Composite Multi-Directional LSTM for Accurate Prediction of Energy Consumption

Abstract

Massive amount of electronic devices usage implies the total energy consumption exponentially. In order to avoid energy depletion issues, an energy prediction scheme should be considered. Previous studies only evaluated the model based on limited performance metrics and left the model robustness behind. In this paper, a novel deep learning (DL) based long short-term memory (LSTM) algorithm is proposed to deal with accurate energy consumption prediction. The proposed CMDLSTM is formed with two groups equipped with bidirectional LSTM (BiLSTM) and LSTM that are able to learn from multi-direction. The proposed model is attached with a dropout layer to avoid overfitting and ReLU as an activation function to extract the feature data. Based on the open-source dataset we used to evaluate the model in various performance metrics, CMDLSTM can accurately predict the energy consumption data compared with the existing DL models.