Short-Term Electrical Peak Demand Forecasting in a Large Government Building Using Artificial Neural Networks

Abstract

The power output capacity of a local electrical utility is dictated by its customers’ cumulative peak-demand electrical consumption. Most electrical utilities in the United States maintain peak-power generation capacity by charging for end-use peak electrical demand; thirty to seventy percent of an electric utility’s bill. To reduce peak demand, a real-time energy monitoring system was designed, developed, and implemented for a large government building. Data logging, combined with an application of artificial neural networks (ANNs), provides short-term electrical load forecasting data for controlled peak demand. The ANN model was tested against other forecasting methods including simple moving average (SMA), linear regression, and multivariate adaptive regression splines (MARSplines) and was effective at forecasting peak building electrical demand in a large government building sixty minutes into the future. The ANN model presented here outperformed the other forecasting methods tested with a mean absolute percentage error (MAPE) of 3.9% as compared to the SMA, linear regression, and MARSplines MAPEs of 7.7%, 17.3%, and 7.0% respectively. Additionally, the ANN model realized an absolute maximum error (AME) of 8.2% as compared to the SMA, linear regression, and MARSplines AMEs of 26.2%, 45.1%, and 22.5% respectively.

Keywords

• neural networks
• energy forecasting
• building management systems
• data logging
• smart grid
• MARSplines
• demand response