An artificial intelligence based approach to predicting syngas composition for downdraft biomass gasification

Highlights: • AI methods estimated syngas composition for downdraft biomass gasification. • Random Forests and least-squares support vector machine models are developed. • First study predicting syngas composition using machine learning classification. • The models have been tested with 5237 data using 10-fold cross-validation. • Temperature distribution significantly affects syngas composition estimation. Artificial neural networks and artificial intelligence based regression techniques have been recently applied to various gasification processes. Although these techniques obtain relatively satisfactory results for predicting gasification products, most of the proposed models are prone to low number of samples in the training data sets, which also lead to overfitting problem. Furthermore, these models may fall into local minima since cross-validation has never been used for predicting gasification products. In this paper, we consider prediction of gasification products as a classification problem by using machine learning classifiers. Two types of classifiers have been proposed, i.e., binary least squares support vector machine and multi-class random forests classifiers, for predicting producer gas composition and its calorific value obtained by woody biomass gasification process in a downdraft gasifier. The proposed approaches have been developed and tested with 5237 data samples using 10-fold cross-validation, where binary and multi-class classifiers achieved over 96% and 89% prediction accuracy values, respectively.