A comprehensive artificial neural network model for gasification process prediction
Simon Ascher, William T. Sloan, Ian Watson, Siming You
Abstract
The viability and the relative merits of competing biomass and waste gasification schemes depends on a complex mix of interacting factors. Conventional analytical methods that are used to aid decision making rely on a plethora of poorly defined parameters. Here we develop a method that eschews the uncertainty in process representation by using a machine learning, data driven, approach to predicting a set of 10 key measures of gasification technology’s performance. We develop an artificial neural network that is novel in its use of both categorical and continuous data inputs, which makes it flexible and broadly applicable in assessing gasification process designs. It is the first model applicable to a wide range of feedstock types, gasifying agents, and reactor options. A strong predictive performance, quantified by a coefficient of determination (R2) of 0.9310, was confirmed. The approach has the potential to generate accurate input data for e.g., cost-benefit analysis (CBA) and life cycle sustainability assessment (LCSA) and thus allow for more transparency in the decisions made by policy makers and investors.