An Artificial Neural Network in Short Term Electrical Load Forecasting of a University Campus: A Case Study
View/ Open
Date
2013Author
Palchak, David
Suryanarayanan, Siddharth
Zimmerle, Daniel
Metadata
Show full item recordAbstract
This paper presents an artificial neural network (ANN) for forecasting the shortterm electrical load of a university campus using real historical data from Colorado State University. A spatiotemporal ANN model with multiple weather variables as well as time identifiers, such as day of week and time of day, are used as inputs to the network presented. The choice of the number of hidden neurons in the network is made using statistical information and taking into account the point of diminishing returns. The performance of this ANN is quantified using three error metrics: the mean average percent error; the error in the ability to predict the occurrence of the daily peak hour; and the difference in electrical energy consumption between the predicted and the actual values in a 24h period. These error measures provide a good indication of the constraints and applicability of these predictions. In the presence of some enabling technologies such as energy storage, rescheduling of noncritical loads, and availability of time of use (ToU) pricing, the possible demandside management options that could stem from an accurate prediction of energy consumption of a campus include the identification of anomalous events as well the management of usage.