Research on Green Water Demand Prediction and Intelligent Irrigation Based on Machine Learning

Abstract

China’s agriculture is currently in a critical phase of transitioning from traditional irrigation to intelligent, efficient, and green agriculture. Faced with the dual pressures of water resource scarcity and food security, developing smart irrigation is an important direction for promoting sustainable agricultural development. This study takes the major wheat-producing areas in Henan Province as the research object and constructs an intelligent irrigation research framework encompassing water demand prediction, feature extraction, multimodel comparison, and multi-objective optimization. First, a sliding window method is employed to construct the prediction dataset, and random forest is used to screen out 10 key meteorological variables including sunshine duration and precipitation. The water demand prediction performance of Random Forest (RF), Convolutional Neural Network (CNN), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM) networks is compared, with model parameters optimized using the Sparrow Search Algorithm (SSA). Results show that the SSA-RF model achieves the best prediction performance. The SHAP method is applied to analyze feature contributions, revealing that precipitation is the main influencing factor of water demand. Predictions indicate higher water demand during the seedling stage and regreening stage of wheat. Furthermore, the NSGA-II multi-objective optimization algorithm is adopted, comprehensively considering irrigation cost, evaporation loss, and precipitation. It determines that sprinkler irrigation is optimal for the regreening, jointing, and grain-filling stages of wheat, while UAV irrigation is optimal for the seedling stage. Three scenarios conventional development, water-saving optimization, and green high-efficiency are set up, and differentiated irrigation water use strategies are proposed.