(365g) Development of Artificial Intelligence Based Models for Biomass Gasification

As a renewable and naturally abundant fuel, biomass has a huge potential for energy generation and subsequent replacement of fossil fuels. A common and very efficient mode of energy generation using biomass is via its gasification. However, designing the gasifier is a tedious task. The characteristics of the feedstock and gasifier conditions have the most influence on the performance of the gasification. Monitoring the performance of the gasifier can be tedious, costly, and time-consuming. The performance of the gasifier is very sensitive to the characteristics and type of the biomass feedstock used. Certain types of gasifiers are exceedingly difficult to scale up due to sensitivity of gasifier performance to operating conditions and process parameters. Therefore, identification of optimal operating conditions for a specific gasifier design employing a certain biomass feedstock can be long-drown-out and expensive. The use of simulation and modeling can save resources required to design the gasifier. Computational fluid dynamics (CFD) can be used to model the gasifier operation. Formulation of and execution of CFD models can be extraordinarily complex due to multitude of homogenous and heterogeneous reactions occurring in a gasifier, with multiple uncontrollable undesired side reactions. Obtaining solutions of these models is very time consuming and requires powerful processors, especially for three dimensional CFD models. The solid particles in the CFD models can only be simulated as perfect spheres, which is a significant limitation since biomass is rarely in perfect spherical shape. In this work, biomass gasification is modeled using four artificial intelligence (AI) based models to find the syngas compositions, and the performance of these models is compared using appropriate evaluation methods. The four models are based on Decision Tree Regression (DTR), Random Forest Regression (RFR), Support-Vector Regression (SVR), and Artificial Neural Network (ANN). The characteristics of biomass (ultimate and proximate analysis) and gasifier conditions (such as equivalence ratio and temperature) are inputs to the models and the outputs of the models include the gasification effluent composition, the gasifier effluent being typically comprised of hydrogen, carbon monoxide, carbon dioxide, methane, and other low carbon number hydrocarbons. Large databases from gasification experiments reported in the literature were used in the development of the four AI models. The databases cover wide ranges of biomass feedstocks and gasifier operating conditions. The four AI models represent the gasifier operations with good accuracy and have reliable predictive ability with respect to data not used in model development. For example, the ANN model has an overall mean square error (MSE) of 0.87. The AI based models will be useful aids in economical design of biomass gasifiers.