IESO Coincident Peak Prediction — Backtesting & Financial Valuation

This notebook simulates operational deployment of the peak prediction model across multiple historical base periods and quantifies the financial value for a hypothetical 10 MW Class A customer.

Key questions:

• How accurately would the model have predicted peak days?
• How often would actual peaks fall within the predicted 3-hour window?
• What is the net financial value after accounting for false alarm costs?

Operational Simulation (2024 Base Period)

For each day in the test period, restrict information to what would be available at 7 AM and generate RED/YELLOW/GREEN alerts.

Confusion Matrix

Alert Calendar Visualization

3-Hour Window Evaluation

For correctly predicted peak days, determine how often the actual peak hour fell within the predicted 3-hour window.

Walk-Forward Backtest (2019–2024)

Retrain model using only data available prior to each base period, then predict that period's peaks. This simulates true operational conditions.

Financial Valuation

Quantify the dollar value of the prediction model for a hypothetical 10 MW Class A customer.