveats

How to read it

Ramp time left: how long each generator can keep ramping up at full rate before it reaches capacity; a bar is full at 120 minutes.

Ramp screen: capability, the ramp the fleet can deliver within the next k minutes, against requirement, the rise in load it must cover by then; crimson where it falls short. The 60-min margin is capability minus requirement at k = 60.

First warning: the first minute the screen finds the fleet short within the next two hours.

Sources

Fleet and look-ahead dispatch: Q. Zhang, L. Xie, L. Zhao, C. Wang, Comparative Assessment of Look-Ahead Economic Dispatch and Ramp Products for Grid Flexibility, PSCC 2026.

The ramp, the screen, the ramp product and duration-aware dispatch: the code of Q. Zhang, A. Looney, C. Tian, X. A. Sun, L. Xie, Duration-Aware Ramp Adequacy Screening, working paper.

At scale, quoted from that paper: on a 2,751-bus synthetic Texas grid, 96 fifteen-minute periods, switching duration-aware dispatch on for 16:00–20:00 cuts shed by 68%, from 2,275 to 723 MWh, for 2.9% more production cost.

Caveats

  • Perfect forecast: every method sees the true net load.
  • Look-ahead and ramp-product days are solved offline with HiGHS; cheapest-first and duration-aware dispatch run live in your browser.
  • Duration-aware windows are fixed in advance from the screen. Switched on all day, it needs no timing and still sheds nothing, for 7.9% more production cost than perfect foresight.
  • A longer product is not always better: products of 25 minutes or less, and of 95 minutes or more, change nothing here, and 50 minutes sheds more than 45 and 75 more than 70.
  • Only an enhanced product, holding every duration up to the window, matches look-ahead dispatch’s feasible region, and only at one step.
  • Load shed is priced at $10,000/MWh here; the PSCC paper’s look-ahead dispatch prices it at $2,500/MWh.