2025 / ENS Partnership (course project)
Electricity Price Forecasting (ENS Partnership)
Time-series forecasting for day-ahead price spreads
Overview
Built a forecasting workflow for electricity price spreads with lag features, calendar effects, and model monitoring. Visuals are illustrative where original files are missing.
Problem
Day-ahead price spreads are volatile and require robust forecasting to inform hedging and operational decisions.
Role
Individual project
Timeline
Spring 2025
Tools
Python / pandas / XGBoost / LightGBM / statsmodels
Tags
Time Series / Forecasting / Energy / ML
Data
- - Hourly market prices with calendar + lag features
- - Feature sets include rolling stats, holiday flags, and price spreads
Approach
- - Built baseline seasonal naive + rolling average models
- - Trained gradient-boosted regressors with lag features
- - Added calibration checks and backtesting windows
Evaluation
- - Illustrative comparison of baselines vs boosted models (labeled)
- - Rolling-window MAE/MAPE tracking and error analysis by season
Results
- - Illustrative metrics show boosted models outperform baselines
- - Generated forecast intervals for operational planning
Deployment & Monitoring
- - Daily batch scoring with drift checks
- - Alerting when residuals exceed control limits
Limitations
- - Illustrative figures used where original assets are missing
- - Weather and demand signals could further improve accuracy
Gallery
Illustrative forecast chart (synthetic)
Forecasting pipeline diagram (concept)
Repro Steps
- - Synthetic visuals are labeled as illustrative
- - Pipeline description included for replication
Next Steps
- - Integrate weather forecasts and load predictions
- - Evaluate probabilistic forecasts with quantile loss
- - Deploy in a lightweight dashboard