Sub-Hourly Load Forecasting for Community-Level Flexible Appliance Management

Menos-Aikateriniadis, Christoforos; Akarepis, Andreas; Kokos, Isidoros; Georgilakis, Pavlos

dc.contributor.author	Menos-Aikateriniadis, Christoforos
dc.contributor.author	Akarepis, Andreas
dc.contributor.author	Kokos, Isidoros
dc.contributor.author	Georgilakis, Pavlos
dc.date.accessioned	2024-09-13T13:52:42Z
dc.date.available	2024-09-13T13:52:42Z
dc.identifier.issn	2161-4393	el
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/60212
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.27908
dc.rights	Αναφορά Δημιουργού-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nd/3.0/gr/	*
dc.subject	Very Short-Term Load Forecasting	en
dc.subject	LSTM	en
dc.subject	Prophet	en
dc.subject	N-BEATS	en
dc.subject	Aggregate Flexible Load	en
dc.title	Sub-Hourly Load Forecasting for Community-Level Flexible Appliance Management	en
heal.type	conferenceItem
heal.classification	Machine Learning	el
heal.classification	Smart Grids	en
heal.classification	Forecasting	en
heal.classification	Energy flexibility	en
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2024-09-09
heal.bibliographicCitation	C. Menos-Aikateriniadis, A. Akarepis, I. Kokos and P. S. Georgilakis, "Sub-Hourly Load Forecasting for Community-Level Flexible Appliance Management," 2024 International Joint Conference on Neural Networks (IJCNN), Yokohama, Japan, 2024, pp. 1-8	en
heal.abstract	Very Short-Term Load Forecasting (VSTLF) at a residential level constitutes a challenge mainly due to the highly volatile end-user consumption patterns as well as data integrity issues that jeopardize the quality of collected data. VSTLF can enable improved flexibility quantification and more efficient residential Demand-Side Management (DSM) for load aggregators, not only on a consumer level but also on an aggregated appliance level. Inspired by the ongoing research debate between statistical-based and data-driven models for VSTLF, this work investigates the performance of a Long Short-Term Memory (LSTM) model with Feed-Forward Error Correction (FFEC), the widely-used XGBoost method and the State-of-the-Art (SoA) models N-BEATS and Prophet. Models were trained and tested on 15-minute aggregate Electric Vehicle (EV) and Air Conditioning (A/C) loads, using real measurements by the Pecan Street dataset. In contrast to common research practices, this work considers solely temporal features and historical consumption for training the models. Results indicate that firstly LSTM with FFEC and secondly XGBoost models capture more accurately sub-hourly load dependencies for both EVs and A/Cs, showcasing their ability to generalize efficiently across different appliances and sparse datasets. N-BEATS model performs adequately well but it cannot outperform the aforementioned models, while Prophet fails to capture very short term dependencies in the data.	en
heal.sponsor	This project has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska Curie grant agreement No 955422.	en
heal.publisher	IEEE	en
heal.fullTextAvailability	false
heal.conferenceName	2024 International Joint Conference on Neural Networks (IJCNN)	en
heal.conferenceItemType	full paper
dc.identifier.doi	10.1109/IJCNN60899.2024.10649928	el