A comparative study on neural network-based prediction of smart community energy consumption

Document Type

Conference Proceeding

Publication Date



© 2017 IEEE. This paper presents a comparative study on designing accurate prediction of future energy consumption at both the household level and the community level. Different Neural Network (NN), including conventional NN, Deep Neural Networks (DNN), and Sliding Window Neural Networks (SWNN), are compared in this work, where a SWNN uses a window of historical data to predict the future energy consumption. Our experimental study shows that the conventional NN can achieve high accuracy in prediction while deep NN does not generate better results. Through data normalization and temporal relationship exploration, SWNN becomes superior to conventional methods and achieves above 99.5% accuracy with a more condensed error distribution.

Publication Title

2017 IEEE SmartWorld Ubiquitous Intelligence and Computing, Advanced and Trusted Computed, Scalable Computing and Communications, Cloud and Big Data Computing, Internet of People and Smart City Innovation, SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI 2017 - Conference Proceedings