Prediction of energy consumption using recurrent neural networks (RNN) and nonlinear autoregressive neural network with external input (NARX)

Recurrent Neural Networks (RNN) and Nonlinear Autoregressive Neural Network with External Input (NARX) are recently applied in predicting energy consumption. Energy consumption prediction for depth analysis of how electrical energy consumption is managed on Tower 2 Engineering Building is critical in order to reduce the energy usage and the operational cost. Prediction of energy consumption in this building will bring great benefits to the Faculty of Electrical Engineering UiTM Shah Alam. In this work, we present the comparative study on the performance of prediction of energy consumption in Tower 2 Engineering Building using RNN and NARX method. The model of RNN and NARX are trained using data collected using smart meters installed inside the building. The results after training and testing using RNN and NARX show that by using the recorded data we can accurately predict the energy consumption in the building. We also show that RNN model trained with normalized data performs better than NARX model.