Influence of Feed Processing Conditions on Pellet Stability, Nutrient Retention and Water Quality in Recirculating Aquaculture Systems

Abstract

The optimization of aquaculture feed processing conditions is essential for improving pellet integrity, minimizing nutrient losses, and maintaining water quality in recirculating aquaculture systems (RAS). Conventional feed processing approaches frequently optimize pellet durability without adequately accounting for nutrient leaching, nitrogenous waste accumulation, and the interactions between processing parameters and aquatic environmental conditions. This study proposes a novel Pellet Stability and Nutrient Retention Optimization Algorithm (PSNROA), integrating multi-objective optimization, machine learning-based predictive modelling, and water-quality constraints to identify suitable aquafeed processing conditions. The proposed framework evaluates extrusion temperature (80–140 °C), feed moisture content (15–35%), screw speed (100–400 rpm), conditioning time (30–180 s), and pellet diameter (2–6 mm), using a controlled experimental design linked to RAS water-quality monitoring. PSNROA combines an XGBoost-based surrogate prediction model with adaptive Pareto-based optimization to maximize pellet water stability, pellet durability index, protein retention, and lipid retention while minimizing nutrient leaching, total ammonia nitrogen, and turbidity. Its predictive and optimization performance is designed to be compared with Response Surface Methodology (RSM), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Non-dominated Sorting Genetic Algorithm II (NSGA-II). Comparative evaluation will employ coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), hypervolume, inverted generational distance, and computational time. Experimental outcomes will be illustrated using multiseries line graphs, grouped bar charts, heatmaps, and scatter plots to establish relationships between processing conditions, pellet properties, and water-quality responses. The proposed hypothesis is that PSNROA can identify processing configurations that deliver superior multi-objective performance compared with conventional optimization methods while satisfying predefined environmental and nutritional constraints. Statistical validation using independent experimental observations, repeated optimization runs, and sensitivity analysis will establish whether these improvements are significant and reproducible. The study is expected to provide an integrated decision-support framework for sustainable aquafeed manufacturing, improved nutrient utilization, reduced water-treatment demand, and enhanced operational efficiency in recirculating aquaculture systems.