Short-Term Forecasting in Electric Power Systems using Artificial Neural Networks

In order to optimize the power flows within microgrids for high economic profitability, as general rule the energy management systems (EMSs) need as input real-time forecasts of time series that present different levels of seasonality and nonlinear correlations with exogenous variables (e.g. load, prices, energy generation). In this work, a fast and simple procedure that constructs real-time prediction intervals (PIs) for these signals is presented. PIs are constructed using artificial neural networks (ANNs) created through a modified lower-upper bound estimation (LUBE) method and trained with the simulated annealing (SA) algorithm. Focus is placed on explaining in detail the important steps of implementation. The effectiveness of the procedure is shown by creating PIs for the power demand of a transmission system operator (TSO). The resulting forecasting model is a centerpiece for the ongoing development of an application for EMSs within microgrids.