Project 08
Forecasting monthly solar PV output across Calgary municipal installations using seasonal decomposition and XGBoost on 2,300+ production records
Streamlit application
Four pages covering production monitoring, multi-step forecasting, facility benchmarking, and return-on-investment analysis.
Production
Monthly kWh output trends per facility with seasonal pattern overlays and year-over-year comparisons
Forecast
Multi-step iterative predictions with confidence intervals for upcoming production months
Facility comparison
Side-by-side benchmarking of solar installations by capacity utilization and output efficiency
ROI
Return-on-investment calculator based on forecast output, energy rates, and installation costs
Key results
R-squared
0.86
XGBoost regressor on temporal holdout set
MAPE
~20%
Mean absolute percentage error across all facilities
Records
2.3K
Monthly production readings from municipal solar installations
Methodology
Fetched solar PV production and site data from Calgary Open Data via Socrata API. Engineered cyclical month encoding, rolling averages (3/6/12-month), and lag features per facility. Compared Ridge Regression, Random Forest, and XGBoost with temporal train-test split and iterative multi-step forecasting.