Applying Support Vector Regression-Based Hybrid Models for Modeling the Gasification Process

Abstract

Modeling the gasification process through machine learning (ML) involves predicting the behavior and performance of gasification systems. Support Vector Regression (SVR) is known as an effective procedure in forecasting continuous variables and is even suitable for the modeling gasification process. Employing optimization algorithms to connect and fine-tune internal model settings can lead to the creation of various hybrid and ensemble models. Generally, employing hybrid models has demonstrated enhanced performance while utilizing cost-effective modeling techniques. In this study, SVR was utilized as a machine learning method, alongside the Crystal Structure Algorithm (CryStAl) and the Population-based Vortex Search Algorithm (PVSA), to fine-tune SVR for accurately assessing CO and CO2 values. After evaluating the outcomes of the proposed models, it was observed that the SVR-PVSA hybrid model outperformed the SVR-CryStAl model, with differences of 1%, 19%, and 57% based on R², RMSE, and MAE indices, respectively for CO and that of 1%, 14%, and 54% for CO2 in terms of R², RMSE, and MAE evaluators, respectively. Furthermore, for predicting both CO and CO2, the SVR-CryStAl hybrid model yielded the highest value, demonstrating superior performance compared to the SVR-PVSA model, with an average difference of 0.6% and 0.9% in terms of the VAF index.

Keywords

• gasification modeling
• machine learning
• support vector regression
• crystal structure algorithm
• population-based vortex search algorithm