"Hourly Electricity Demand and Weather Data for Major U.S. Cities (2019\u20132023)"

"This dataset provides a curated, multi-regional hourly electricity demand and weather resource for 141 U.S. metropolitan areas spanning five years (2019\u20132023), constructed to directly support AI and machine learning model development in power and energy systems. Each of the 141 city files synchronizes three independently sourced data streams \u2014 county-level megawatt demand from the NREL\/OEDI Hourly Electricity Demand dataset (Obika et al., 2025), ground-truth meteorological observations from the NOAA Integrated Surface Database, and satellite solar irradiance from the NREL National Solar Radiation Database (NSRDB) \u2014 into a single analysis-ready time series of 43,824 hourly records and 22 columns.Physics-informed features are pre-engineered into each file, including apparent temperature (heat index), 24-hour thermal memory, cooling and heating degree days, wind speed extrapolated to 100-meter turbine hub height via the Hellman power law, behind-the-meter rooftop solar proxy, 3-hour solar volatility, and cyclical encodings of hour, day-of-week, and month.Of the 141 cities, 101 are pristine (zero missing values, zero imputation) and 40 are intentionally quarantined with documented data quality issues, providing a realistic benchmark set for imputation and denoising research. Dataset utility is validated through five distinct machine learning experiments conducted across all 101 pristine cities on a held-out 2023 test year:- Autoregressive Load Forecasting: ARX-LSTM achieved a mean R\u00b2=0.993 and MAE=23.66 MW, establishing a strong baseline for sequence-to-sequence grid demand prediction.- Thermodynamic Load Forecasting: Weather-only LSTM (R\u00b2=0.853, MAE=164.91 MW) isolates the contribution of meteorological drivers without autoregressive feedback.- Baseline Benchmarking: Random Forest load forecasting (R\u00b2=0.864, MAE=159.94 MW) provides interpretable feature importance rankings that reveal dominant climate drivers \u2014 thermal inertia in humid climates versus solar irradiance in desert regions.- Grid Volatility: Grid ramp rate forecasting via LSTM (R\u00b2=0.853, MAE=28.41 MW\/h) supports flexibility and storage dispatch planning.- Renewable Potential: Solar GHI forecasting via LSTM (R\u00b2=0.881, MAE=46.56 W\/m\u00b2) enables renewable integration and behind-the-meter solar estimation.Together, these experiments confirm that the dataset supports load forecasting, renewable energy forecasting, grid flexibility analysis, feature engineering research, and climate-driven demand modeling across diverse U.S. regional grid conditions."