Unveiling the potential of operating time in improving machine learning models’ performance for waste biomass gasification systems

Biomass is a promising renewable energy source, especially regarding local energy production. Gasification is an outstanding thermochemical process for syngas production from biomass. The optimal modeling and control of syngas production from biomass gasification is vital in planning small-scale energy supply. Artificial intelligence methods are widely applied to reveal the nonlinear relations between the gasification parameters and the reaction products. In this study, special attention is paid to the effect of operating time as a predictor on the performances of the models formed, as its introduction provides the phase information to the model which allows the combination of different sequential experimental data. Unlike most preceding studies in which hold-out is preferred, nested cross-validation is utilized for model evaluation emphasizing the unbiased nature of the latter method. Least squares, gaussian kernel, generalized additive models, random forest, and artificial neural networks (ANN) are utilized to model the process's product syngas composition and high heating value (HHV). The significance of operating time on the gasification system is also reflected on the feature importance analysis which is performed. Along with the operating time, temperature is also shown to play a vital role on the hydrogen yield and calorific value of the products.