Nine Mile Point — LMP Price Forecasting Dashboard

A real-time data pipeline and forecasting model for electricity prices at the NINEMILE pricing node in Oswego, NY — the grid interconnection point for Nine Mile Point Nuclear Power Plant, operated by Constellation Energy.

Built as a portfolio project targeting quantitative analyst and risk analytics roles in energy trading.

---

What this is

This project pulls real hourly Locational Marginal Price (LMP) data from PJM's Data Miner API for the Nine Mile node, stores it in a local SQLite database, trains a price forecasting model, and displays everything in a live Streamlit dashboard.

LMP is the actual dollar price Constellation receives for each megawatt-hour generated at Nine Mile — the same number Constellation's commercial team in Baltimore watches and hedges against daily. This project models that price, analyzes congestion patterns, and forecasts the next 24 hours.

---

Project structure

ninemile-lmp-dashboard/
├── fetch_data.py     # Pulls real LMP data from PJM Data Miner API → SQLite
├── model.py          # Feature engineering + forecasting model training
├── dashboard.py      # Streamlit dashboard (price history, congestion, forecast)
├── .gitignore        # Excludes .env, database, and model files
└── README.md

---

Quick start

1. Install dependencies

pip install requests pandas numpy scikit-learn streamlit

2. Set up credentials

Create a .env file in the project folder:

PJM_USERNAME=your_pjm_username
PJM_PASSWORD=your_pjm_password
PJM_API_KEY=your_pjm_api_key

Getting access:

• Register at accountmanager.pjm.com — free, non-member account works
• Request PJM Public access during registration
• Find your API key in the Network tab of browser DevTools when using Data Miner

3. Fetch data

python fetch_data.py

Pulls 90 days of real hourly LMP data for the NINEMILE node (pnode_id: 1067164095) and saves to a local SQLite database.

4. Train the model

python model.py

Engineers features, compares Ridge Regression vs Gradient Boosting, saves the best model. Prints MAE, RMSE to terminal.

5. Launch the dashboard

streamlit run dashboard.py

Opens at http://localhost:8501

---

What the model does

Forecasts hourly LMP ($/MWh) for the next 24 hours using time-series features:

Time features

• Hour of day and month, encoded cyclically as sin/cos so hour 23 and hour 0 are treated as close together
• Day of week, weekend flag, business hours flag

Lag features

• LMP 1 hour ago, 24 hours ago, 168 hours ago (same hour last week)
• These are typically the strongest predictors in energy price time-series

Rolling statistics

• 24-hour rolling mean and standard deviation
• 7-day rolling mean

Models compared: Ridge Regression, Gradient Boosting Regressor
Train/test split: 80/20 chronological — older data trains, most recent 20% tests
Evaluation metrics: MAE and RMSE (in $/MWh)

Note: LMP forecasting is genuinely difficult. Prices spike unpredictably due to transmission events, weather, and generation outages. The model captures daily and weekly patterns well but will underestimate spikes — a known limitation of regression approaches on price data.

---

Dashboard features

• LMP price history — Total LMP vs System Energy Price, showing the spread caused by congestion
• Congestion analysis — Hours of positive vs negative congestion, worst congestion events
• 24-hour forecast — Model prediction overlaid on recent actuals
• Price shape — Average LMP by hour of day and day of week
• Model performance — Actual vs predicted chart with MAE and RMSE metrics

---

Why this matters — energy market context

Nine Mile Point (Units 1 & 2) and FitzPatrick Nuclear Power Plant are adjacent baseload generators in Oswego, NY, both operated by Constellation Energy. They run at full output around the clock and sell every megawatt-hour into the PJM real-time market at the LMP for their grid node.

LMP has three components:

• System energy price — the grid-wide clearing price, set by the marginal generator (usually a gas plant)
• Congestion price — positive when transmission lines have spare capacity, negative when they're constrained. Sustained negative congestion means Nine Mile earns less than the grid-wide price because the lines out of Oswego can't carry all the power being generated
• Marginal loss price — small adjustment for electrical losses over long distances

Why Constellation's Baltimore team cares:
The commercial/trading team hedges Nine Mile's exposure to LMP volatility using forward contracts, options, and Financial Transmission Rights (FTRs). Accurately forecasting LMP — and specifically congestion patterns at the Oswego node — informs decisions about how much output to hedge, at what price, and how far forward. This is the core function of the Risk Analytics and Quantitative Analyst teams.

---

Data source

PJM Data Miner API — api.pjm.com/api/v1/rt_hrl_lmps
Node: NINEMILE · pnode_id: 1067164095 · Zone: ATSI · Updated hourly by PJM

---

Planned improvements

• Add weather data (temperature) as a model feature — the single strongest driver of electricity demand and therefore LMP
• Pull day-ahead LMP alongside real-time to model the DA/RT spread
• Build a simple VaR (Value at Risk) calculator based on historical LMP volatility
• Deploy to Streamlit Cloud for a live public URL

---

Built by Dylan Pettinelli — portfolio project for quantitative analyst roles in energy trading