Load Forecasting in a Smart Grid Oriented Building

This research work has addressed demand side flexibility in a smart grid oriented building. The principal purpose has been to build a short term forecasting model that will predict the next hour consumption. Three advanced methods of forecasting have been investigated for this purpose, the ARIMA (Autoregressive Integrated Moving Average) model, Artificial Neural Networks (ANN) and Support Vector Machines (SVM). Historic time series of loads and consumption data of Østfold University College in Halden have been used for seeding and tested for trends, seasonality, cyclic characteristics and randomness. Accuracy for all three methods are fair and can be applied for the purpose. Priority is placed on ARIMA due to its transparency. This provides a simple form of explanation. The next hour predictions obtained will yield sufficient information and latitude to change operational strategies and move loads or substitute imported electricity with energy produced from local resources.