A Physics-Informed Neural Network Framework Integrating Soft and Hard Constraints for Predicting Biomass Gasification...

Machine learning methods have demonstrated promising applications in biomass gasification modeling. However, conventional machine learning models primarily rely on experimental data and do not account for the reaction mechanisms of gasification. When data samples are insufficient, the correlations learned by the model can deviate substantially from mechanistic laws. In this study, a method for predicting biomass gasification product distribution based on physics-informed neural networks (PINNs) was proposed to apply on a biomass gasification problem with a small data size. This method seamlessly integrates real experimental data with prior mechanistic knowledge by embedding boundary constraints and monotonic relationships among key variables into an artificial neural network (ANN). Hard constraints were enforced via a normalized output layer, while soft constraints were applied as a monotonicity penalty in the loss function. Results show that the proposed PINN model achieves a coefficient of determination (R ²) greater than 0.89 and a root-mean-square error (RMSE) less than 4%. The overall predictive accuracy is superior to that of other three purely data-fitting machine-learning modelsrandom forest (RF), support vector machine (SVM), and ANN. Furthermore, the PINN model strictly adheres to boundary constraints and prior mechanistic monotonic relationships, exhibiting better interpretability and generalization capabilities.