Prediction of Electric Energy Consumption using Recurrent Neural Networks

Abstract

The prediction of power consumption is a complex and important task for a smart home, a city, and a country. However, deep learning plays a substantial role in predicting electric energy consumption more efficiently. Recurrent Neural Network (RNN) of deep learning is well capable of handling time-series datasets and predicting the electricity consumption better than the machine learning approaches such as ARMA, SVR. In this work, we use a large electricity consumption dataset of Dominion Virginia Power (DOM) using the proposed RNN approach to identify the hidden patterns of the dataset as well as predict electric power consumption. The results from the proposed approach are compared with the above-mentioned approaches to validate the performance to unveil the hidden patterns and predict the consumption behaviors. The accuracy is varied from 94.02% to 96.86% based on the number of epochs. Also, the error matrices like MSE, RMSE, MAE, and MAPE are demonstrated to validate its robustness for the prediction of electricity consumption. understanding the demand for electric power.