PJM Interconnection
PJM demand forecasts. 1.7% MAPE.
56.4% lower error than the ISO baseline. Direct PJM Data Miner 2 integration for load, generation, and market signals across the world's largest wholesale electricity market.
Why PJM is different
The largest U.S. grid, with capacity risk priced in.
PJM serves 65M customers across 13 states and Washington D.C. Peak load is around 180 GW — roughly 30% larger than ERCOT. The footprint runs from Chicago (ComEd) through Pennsylvania, New Jersey, Maryland, Virginia, and into the Carolinas (Dominion), which means a single day-ahead forecast has to absorb very different load shapes: old industrial Midwest, dense East Coast residential, and the fastest-growing data-center footprint in the country.
The capacity market is what makes accuracy financially directly measurable. RPM auctions clear on a one-year-ahead basis using PJM's own load forecast; a load-serving entity that over-procures capacity pays for reserves it doesn't need, and one that under-procures faces deficiency charges. One percentage point of MAPE on the planning-year peak forecast moves real capacity position by hundreds of MW and directly affects clearing outcomes in the Base Residual Auction.
Zonal LMP is the second source of forecast leverage. ComEd, PECO, Dominion, AEP, BGE, and the other PJM zones settle at different prices — a trader or scheduler needs zone-aware forecasts, not just an RTO total. Data center growth in Northern Virginia has made the Dominion zone the most talked-about forecasting problem in the country right now, and it's where our monthly re-training shows up most visibly.
If you already model nodal or zonal price, think of Gramm as a sharper load input to whatever LMP, congestion, or capacity model you run — not a replacement for the market data you're already paying for. PJM's own day-ahead load forecast is what most price models carry today; swapping in a 1.70% MAPE input where the ISO baseline is 3.90% tightens the downstream price estimate without changing anything else in the pipeline.
Why PJM
The largest wholesale market deserves the best forecasts.
A 2.20 percentage point MAPE reduction over the ISO baseline, measured across 8,760 hours. PJM serves 65 million people across 13 states — accuracy at this scale matters.
5-minute instantaneous load, hourly verified metered load, and generation by fuel type pulled directly from PJM’s Data Miner 2 API — not through third-party scrapers.
Hourly generation data for gas, nuclear, coal, hydro, solar, and wind across PJM. Track thermal fleet dispatch patterns and renewable penetration in real time.
Benchmark
PJM benchmark details.
Full-year evaluation against PJM's own published day-ahead forecast. Every hour, every season, no cherry-picking.
Worked example
Translating MAPE to capacity-auction exposure on PJM
An LSE serving 8 GW of peak load in the Dominion zone. PJM's published DA baseline runs at ~3.9% MAPE; our evaluation shows Gramm at 1.70% MAPE — a 2.2 pp reduction. For RPM capacity auctions, even a small MAPE delta on the planning-year peak translates directly to capacity position dollars:
Illustrative for an 8 GW peak LSE in PJM. Capacity exposure scales with RPM clearing prices, which vary by auction and by zone. Exact impact depends on whether the forecast delta translates to over- or under-procurement, and on your hedging position. Zone-level forecasts for the full capacity delta are available on Enterprise.
Use cases
PJM market participants.
Capacity market planning
PJM’s capacity market (RPM) depends on accurate load forecasts for reserve margin calculations. Better demand forecasts improve your capacity position sizing and reduce unforced capacity exposure.
LMP-aware scheduling
Combine demand forecasts with generation mix signals to anticipate congestion patterns. Know when gas ramp-up drives LMP spikes at your node before the real-time market settles.
Multi-zone portfolio management
PJM spans 13 states with distinct load profiles. Our forecasts cover the full RTO with zone-level granularity, so you can optimize across ComEd, PECO, Dominion, and AEP simultaneously.
Quick start
One request. Full day-ahead forecast.
Most teams integrate in under an hour. Bearer token auth, JSON response, prediction intervals included.
View full API docsFAQ
Questions PJM participants actually ask.
Do you forecast the RPM planning-year peak used for capacity auctions?
Yes. We provide forward-looking peak-hour forecasts aligned to PJM's planning year (June-May) for the RPM Base Residual Auction and Incremental Auctions. These are available on Enterprise given the auction-sensitivity of the data.
Which zones are covered with zone-level granularity?
All 20 PJM zones and transmission zones: AE, AEP, APS, ATSI, BGE, COMED, DAY, DEOK, DOM, DPL, DUQ, EKPC, JCPL, METED, PENELEC, PEPCO, PPL, PSEG, RECO, and the aggregated RTO totals. Zone-level access is on Enterprise; system-total is on all plans.
How do you account for the Virginia / Dominion data center load growth?
We re-train monthly and our model weights shift based on the most recent 24 months of actual data, which captures the data-center buildout as it happens. For Enterprise customers we can also incorporate customer-provided forward load scenarios (new customer connections, retail migration) as explicit inputs rather than having to infer them from trend.
Can you deliver forecasts aligned to PJM's DA bidding window?
Yes. Forecasts are available by 10:30 AM EPT on the day before operating day, ahead of the 11:00 AM PJM DA market bid deadline. Forecasts can be re-pulled via API or pushed via webhook or SFTP on Enterprise.
Do you integrate with PJM Data Miner 2 authentication?
We pull from Data Miner 2 on our own credentials; our API abstracts over DM2 entirely. You don't need a Data Miner 2 account or key to use Gramm. If you want DM2-aligned field naming and CSV formats for compatibility with existing pipelines, that's available as a response format option.
How does the forecast behave during a summer capacity emergency?
Honestly, worse than in steady-state hours — everyone's forecast does. What differentiates our model (per the state space paper) is that it degrades gracefully rather than catastrophically. During the July 2024 heat wave hours when the PJM baseline forecast spiked to 5-6% MAPE, our backtested error stayed in the 2-3% range. That's still worse than our average, but it's the range that matters for load-serving entities sizing their capacity position.
Get started
Start evaluating PJM forecasts.
Free plan includes 100 req/hr across all 7 grids. No credit card required.