Application of artificial intelligence in modeling of nitrate removal process using zero-valent iron nanoparticles-loaded carboxymethyl cellulose.

This study explores nitrate reduction in aqueous solutions using carboxymethyl cellulose loaded with zero-valent iron nanoparticles (Fe-CMC). The structures of this nano-composite were characterized using various techniques. Based on the characterization results, the specific surface area of Fe-CMC measured by the Brunauer-Emmett-Teller analysis were 39.6 m/g. In addition, Scanning Electron Microscopy images displayed that spherical nano zero-valent iron particles (nZVI) with an average particle diameter of 80 nm are surrounded by carboxymethyl cellulose and no noticeable aggregates were detected. Batch experiments assessed Fe-CMC's effectiveness in nitrate removal under diverse conditions including different adsorbent dosages (C, 2-10 mg/L), contact time (t, 10-1440 min), initial pH (pH, 2-10), temperature (T, 10-55 °C), and initial concentration of nitrate (C, 10-500 mg/L). Results indicated decreased removal with higher initial pHi and C, while increased Cs and T enhanced removal. The study of nitrate removal mechanism by Fe-CMC revealed that the redox reaction between immobilized nZVI on the CMC surface and nitrate ions was responsible for nitrate removal, and the main product of this reaction was ammonium, which was subsequently completely removed by the synthesized nanocomposite. In addition, a stable deviation quantum particle swarm optimization algorithm (SD-QPSO) and a least square error method were employed to train the ANFIS parameters. To demonstrate model performance, a quadratic polynomial function was proposed to display the performance of the SD-QPSO algorithm in which the constant parameters were optimized through the SD-QPSO algorithm. Sensitivity analysis was conducted on the proposed quadratic polynomial function by adding a constant deviation and removing each input using two different strategies. According to the sensitivity analysis, the predicted removal efficiency was most sensitive to changes in pH, followed by C, T, C, and t. The obtained results underscore the potential of the ANFIS model (R = 0.99803, RMSE = 0.9888), and polynomial function (R = 0.998256, RMSE = 1.7532) as accurate and efficient alternatives to time-consuming laboratory measurements for assessing nitrate removal efficiency. These models can offer rapid insights and predictions regarding the impact of various factors on the process, saving both time and resources.