Abstract:
Global energy usage has increased due to industrialization and growing energy demand. Current studies revealed that residential building consumes a larger portion of energy generated. The recent deployment of smart meters in the residential sector has made it possible to acquire datasets for various analysis. Accessing the real-time electricity consumption of each household will enable both the distribution and the consumer side to extract useful information for effective energy management. Predicting energy usage will help in improving electricity generation plans and ensure efficient consumer energy
management. This study develops an energy consumption predictive model from
accumulated energy data for future energy forecast and management in residential buildings. A smart energy meter was deployed to retrieve and store the energy consumption of a particular building to enable proper energy management. The energy consumption was forecasted using three deep learning methods chosen from a review of several works on this topic: Convolutional Neural Networks (CNN), AutoRegressive Integrated Moving Average (ARIMA), and Long Short-Term Memory (LSTM). Energy prediction was implemented in two phases: hourly and daily prediction. Results indicate that the CNN model outperforms all other models by recording the smallest
Root Mean Square Error (RMSE) value of 0.66 compared to 0.75 and 0.82 for LSTM and ARIMA respectively. While LSTM model yields the most accurate result in predicting daily energy consumption by recording the smallest RMSE value of 0.082 compared to 0.86 and 0.086 for CNN and ARIMA respectively.
