Transformer-Based Time Series Prediction for Early Detection of Environmental Uncertainty in Marine Aquaculture
Advances in Nonlinear Variational Inequalities · 02.02.2026 · 29 (1s) · 47-73
DOI: 10.52783/anvi.v29.7256 ↗Kurz gefasst
Fachartikel in der Zeitschrift Advances in Nonlinear Variational Inequalities, online erschienen am 02.02.2026. Verfasst von Vu Ngoc Kien.
Abstract (Originalsprache)
Marine aquaculture systems are highly sensitive to environmental fluctuations, necessitating advanced predictive models for early anomaly detection and management. Environmental uncertainty remains a critical challenge in marine aquaculture, where fluctuations in water quality parameters can trigger disease outbreaks, reduce productivity, and disrupt ecosystem stability. Early detection of such uncertainties is essential for enabling proactive management and ensuring sustainable aquaculture operations. This research proposes a Transformer-based time series prediction framework for the early detection of environmental anomalies in marine aquaculture systems. The proposed framework leverages Marine Aquaculture Risk Time Series data collected from Internet of Things (IoT)-enabled sensor networks deployed across aquaculture farms, monitoring key environmental variables and it contains 4000 samples, followed by preprocessing steps including noise filtering, missing value imputation, and normalization. The data are fed into a Gases Brownian Motion Optimized Temporal Fusion Transformer (GBMO-TFT), GBMO optimizes Transformer hyperparameters for stable convergence and improved prediction, while TFT learns long-term temporal dependencies. The proposed framework is implemented using Python. Experimental results demonstrate that the Transformer-based model outperforms traditional approaches, achieving an accuracy of 97.56%, precision of 97.78%, and F1-score of 97.65%. Furthermore, the system enables forecasting and supports the generation of spatial risk maps for identifying high-risk zones within aquaculture environments. The proposed framework provides a scalable and intelligent decision-support tool for early warning and adaptive management in marine aquaculture. By integrating advanced Deep Learning (DL) with IoT-based monitoring, this research contributes to the development of sustainable, data-driven, and resilient aquaculture systems.
Zitieren
APA
Kien, V. N. (2026). Transformer-Based Time Series Prediction for Early Detection of Environmental Uncertainty in Marine Aquaculture. Advances in Nonlinear Variational Inequalities, 29(1s), 47-73. https://doi.org/10.52783/anvi.v29.7256
Vancouver
Kien VN. Transformer-Based Time Series Prediction for Early Detection of Environmental Uncertainty in Marine Aquaculture. Advances in Nonlinear Variational Inequalities. 2026;29(1s):47-73. doi:10.52783/anvi.v29.7256
BibTeX
@article{Kien2026,
author = {Kien, Vu Ngoc},
title = {Transformer-Based Time Series Prediction for Early Detection of Environmental Uncertainty in Marine Aquaculture},
journal = {Advances in Nonlinear Variational Inequalities},
year = {2026},
volume = {29},
number = {1s},
pages = {47--73},
doi = {10.52783/anvi.v29.7256}
}
EUVIMED