A systematic approach based on artificial intelligence techniques for simulating the ammonia removal by eighteen deep eutectic solvents

Abstract

The current research attempts to model the ammonia removal capacity of deep eutectic solvents (DES) employing artificial intelligence (AI) techniques. Four well-known AI classes apply to estimate the ammonia dissolution amount in eighteen DESs. The DES chemistry (type, molar concentration of components, and water dose), equilibrium temperature/pressure, and DES-ammonia absorption enthalpy help AI models estimate the ammonia solubility in DESs. The combination of trial-and-error and statistical analysis determines the best structure of AI models first. Then, several well-established model selection strategies apply to find the highest-accurate AI class. It was found that the adaptive neuro-fuzzy inference system (ANFIS) is more reliable than the other tested AI models. The designed ANFIS model precisely estimates 1356 experimental samples of ammonia removal by DESs with the mean absolute relative deviation percent of 3.51, mean absolute deviation percent of 0.121, root mean squared error of 0.216, and correlation of determination of 0.9980. The selected AI model monitors the influence of DES components, molar dose, equilibrium pressure, and equilibrium pressure on ammonia elimination by these green solvents. Both trend and relevancy investigations approved that the ammonia absorption tendency of DESs intensifies by increasing the equilibrium pressure, absorption enthalpy, the number of moles of hydrogen bond acceptor (HBA), and molecular weight of hydrogen bound donor (HBD). In addition, increasing the equilibrium temperature, water dosage in DES structure, HBA molecular weight, and the number of HBD moles decreases ammonia removal by DESs.