High-precision multi-target prediction and interpretability analysis of biomass gasification via ensemble machine learning.

Machine learning exhibits notable advantages in simulating complex thermochemical processes such as gasification, with its robust nonlinear fitting capabilities effectively capturing the coupling relationships among multiple variables. In this study, missing values are addressed through model imputation, combined with categorical variable encoding and rigorous standardization, thereby significantly enhancing data quality. On this foundation, predictions of biomass gasification product composition, yield, and process efficiency were performed using six mainstream machine learning models under four hyperparameter optimization strategies, and the performance of these models was evaluated. Following optimization, the coefficient of determination (R2) for most models exceeded 0.90. Additionally, various ensemble methods are compared to integrate the strengths of individual models, aiming to achieve more stable and superior overall performance in multi-target prediction tasks. The predictive accuracy of machine learning in gasification applications is further improved by ensemble models, with the Weighted Averaging ensemble attaining an average test R2 of 0.9251, markedly outperforming single-model approaches. High-precision prediction methods, coupled with global and local interpretability techniques, provide more reliable and transparent decision support for policymakers and investors in the selection and design of gasification technologies and processes.