An Unbiased Climate-Informed Tool for Organizational Energy Policy Development and Budget Forecasting

The prospect of changing climate has driven facility owners, companies, market sectors, and governments to question how gradual trends in climate may affect asset resilience. Climate uncertainty poses a unique challenge for coupled human-natural systems; energy consumption is no exception. Managers should value skillful energy-use predictions at various time scales to mitigate the need to borrow funds from other operations and maintenance budget areas to offset periods where actual energy cost outpaces projections. Developing skillful forecasts with open-source data allows facility and campus managers to develop adaptive policies and realistic budgets for energy use. This research uses a statistical model-based approach to predict daily energy consumption for a medium-sized community (population: 26,500). The skill of several model configurations is tested using combinations of consumption periodicity, climate, and temporal state variables. The modeling framework consists of cross-validated principal component regression followed by post-prediction statistical bias correction. Ensemble predictions are validated using ranked probability skill score (RPSS), while deterministic prediction skill is measured by MAPE and contingency tables. Hindcasts explain more than 70% of variability in hourly energy consumption. The top-performing model achieves an RPSS of over 50% and is skillful year-round. A forward-looking model, based on the top-performing hindcast configuration, is informed by CMIP5 climate projections. The cumulation results suggest that incorporation of forward-looking projections could be useful for long-term utility policy development and budgetary planning.