Knowledge Bank

Home energy management systems (HEMS) and other demand side management strategies are necessary for the establishment of stable and sustainable energy grid. HEMS schedule the timing of deferrable loads within homes by minimizing energy costs according to a time variable price for electricity while maintaining occupant comfort. Scheduling these loads establishes a challenging optimization problem, in which the controller must handle significant operational uncertainty, including error in weather forecasts and load requests from residents. To properly evaluate HEMS performance, it is necessary to do so using conditions that accurately mimic realworld forecast uncertainty. However, there is a lack of open access historical weather forecast data, leading researchers to rely upon synthetic weather forecasts.

This thesis aims to improve the accuracy of HEMS performance evaluation by generating synthetic weather forecasts with uncertainty that is more accurate to real-world conditions. Live weather forecast uncertainty was analyzed by recording data from an AccuWeather API and comparing to weather observations from the National Oceanic and Atmospheric Administration. A long-short term memory model was developed to generate synthetic weather forecasts with realistic uncertainty. This analysis showed that real-world weather forecast uncertainty is dependent upon time of day, location, and the lead time between a weather forecast and its prediction. While the model developed to generate synthetic weather forecast does not accurately match real-world uncertainty, it establishes a methodology which can be improved upon and implemented in future HEMS evaluation.