Prediction of Energy Consumption Peak in Household by using LSTM & MLP

Abstract

Accurate peak load forecasting plays a key role in operation and planning of electrical power generation. To minimize the operating cost, electric suppliers use forecasted peak load to control the number of running generator units. One of the most precise load forecasting methods is deep neural networks (DNNs), which is categorized under artificial neural networks (ANNs). In the past few decades, DNNs have appeared as a powerful tool in machine learning filed. DNNs have been shown to significantly outperform the other traditional methods in many applications, and they have completely revolutionized some fields. Given their success in other machine learning problems, DNNs are applied in energy forecasting. ANN has recently applied on short-term load forecasting in electrical utilities. In this thesis, two ANN algorithms for predicting peak load has been used. Multilayer Perceptron and Long Short-Term Memory. Then, the performance of the models was compared to find out the error in peak load forecasting. Error here refers to the difference between actual loads and predicted ones. The result based on in our study revealed that Long Short-Term Memory has less mean absolute percentage error (MAPE) in compare with Multilayer Perceptron.