Dynamic Viscosity of Polyethylene Glycol (<scp>PEG</scp>): Data Assessment, Sensitivity Analysis and Robust Modeling via Artificial Intelligence Methods
Annotatsiya
ABSTRACT Accurate determination of physio‐chemical properties of polyethylene glycol as a green non‐toxic solvent is vital for any chemical and energy process involving polyethylene glycol. Machine learning algorithms are known as robust methods for prediction purposes. This study put forward least squares support vector machine scheme optimized with either particle swarm optimization, genetic algorithm and coupled simulated annealing optimization methods to precisely carry out the prediction task of polyethylene glycol viscosity in terms of its influencing parameters of pressure, temperature and polyethylene glycol molecular weight using a dataset containing experimental pertinent values. An outlier detection algorithm is made use of to confirm the data reliability for model development. In addition, sensitivity study is done to ascertain the relative impacts of each input factor on polyethylene glycol viscosity. The results showed that least squares support vector machine optimized with coupled simulated annealing is the most accurate model for the PEG viscosity prediction task with coefficient of determination of 0.997, average absolute relative error percent of 0.653 and mean square error of 8.005 for all datapoints. In addition, it was found that unlike temperature, we observe an indirect correlation of pressure and polyethylene glycol molecular weight with the pertinent viscosity data. The robustness of the developed least squares support vector machine optimized with coupled simulated annealing is further approved as it outperforms a renowned correlation that is often used for general viscosity prediction.