Prediction of Slag Viscosity Based on Machine Learning for Molten Gasification of Hazardous Wastes

Abstract

Experimental data from viscosity measurements of 124 glassy slags were used to drive and develop machine learning models that could be used for direct or indirect viscosity prediction. Samples were categorized according to the content of chemical components or general competitive neural network. The direct viscosity prediction using artificial neural network models of different kinds of slag samples was established. The prediction average error and maximum absolute error in the corresponding models were significantly smaller than the artificial neural network without categorizing the samples. Moreover, the viscosity curve for each glassy slag was fitted by a general formula, and the corresponding parameters were obtained. The principal component analysis (PCA)–particle swarm optimization (PSO)–back propagation (BP) neural network models for predicting parameters were proposed. This indirect approach was considered to successfully overcome the limitations of temperature and viscosity ranges in direct prediction while delivering smooth viscosity curves.