Multivariate Short-term Load Forecasting Using Deep Learning Algorithms

Abstract

Load forecasting is important in energy market. In fact electricity is a type of energy that cannot be stored, thus it is more important in electrical energy. The facilities need to balance between electricity generation and consumption by making plans. Computer-aided forecasting models are developed to reduce the effects of factors that disrupt this supply-demand balance. Generally, daily, weekly and monthly forecasts are made in demand forecast. In this study, hourly demand estimation is made. By using the past 24-hour consumption data and weather data such as temperature, humidity, wind speed and radiation in Konya, the next hour's consumption value was tried to forecast. Forecasting models were created using deep learning algorithms such as RNN, LSTM and GRU and the most successful model was determined by comparing the models.