Very short-term load forecasting based on NARX recurrent neural networks

Time series forecasting is an important task in various fields of science, like economy, engineering and other areas that use historical data to predict future problems. In this context, Artificial Neural Networks have shown promising results for this task, when compared with the traditional statistical techniques. Thus, this research aims to evaluate the performance of NARX-neural network (Nonlinear Autoregressive Model with Exogenous Input) for the purpose of performing load forecasting for very short-term data from distribution substations. The cross validation was applied to evaluate different topologies. It is important to mention that the data was obtained by measures done in Brazilian substations located at two different cities. The results show the contribution of the paper once it demonstrates the efficiency of the NARX-neural network compared with Feedforward and Elman neural networks, which are widely used to predict times series.