Short-term electric load forecasting using neural networks: A comparative study

Abstract

Forecasting the short-term electrical demands of a distribution system is essential for efficient electric system planning and operation. A balance between generation and demand is the main requirement for efficient operation by the electric utilities. Therefore, a load forecast as accurately as possible is important to offer a good service with adequate economic viability. In this work, four types of short-term load forecasting based on neural networks are proposed and compared. A Nonlinear autoregressive neural network (NAR), a Nonlinear autoregressive neural network with external input (NARX), a Feedforward neural network (Feedforward), and a Time delay neural network (Timedelay) are modeled and simulated. A comparison among proposed neural networks is done using metrics established in the evaluation of neural networks, such as Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and Correlation Coefficient (R). The data used in this work was extracted from the Operating Center of a local company of distribution of electric energy. These data are pretreated using curve smoothing techniques, and then, they are used to train the selected neural networks. Moreover, a simulation using the Prophet, a Facebook® procedure to predict time series, was made to compare the results with the mentioned neural networks.