Integrating machine learning for multi-objective optimization of biomass conversion to hydrogen

The biomass-to-H₂ (BTH) process is considered an important source of green H₂, but the involvement of various biomass species and complex operating parameters poses significant challenges in practical operation and optimization. Herein, Aspen Plus was employed for data augmentation along with machine learning (ML) models to establish a hybrid ML model, which achieved an average R² greater than 0.999 and an average RMSE of 0.322 in predicting BTH outputs. Furthermore, integrating the hybrid ML model with the economic-environmental evaluation program enabled multi-objective optimization of the BTH process. Results revealed that the lowest cost achieved was 1.13 USD/kgH₂ with corresponding carbon emissions of 4.12–4.63 kgCO₂e/kgH₂. However, there was a tradeoff between cost and carbon emissions. By controlling the H₂ yield within the G3 range (200–300 kg/h), a low cost of 1.32 USD/kgH₂ and low carbon emissions of −0.23 kgCO₂e/kgH₂ were simultaneously achieved. Overall, this work proposed a new strategy for multi-objective optimization in H₂ production, coupling the hybrid ML model-driven accuracy prediction with an interactive platform.