Deep Learning-Based Power Usage Forecast Modeling and Evaluation

The growing Internet of Things (IoT) provides significant resources to be integrated with critical infrastructures to enable cyber-physical systems. More specifically, the deployment of smart meters for electricity usage monitoring in the smart grid can provide granular and detailed information from which power load forecasting can be carried out. However, the accurate prediction of long-term power usage remains a challenging issue. In light of many recent advances, deep learning has the potential to significantly improve the ability to assess data and make predictions, and is already rapidly changing the world we live in. As such, in this paper, we consider the use of deep learning, via Recursive Neural Network (RNN) and Long Short-Term Memory layers, for the long-term prediction of localized power consumption. In particular, we consider the optimization of both data feature sets and neural network models, developing three model-feature combinations to maximize prediction accuracy and minimize error. Through detailed experimental evaluation, our results demonstrate the ability to achieve highly accurate predictions over periods as large as 21 days through the integration of correlated features. (C) 2019 The Authors. Published by Elsevier Ltd.