Sustainable synthesis of activated carbon from water hyacinth via one-step carbonization-H3PO4 activation: Optimization and modeling via RSM-artificial intelligence approach

This study explores the green synthesis of activated carbon (AC) from water hyacinth (WH), an invasive aquatic weed, using a one-step H₃PO₄ activation method, addressing both environmental waste management and sustainable material production. The present work aimed to optimize AC yield and surface area by evaluating the effects of impregnation ratio (0.4–1.0), temperature (600–800°C), and activation time (60–120 mins) via Response Surface Methodology (RSM) and advanced Artificial Intelligence (AI) techniques, including Artificial Neural Networks (ANN), ANN-Particle Swarm Optimization (PSO-ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS), and XGBoost. Characterization via XRD, FTIR, SEM, and BET confirmed the AC’s amorphous structure, functional groups, and porous morphology. RSM revealed that a two-factor interaction (2FI) model best described yield (R² = 0.9844), while a quadratic model fit surface area (R² = 0.9992). AC with a high surface area of 1025 m²/g and a yield of 44.5 % was achieved under optimum conditions (impregnation ratio of 0.785, temperature of 734°C, and time of 72 mins). The overall yield and surface area prediction accuracy followed the PSO-ANN>ANN>ANFIS>XGBoost order. PSO-ANN outperformed other models (R² = 0.9990 for surface area and 0.9767 for yield). The surface area was predicted more accurately than the yield response due to its less complex relationship with inputs. Based on cost constraints and performance requirements, the trade-off between yield and surface area is fundamental in AC production. This study offers a paradigm shift of invasive WH management, using it as a sustainable source of AC for industrial applications. • Invasive water hyacinth proved viable as activated carbon (AC) precursor. • One-step H 2 PO 4 activation produced AC: yield (44.5 %) and surface area (1025 m²/g). • Yield (56 %) and surface area (72 %) greatly depend on impregnation ratio. • Prediction accuracy followed PSO-ANN>ANN>ANFIS>XGBoost order. • Surface area was predicted more accurately by all models than yield response.