Evolutionary Echo State Network Architecture Search for Time-Series Prediction
IEEE Access · 2026 · 14 · 124684-124702
DOI: 10.1109/access.2026.3722373 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Alok Yadav und 2 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: School of Information Technology, King Mongkut’s Institute of Technology Ladkrabang.
Abstract (Originalsprache)
Echo State Networks (ESNs) are efficient recurrent models for temporal dynamics, but their performance depends strongly on component choice, architecture, and hyperparameters, which are often selected manually or optimized separately. We propose Evolutionary ESN Architecture Search (EESNAS), a hybrid framework that uses a genetic algorithm for structural exploration and Bayesian optimization for local hyperparameter refinement. EESNAS is evaluated on six benchmark time-series tasks under one-step and autoregressive protocols against reservoir-computing, recurrent-network, and ablation baselines. It improves predictive accuracy and run-to-run stability on most datasets while remaining practical relative to exhaustive hyperparameter search. Statistical comparisons and ablations show that structural search and hyperparameter tuning provide complementary, dataset-dependent gains, supporting coordinated empirical adaptation of ESN topology, component choice, and local hyperparameters.
Förderung
- School of Information Technology, King Mongkut’s Institute of Technology Ladkrabang
Zitieren
APA
Yadav, A., Pasupa, K., & Loo, C. K. (2026). Evolutionary Echo State Network Architecture Search for Time-Series Prediction. IEEE Access, 14, 124684-124702. https://doi.org/10.1109/access.2026.3722373
Vancouver
Yadav A, Pasupa K, Loo CK. Evolutionary Echo State Network Architecture Search for Time-Series Prediction. IEEE Access. 2026;14:124684-124702. doi:10.1109/access.2026.3722373
BibTeX
@article{Yadav2026,
author = {Yadav, Alok and Pasupa, Kitsuchart and Loo, Chu Kiong},
title = {Evolutionary Echo State Network Architecture Search for Time-Series Prediction},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {124684--124702},
doi = {10.1109/access.2026.3722373}
}
EUVIMED