Process Simulation and Reverse Optimization of Modified Biochar Adsorption of Salt and Minerals in Water: Prediction Model of Adsorption Efficiency Based on BP Neural Network

<title>Abstract</title> <p>Removing salts and mineral ions from high‑salinity surface water, groundwater, and unconventional resources is a key challenge. Modified biochar offers adjustable structure and low cost, yet studies rarely address synergistic adsorption of multiple ions or provide a complete chain from modification to performance prediction and reverse optimization. Here, corn straw biochar was prepared unmodified (BC), sulfuric acid‑modified (SBC), and iron‑modified (FeBC). Batch adsorption experiments used Heze farmland drainage to link material structure with adsorption performance. A BP neural network model was built with adsorbent type, dosage, pH, temperature, and reaction time as inputs and adsorption efficiency as output. Hidden layer neurons were optimized to improve accuracy. Particle swarm optimization (PSO) enabled reverse parameter optimization. FeBC gave the best performance due to high specific surface area (315.2 m²/g), oxygen‑containing functional groups (3.18 mmol/g), and iron loading (15.6 mg/g). The BP model achieved R² of 0.935 (training) and 0.930 (validation), with prediction relative error below 1.2%. Reverse optimization identified optimal FeBC conditions: dosage 0.8 g/L, pH 7.5, 35°C, 180 min, yielding actual adsorption efficiency of 91.8% and a cost‑benefit ratio of 57.4%/yuan. This work provides both an efficient adsorbent and a data‑driven optimization method for water salinity removal, with demonstrated engineering feasibility and economic value.</p>