Short-Term Load Forecasting using Artificial Neural Networks and Multiple Linear Regression

Abstract

In this paper, two methods for short-term load forecasting are compared; namely, artificial neural networks (ANNs) and multiple linear regression (MLR). Only input features that had a very large correlation with the load were used. Historic load data are shown to have the strongest correlation with the current load data than other weather variables such as temperature and humidity. Simulation results show that the MLR give better results for the seasonal forecasts, whereas the ANN showed an overall lower mean absolute percentage error (MAPE) for the daily forecasts.