Modeling the impact of some independent parameters on the syngas characteristics during plasma gasification of municipal solid waste using artificial neural network and stepwise linear regression methods

Thermal plasma gasification is considered as an attractive technology to produce high quality syngas from municipal solid waste (MSW). It is imperative to study the effect of operating parameters on syngas quality and find a practical way to predict syngas characteristics. This paper compiled 112 research cases to develop quantitative models for 8 kinds of syngas characteristics and explored the simultaneous effects of input parameters during the plasma gasification by applying stepwise linear regression (SLR) and artificial neural network (ANN) methods. The SLR model has a higher predictive accuracy than the ANN model for gas yield, volume fraction of CH4 and CO2, as well as mechanical gasification efficiency (MGE), with R-testing(2) = 0.659-0.916. The ANN model demonstrates better performance than the SLR model for low heating value (LHV), dry gas ratio, volume fraction of H-2 and CO, with R-testing(2) = 0.807-0.939. According to the importance analysis, flow rates of the work gas-N-2, feedstock type, flow rates of the work gas-steam, and input power are the most critical parameters for LHV, gas yield, and volume fraction of CH4 and H-2, respectively. Input power and specific energy requirements (SER) are the most important factors affecting volume fractions of H-2 (25.7-57.3 vol%) and input power plays a dominant role. The models developed in this study could enhance understanding of plasma gasification and are unique to considering multiple input parameters together.

Automated summary

Thermal plasma gasification is an attractive technology for producing high quality syngas from MSW. This study developed quantitative models for syngas characteristics and explored the effects of input parameters using stepwise linear regression and artificial neural network methods. The results showed that the SLR model performed better for gas yield, volume fractions of CH4 and CO2, and mechanical gasification efficiency, while the ANN model had better performance for LHV, dry gas ratio, and volume fractions of H-2 and CO.