A new hybrid neural network based on gated recurrent unit and simple exponential smoothing for forecasting

Deep recurrent artificial neural networks can be easily adapted to the forecasting problem due to the dynamic structure in their architecture. In this paper, two different hybrid neural network architectures based on gated recurrent units and exponential smoothing are proposed for forecasting. The a...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Computing Jg. 107; H. 10; S. 196
Hauptverfasser: Kolemen, Emine, Egrioglu, Erol, Bas, Eren, Turkmen, Mustafa
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Vienna Springer Vienna 01.10.2025
Springer Nature B.V
Schlagworte:
Algorithms
Artificial Intelligence
Artificial neural networks
Computer Appl. in Administrative Data Processing
Computer Communication Networks
Computer Science
Deep learning
Evolutionary computation
Forecasting
Information Systems Applications (incl.Internet)
Machine learning
Neural networks
Particle swarm optimization
Regular Paper
Smoothing
Software Engineering
Stream flow
Time series
Traffic flow
Water resources
Wind power
Exponential smoothing
Gated recurrent unit
Differential evolution algorithm
Forecasting
Sustainable water resources
Particle swarm optimization algorithm
ISSN:0010-485X, 1436-5057
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Deep recurrent artificial neural networks can be easily adapted to the forecasting problem due to the dynamic structure in their architecture. In this paper, two different hybrid neural network architectures based on gated recurrent units and exponential smoothing are proposed for forecasting. The architectures combine gated recurrent units and structures inspired by exponential smoothing models. In addition, training algorithms based on differential evolution and particle swarm optimization algorithms are proposed separately for training. Different strategies are used to solve overfitting and local optimum problems in these training algorithms. The performance of the proposed method is applied to sustainable water resources. The analysis results show that the proposed deep neural network methods have superior forecasting performance than many artificial neural networks.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0010-485X
1436-5057
DOI:10.1007/s00607-025-01549-1