A comparison of neural network-based methods for load forecasting with selected input candidates

Abstract

Accurate models for electric power load forecasting are essential to the operation and planning of the electricity in the energy management system(EMS). In the economic dispatch problem, load forecasting helps an electric utility to make important decisions including the battery charging schedule and the generation of electric power. This paper considers several methods for load forecasting based on the neural networks by using selected input candidates. The input variables are analyzed on the characteristic of correlation between weather data and an electrical load. We compare three neural network-based models for predicting the electrical load as follows : Feed Forward Neural Network model(FFNN), Recurrent Neural Network model(RNN) and Neural Network-based nonlinear autoregressive exogenous model(NARX). The subject of load forecasts is the building of the department of mechanical engineering in GIST, Gwangju, Republic of Korea in 2014. Under the proposed models, the predicted load data could be obtained from the selected input data. The simulation results are compared to each model and it shows that the predicted data is accurate and effective.