Total congestion in PJM increased by over $840 million in the first nine months of 2025 alone—reaching a total of $2.23 billion.
I decided to map this chaos. I analyzed every 5-minute interval, mapped the congestion costs to specific substations, and identified the Top 100 most congested lines in PJM.
The Discovery
While doing the mapping, I noticed something that surprised me: a huge percentage of the highest-cost constraints aren't the massive 345/500kV backbones, but relatively short segments. For the engineers in the room: we know DLR isn't a silver bullet for every constraint. On long-haul lines, you usually hit Surge Impedance Loading (SIL) or stability limits long before you sag the conductor to its limit. But on these short, thermally-limited segments, the ampacity is the only thing standing between a clear path and a $1,000/MWh price spike. This is where AAR and DLR have the highest "delta" and the biggest market impact.
The Challenge: Beating the Market
With FERC Order 881 now in effect, PJM has fully implemented Ambient Adjusted Ratings (AAR) into its congestion management. This means line limits now fluctuate hourly based on temperature.
My goal for Part 2: Can I use Gridraven Claw to predict these ratings for the Top 100 lines before PJM publishes them?
The Workflow
1. The Dataset: Focus on the 100 lines driving the highest congestion rent running in Claw.
2. The "Inference" Engine: Since exact conductor specs are rarely public, I’m using the brute-force tool from Part 1.1 to find the best-fit thermal parameters
3. Shadow Ratings: Running thermal models using only public geometry (OSM) and hyperlocal weather to generate a rating forecast up to 60 hours in advance.
Systematic Validation
By forecasting DLR/AAR for the Top 100 congested segments 24–36 hours ahead of the DA market clearing, we shift from reactive monitoring to predictive modeling of thermal constraints.
I am currently in the data-collection phase to establish high-confidence baselines for these specific line geometries. Once I have a statistically significant sample size, I will be conducting a formal back-test, benchmarking these "Blind Predictions" against the published PJM Day-Ahead ground truth.
