Enhanced deep learning model for predicting hydraulic performance in recycled porous pipe irrigation systems

Area of Science:

• Agricultural Engineering
• Materials Science
• Artificial Intelligence

Background:

• Irrigation systems are crucial for agriculture, but water scarcity necessitates efficient water use.
• Recycled materials offer a sustainable alternative for irrigation components.
• Accurate prediction of discharge rates is vital for optimizing irrigation scheduling.

Purpose of the Study:

• To evaluate the hydraulic performance of two types of recycled porous irrigation pipes (Type A and Type B).
• To develop and compare deep learning models for predicting discharge rates in these systems.
• To assess the potential of AI in enhancing irrigation efficiency using recycled materials.

Main Methods:

• Laboratory experiments measured discharge rates, coefficient of variation (CV), and emission uniformity (EU) under varying pressures and pipe lengths.
• Four deep learning models (MLP, LSTM, DNN, ANN) were trained to predict discharge rates.
• Generative Adversarial Networks (GANs) were used for synthetic data augmentation.

Main Results:

• Type B recycled pipes demonstrated significantly better hydraulic performance (lower CV, higher EU) than Type A at 80 kPa.
• Strong correlations (R²=0.95-0.97) were observed between discharge and pressure.
• The Enhanced Multilayer Perceptron (MLP) deep learning model achieved the highest prediction accuracy (R²=0.9891).

Conclusions:

• Recycled porous pipes, particularly Type B, can effectively improve irrigation system performance and water efficiency.
• Deep learning models, especially Enhanced MLP, show high potential for accurate discharge rate prediction in irrigation systems.
• Integrating hydraulic evaluation with AI modeling supports sustainable water management in water-scarce environments.