Use of Convolutional Neural Networks and Long Short-Term Memory for Accurate Residential Energy Prediction

Abstract

With the deployment of smart meters on the residential level, consumers now possess more options for controlling the electrical consumption of their electrical appliances. So, consumers can better plan for and control how much electricity they use if they know how much electricity they use every day. Today's electrical systems must properly estimate consumer energy use, which can lead to a better understanding of the actual power consumption patterns that consumers experience. This paper addresses methodologies based on machine learning tools used to improve electrical system load forecasting by applying Long Short-Term Memory and Convolutional Neural Networks on a dataset containing 2 months, (i.e. from 1-1-2022 to 1-3-2022), of six-second regularly spaced measurement samples obtained from a lab designed smart meter placed in a residential house. This study also looks at how well the proposed LSTM-CNN model can predict home consumption based on data from two months.