Solar-Aware Machine Learning Framework for Short-Term Electricity Demand Forecasting in Delhi Using Weather and Calendar Variables
Aditya Sridhar · 3rd International Conference on Signal Processing and Computer Vision (SIPCOV-2026), NIT Puducherry
Short-term forecasting of Delhi's electricity demand at 5-minute resolution, over 18 months of data, using only weather and calendar variables.
Rooftop solar output is not measured, yet it shapes the daytime dip and evening ramp of the duck curve. A solar-availability proxy lets the models capture that shape without any PV data.
Six models are benchmarked under leakage-free, time-aware validation. The best reached 0.77% MAPE and R² 0.9977, and the tabular learners beat the LSTM.
- Data
- 18 months, 5-minute load, Delhi
- Models compared
- 6
- Best MAPE
- 0.77%
- Best R²
- 0.9977