Development of an Efficient Electricity Consumption Prediction Model using Machine Learning Techniques

Abstract

Electricity consumption has continued to go up rapidly to follow the rapid growth of the economy. Therefore, detecting anomalies in buildings' energy data is considered one of the most essential techniques to detect anomalous events in buildings. This paper aims to optimize the electricity consumption in households by forecasting the consumption of these households and, consequently, identifying the anomalies. Further, as the used dataset is huge and published publicly, many research used part of it based on their needs. In this paper, the dataset is grouped as daily consumption and monthly consumption to compare the network topologies of all other works that used the same dataset with the selected part. The proposed methodology will depend basically on long short-term memory (LSTM) because it is powerful, flexible, and can deal with complex multi-dimensional time-series data. The results of the model can accurately predict the future consumption of individual households in a daily or monthly consumption base, even if the household was not included in the original training set. The proposed daily model achieves Root Mean Square Error (RMSE) value of 0.362 and mean absolute error (MAE) of 19.7%, while the monthly model achieves an RMSE value of 0.376 and MAE of 17.8%. Our model got the lowest accuracy result when compared with other compared network topologies. The lowest RMSE achieved from other topologies is 0.37 and the lowest MAE is 18% where our model achieved RMSE of 0.362 and MAE of 17.8%. Further, the model can detect the anomalies efficiently in both daily electricity consumption data and monthly electricity consumption data. However, the daily electricity consumption readings are way better to detect anomalies than the monthly electricity consumption readings because of the different picks that appear in the daily consumption data.