Assessing deep learning performance in power demand forecasting for smart grid
by Hengshuo Liang; Cheng Qian; Wei Yu; David Griffith; Nada Golmie
International Journal of Sensor Networks (IJSNET), Vol. 44, No. 1, 2024

Abstract: This paper addresses the issue of forecasting power demands via deep learning (DL) techniques in smart grid (SG). Assessing proper DL models for power demand forecasting requires the consideration of factors (e.g., data pre-processing, computational resource usage, the complexity of learning models). We employ a two-tiered approach to carry out both short-term and long-term forecasting. Short-term forecasting emphasises model accuracy, while long-term forecasting assesses model robustness. Our evaluations utilise temporal fusion transformers (TFT) and the neural hierarchical interpolation for time series (N-HiTS)-based predictors, tested on a publicly available dataset. Our findings confirm that while TFT and N-HiTS perform similarly in short-term forecasting tasks, TFT displays superior robustness and accuracy in long-term forecasting tasks. The TFT model requires substantial computational resources, especially video RAM (VRAM), for a longer input data stream. Conversely, N-HiTS, though less confident in long-term forecasting, is shown to be more resource-efficient for handling longer input data streams.

Online publication date: Tue, 30-Jan-2024

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Sensor Networks (IJSNET):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com