Toggle Main Menu Toggle Search

Open Access padlockePrints

Comparison of Neural Network Based Approaches for Short-term Residential Energy Load Forecasting

Lookup NU author(s): Dr Xiang XieORCiD


Full text for this publication is not currently held within this repository. Alternative links are provided below where available.


Neural network-based techniques have been used extensively by researchers and practitioners in the energy sector for the last few decades, to solve specific problems such as unit commitment, battery scheduling, energy trading, load forecasting. In particular, the energy load forecasting is vital to enhance the flexibility in energy management. However, a thorough and multi-faceted comparison of the performance of these neural network-based approaches on energy load forecasting is missing from the current literature. In this paper, four typical discrete-time neural network architectures, convolutional, vanilla feedforward, gated recurrent unit and long short-term memory neural network, as well as a novel continuous-time architecture are introduced to realize short-term energy load forecasting. These architectures are tested on benchmark datasets of electricity consumption data mainly from residential and other facilities. Their performances are evaluated based on three metrics, prediction accuracy, maximum error and computational efficiency. The results indicate that, although slower than a vanilla feedforward neural network, the gated recurrent unit architecture outperforms its competitors in terms of the accuracy. Furthermore, in the grid-level application scenarios where the coordination between facilities is necessary, neural ordinary differential equation enhanced autoencoder is preferable due to its continuous timeline property. Furthermore, the analysis shows that the incorporation of weather predictors and the level of aggregation determine the forecasting accuracy to a great extent, and thus should be taken into consideration wisely.

Publication metadata

Author(s): Tsianikas S, Xie X, Puri RS, Parlikad AK, Coit DW

Publication type: Conference Proceedings (inc. Abstract)

Publication status: Published

Conference Name: The IISE Annual Conference & Expo 2022

Year of Conference: 2022

Online publication date: 21/05/2022

Acceptance date: 10/01/2022

Date deposited: 11/04/2023

Publisher: Institute of Industrial and Systems Engineers, IISE


Library holdings: Search Newcastle University Library for this item

ISBN: 9781713858072