Answers are grounded in this paper and cite their sources.
Anomaly detection and fault diagnosis in intelligent irrigation systems are critical for ensuring food security and sustainable water resource management, particularly for water-intensive crops such as rice. However, existing methods exhibit limited adaptability to dynamic agricultural environments and insufficient interpretability for maintenance decision-making. This study proposes an interpretable dual-channel XGBoost-attention model that integrates ensemble learning with attention mechanisms to address these challenges. The model was validated using 164413 irrigation records from operational rice fields, demonstrating superior performance with 98.67% accuracy, 98.52% precision in binary classification tasks. Under extreme class imbalance conditions, the proposed model maintained robust anomaly detection capabilities, achieving F1-scores ranging from 86.67% to 97.73% across imbalance ratios from 1:1 to 1:10, outperforming traditional machine learning models and deep learning architectures. In multi-class fault diagnosis, the model achieved area under the receiver operating characteristic curve (ROC-AUC) values of 0.9311, 0.9280, and 0.9500 for the three most prevalent anomaly types, representing 14.8%–18.0% improvements over baseline XGBoost. SHAP analysis revealed that temporal features and execution delay metrics were the primary determinants of anomaly classification, with generalized additive model (GAM)-smoothed response curves establishing quantitative thresholds for automated maintenance protocols. The deployed web-based diagnostic system enables real-time fault identification with actionable maintenance recommendations, offering a practical solution for precision irrigation management in resource-constrained agricultural environments.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: