NARX Recurrent Neural Network Based Short Term Residential Load Forecasting Considering the Effects of Multiple Weather Features

As smart grids are expected to revolutionize the current electrical systems, short-term load forecasting (STLF) has emerged as a critical issue that must be solved before the smart grid's applications can be fulfilled. Recently, the rise of big data combined with machine learning has made neural network a viable solution for STLF. However, the impact of weather parameters for residential loads forecast is rarely investigated, despite the fact that it is an essential element that impacts power consumption patterns. In this paper, a Nonlinear Autoregressive with Exogenous (NARX) input recurrent neural network is designed for STLF while considering the effects of different weather features. Analysis based on correlation heatmaps has been carried out to determine the relations of the weather features with the load power consumption as well as with the other weather features. Accordingly, based on the correlation heatmaps, two weather features namely dew point and wind speed have been chosen for accurate predictions. Moreover, a prediction comparison has also been performed to demonstrate how high correlation among the weather features can lead to overfitting, noisy, and less accurate power consumption predictions. The proposed models' load power consumption performance is tested on 108 residential loads using the UMass Smart∗ Dataset.