Evaluation of Forecasting Methods for Very Small-Scale Networks

Increased levels of electrification of home appliances, heating and transportation are bringing new challenges for the smart grid, as energy supply sources need to be managed more efficiently. In order to minimize production costs, reduce the impact on the environment, and optimize electricity pricing, producers need to be able to accurately estimate their customers’ demand. As a result, forecasting electricity usage plays an important role in smart grids since it enables matching supply with demand, and thus minimize energy waste. Forecasting is becoming increasingly important in very small-scale power networks, also known as microgrids, as these systems should be able to operate autonomously, in islanded mode. The aim of this paper is to evaluate the efficiency of several forecasting methods in such very small networks. We evaluate artificial neural networks (ANN), wavelet neural networks (WNN), auto-regressive moving-average (ARMA), multi-regression (MR) and auto-regressive multi-regression (ARMR) on an aggregate of 30 houses, which emulates the demand of a rural isolated microgrid. Finally, we empirically show that for this problem ANN is the most efficient technique for predicting the following day’s demand.