Predicting city-scale daily electricity consumption using data-driven models

• We studied how city electricity use is influenced by weather and COVID-19 pandemic. Seven data-driven models were applied evaluated for data of three cities. Gradient boosting tree model delivers the most accurate prediction with CVRMSE 4%?6%. 1 °C increase ambient temperature drives up cities usage around 5%. curtailment reduced city-scale 2%?12%. Accurate demand forecasts that account impacts extreme events are needed to inform electric grid operation utility resource planning, as well enhance energy security resilience. Three common used predict daily usage: linear regression models, machine learning time series data, tabular data. In this study, we developed compared seven models: (1) five-parameter change-point model, (2) Heating/Cooling Degree Hour (3) decomposed implemented Facebook Prophet, (4) Boosting Machine Microsoft lightGBM, (5) widely-used (Random Forest, Support Vector Machine, Neural Network). metropolitan areas in United States: Sacramento, Los Angeles, New York. Results show can area's use, a coefficient variation root mean square error (CVRMSE) less than 10%. The lightGBM provides results, on test dataset 6.5% 4.6% 4.1% York area. These further explore (e.g., heat waves) unexpected public health pandemic) influence each city's demand. weather-sensitive component accounts 30%–50% total usage. Every degree Celsius summer leads about 5% (4.7% 6.2% 5.1% York) more base load areas. pandemic demand: pre-pandemic same months 2019, during 2020 decreased 10% April started rebound summer.