Machine Learning-Driven Synthesis of Urea-Formaldehyde Fertilizers for Precision Agriculture: Tailoring Nitrogen Release via ANN-GA Optimization

Abstract

Urea-formaldehyde (UF), a prominent slow-release nitrogen fertilizer, faces challenges in production optimization to efficiently meet the varying slow-release needs of different crops. This study employed response surface methodology (RSM) analysis combined with an artificial neural network-genetic algorithm (ANN-GA) prediction to refine the UF polymerization process. Key factors influencing the polymerization process and the slow-release properties of UF products were identified as the urea/formaldehyde molar ratio (U/F) and reaction pH. The ANN-GA model demonstrated superior prediction accuracy over the RSM model, achieving coefficient of determination (R²) values of 0.9968 for cold water-insoluble substances (CWI) and 0.9979 for hot water-insoluble substances (HWI), representing improvements of 0.6% and 0.43%, respectively. By utilizing a fitness function that incorporated the UF activity index as the objective, the model optimized process parameter combinations, yielding relative errors below 4% between predicted and experimental values. The ANN-GA model facilitated precise control over UF polymerization, enabling the synthesis of short-cycle slow-release UF derived from methylenediurea (MDU) for rapid nutrient delivery and long-cycle UF based on trimethylenetraurea (TMTU) for sustained nutrient release. This study introduces a novel framework for regulating fertilizer manufacturing processes in precision agriculture, employing a “demand-driven → algorithmic optimization → targeted synthesis” approach that provides quick and adaptive solutions.