Long Short-Term Memory-Based Model Approach for Energy Consumption Prediction to Trigger High Consumption Alert

Abstract

In the modern era, electricity is crucial in daily life, as countless technologies require electric energy. This huge consumption of electrical energy leads to energy waste that has become a frequently discussed topic. Waste in energy consumption also leads to global warming and climate change to our world. Environmental pollution also increases with the consumption of electrical energy. This topic has received attention from the government and Tenaga Nasional Berhad (TNB), which is the largest electricity producer in Malaysia, and also has been emphasized in the Eleventh Malaysia Plan. Therefore, energy consumption analysis is needed to identify trends in electricity usage at a particular place and its diversity of users. From this analysis, models of energy consumption behavior will be formed based on the collected data which can be utilized for the prediction of daily energy consumption depending on the different places and users. In this project, a machine learning technique, specifically the Long Short-Term Memory (LSTM) method is employed to learn and predict energy consumption patterns. An energy consumption prediction model is built from the collected data and used for predicting future consumption behavior for different localities. The model’s predictive performance of the prediction is measured by using Mean Absolute Percentage Error (MAPE) and R-square Regression score. The main purpose of such energy usage analysis and prediction is to give alerts of unusually high consumption behavior so that precautions can be taken. When actual energy consumption is higher than the predicted at certain threshold value, the alert system will be triggered to alert the users that they have exceeded the common trend of their energy usage. Precautions taken based on the alerts and also predicted patterns can create awareness among users and hence helps to reduce the utility bill. The alert system triggered when actual consumption is higher than predicted and a simple message box popped up to inform the users that they have exceeded high energy consumption. The accuracy for both MAPE and R-square regression shows their best accuracy after the energy consumption prediction.