Optimizing irrigation decisions with Seq2Seq modeling and deep reinforcement learning
Computers and Electronics in Agriculture · 2026-04 · 245 · 111448
DOI: 10.1016/j.compag.2026.111448 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computers and Electronics in Agriculture, 2026-04. Verfasst von Adityo Julian Tri Wibowo Mukti und 2 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 68 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This study integrates advanced soil moisture modeling with deep reinforcement learning (DRL) to develop a data-driven irrigation management system tailored for precision agriculture. Among six forecasters included in this study (LSTM, BiLSTM, LSTM-CNN, LSTM/BiLSTM with attention, Seq2Seq), Seq2Seq achieved the best accuracy (average MSE = 0 . 276 , RMSE = 0 . 500 , MAE = 0 . 337 , R 2 = 0 . 965 ), improving over an LSTM baseline by 35.5% (MSE), 13.6% (RMSE), and 14.5% (MAE). The Seq2Seq model is integrated in DRL environment to emulate soil moisture dynamics, and Proximal Policy Optimization (PPO) agents are trained under multiple reward designs explicitly aligned with farmer-oriented, short-term operational goals to assess how reward design influences irrigation behaviors. In this study, minimizing water consumption and the difference between soil moisture level and field capacity on every irrigation, provided the most balanced result. It reduces total water usage by 42.4% compared to the baseline reward design while minimizing the risk of under-irrigation. The study also underscores that reward structures have a significant influence on irrigation behavior which highlights the need for a more comprehensive evaluation strategy for DRL agents to ensure both water efficiency and crop health.
Zitieren
APA
Mukti, A. J. T. W., Ertal, C., & Delibasi, A. (2026). Optimizing irrigation decisions with Seq2Seq modeling and deep reinforcement learning. Computers and Electronics in Agriculture, 245, 111448. https://doi.org/10.1016/j.compag.2026.111448
Vancouver
Mukti AJTW, Ertal C, Delibasi A. Optimizing irrigation decisions with Seq2Seq modeling and deep reinforcement learning. Computers and Electronics in Agriculture. 2026;245:111448. doi:10.1016/j.compag.2026.111448
BibTeX
@article{Mukti2026,
author = {Mukti, Adityo Julian Tri Wibowo and Ertal, Cem and Delibasi, Akin},
title = {Optimizing irrigation decisions with Seq2Seq modeling and deep reinforcement learning},
journal = {Computers and Electronics in Agriculture},
year = {2026},
volume = {245},
pages = {111448},
doi = {10.1016/j.compag.2026.111448}
}
EUVIMED