Load Demand Forecasting Using a Long-Short Term Memory Neural Network

Electric power load forecasting is very important for the operation and the planning of a utility company. Decisions of the electric market, electric power generation, load switching, and infrastructure development depend on load forecasting. There are many methods for load forecasting using statistical models, machine learning models and hybrid models. In this work, a Long-Short Term Memory Neural Network (LSTM NN) is used for short-term load forecasting, ranging from 1 h to 2 h ahead. Electric power demand time series provided by the National Center of Energy Control (CENACE) is used to train and validate the network. Results are compared with reported values from CENACE and Mean Absolute Percentage Errors (MAPEs) are calculated.