Development of data-driven models for fluidized-bed coal gasification process

Abstract

Coal gasification is one of the most viable and practical clean coal technologies for power generation. In this technology, the low calorific value fuel (LCVF) gas containing mainly carbon monoxide (CO), hydrogen (H₂) and carbon dioxide (CO₂) are obtained by blowing a mixture of air and steam through a bed of coal at atmospheric or elevated pressures. In this paper, coal gasification data from 18 fluidized bed gasifiers (FBGs) operating at a steady-state and located in India and other countries have been used to develop two types of exclusively data-driven steady-state FBG models. Specifically, multivariate regression (MVR) and multilayer Artificial Neural Network (ANN) strategies have been employed for the prediction of gas produced rate and heating value of the product gas. Both types of models use six inputs describing coal properties and gasification process parameters namely fixed carbon, volatile matter, mineral matter, air feed per kg of coal, steam feed per kg of coal and temperature. The output prediction accuracy results of the developed models indicate that the ANN-based models outperform the corresponding MVR models. The models presented here can be gainfully employed for evaluating steady-state performance of an FBG as well as the thermal potential of a coal prior to its usage for the generation of the LCVF gas. These models would also assist in the design of a gasifier for coals with varying ash content.