README.md

metadata

title: Time Series Forecasting - ERCOT Electricity Market
emoji: ⚡
colorFrom: blue
colorTo: purple
sdk: streamlit
sdk_version: 1.52.1
app_file: app.py
pinned: false
python_version: 3.11

Time Series Forecasting Application

Zero-shot time series forecasting application for ERCOT electricity market data using state-of-the-art pretrained models.

Features

• ⚡ Live ERCOT Data: Fetches real-time electricity price data from ERCOT Day-Ahead Market (180+ days)
• 🤖 Multiple Models: Choose from 7 pretrained forecasting models:
  • Chronos-2 (46M - 120M parameters) - Amazon's latest models
  • Chronos-T5 (8M - 710M parameters) - Original Chronos family
  • TiRex (35M parameters) - NX-AI's xLSTM-based model
• 📊 Backtesting: Automatic train/test split with performance metrics (MAE, RMSE, MAPE)
• 📈 Interactive Visualization: Historical context, actual values, and forecasts with date-based axes
• 🎯 Zero-Shot Forecasting: No training required - models work out-of-the-box
• 💻 Easy-to-Use Interface: Built with Streamlit for intuitive interaction

Usage

1. Select a forecasting model from the dropdown
2. Choose data source (ERCOT or sample data)
3. Set the forecast horizon (number of time steps)
4. View backtesting results with error metrics and comparison plots

Models

Chronos-2

Amazon's latest time series foundation models offering state-of-the-art zero-shot forecasting performance.

Chronos-T5

Original Chronos family based on T5 architecture, available in multiple sizes for different accuracy/speed tradeoffs.

TiRex

NX-AI's xLSTM-based model optimized for both short and long-term forecasting with excellent benchmark performance.

Data Source

• ERCOT: Day-Ahead Market Settlement Point Prices (SPP) from the Electric Reliability Council of Texas
• Sample Data: Synthetic electricity price data for testing

Links

• Chronos Forecasting
• TiRex Model
• ERCOT Data