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

This study evaluates the hydraulic performance of irrigation systems using recycled porous pipes and the predictive modeling capabilities of deep learning algorithms for discharge rates, comparing Type A (recycled rubber-polyethylene blend) and Type B (recycled rubber only). Laboratory experiments measured discharge rates, coefficient of variation (CV), and emission uniformity (EU) across pressures (20-80 kPa) and pipe lengths (3-9 m). Results showed strong discharge-pressure correlations (R² = 0.95-0.97). Type B achieved superior performance at 80 kPa, with lower CV (8.80%) and higher EU (87.25%) versus Type A (CV = 9.54%, EU = 84.60%), indicating enhanced flow efficiency. Statistical analysis confirmed significant differences (p < 0.05) between pipe types. Four deep learning models-Enhanced Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), Deep Neural Network (DNN), and Artificial Neural Network (ANN)-were developed to predict discharge rates based on pressure, pipe length, and material type. Synthetic data augmentation (GANs) was used to overcome limited experimental samples.The Enhanced MLP mode achieved the highest predictive accuracy (R² = 0.9891, RMSE = 0.2762), outperforming all other models.This integration of hydraulic evaluation and AI modeling supports real-time irrigation scheduling, enhances water efficiency in water-scarce regions, and highlights the critical influence of material choice.