Interpretable machine-learning model with a collaborative game approach to predict yields and higher heating value of torrefied biomass

Abstract

Torrefaction is a treatment process for converting biomass to high-quality solid fuels. The investigation and interpretation of this process on highly dimensional, non-linear relationships as large datasets are limited. In this work, machine learning (ML) in combination with collaborative game theory (Shapley additive explanation, SHAP) was applied to develop an interpretable model in predicting solid yields (SY) and higher heating values (HHV) of solid products from biomass torrefaction using 18 independent input features from operating conditions, feedstock characteristics and torrefaction reactor properties. Three novel ML algorithms were evaluated, based on 10-fold cross-validation, with 5 different sets of input features. A gradient tree boosting (GTB) model was found to have the highest prediction accuracy R2 of 0.93 with root mean square error (RMSE) of 0.06 for SY while about 0.91 R2 with 0.79 RMSE for HHV. With the powerful SHAP algorithm, a new framework was proposed to interpret/explain the GTB model performance and highlight the highly influential features for the system of biomass torrefaction in both local and global points of view. Interactions for any pair of the features on the GTB model can be achieved. This application of ML with SHAP is a useful tool for researchers on biomass conversion.