This is the authors' abstract. We don't add key points for this paper.
Robust forecasting of water-quality dynamics is a core requirement for intelligent decision-support systems in aquaponics, where multiple environmental variables interact nonlinearly and evolve over extended time horizons. Existing approaches are largely limited to univariate or short-term predictions, which fail to represent the coupled, long-range temporal behaviors. This study formulates water-quality estimation as a fully multivariate, multi-horizon forecasting problem, and introduces a model based on a temporal convolutional network (TCN) and a Bidirectional Gated Recurrent Unit (BiGRU) with attention (TCN-BiGRU-Attn) for multivariate, multi-horizon forecasting. The proposed hybrid architecture integrates dilated causal temporal convolutions for efficient modeling of long-range dependencies, bidirectional gated recurrent units for preserving sequential continuity, and a temporal self-attention mechanism that dynamically emphasizes informative historical states. Moreover, the multi-head forecasting design enables parameter-specific predictions while maintaining a shared latent temporal representation, allowing the model to capture common system dynamics and variable-specific behaviors jointly. To evaluate the performance of the proposed TCN-BiGRU-Attn architecture, extensive experiments are conducted across multiple datasets for aquaponic ponds, with temporal resolutions and forecasting horizons ranging from minutes to hours. The results demonstrate that TCN-BiGRU-Attn consistently and significantly outperforms state-of-the-art recurrent, convolutional, and hybrid baselines across all evaluated water-quality parameters. The proposed hybrid model achieves substantial reductions in error for long-horizon forecasts and chemically sensitive variables, while exhibiting markedly slower error growth as the prediction horizon increases. These empirical findings confirm the robustness, scalability, and practical value of the proposed TCN-BiGRU-Attn approach.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Saleh et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: