The Value of Forecasting: The Effect of Building Load Forecast Errors on the Performance of an Optimal Energy Management System

Distributed renewable resources and microgrids will be a crucial technology during the energy transition. Fully utilizing these to reduce consumer electric costs will often necessitate forecasting electric load, required to solve the unit commitment problem. Traditional forecasting methods and also modern machine learning algorithms are often trained and evaluated on mean squared error, which may not be the best choice for designing forecasts that work effectively with a downstream decision maker: an optimal energy management system. This novel work constructs several different forecasts with different error distributions to study the error impact on simulated optimal battery dispatch in a building microgrid. The findings are that forecast error is most important during peak load periods and peak price periods, especially with consumer demand charges in place, and a zero-bias or random Gaussian error distribution is necessarily advantageous.