Author: Wei Wu (National Cheng Kung University) - A physics-informed neural network (PINN) for modeling the gasification process of Municipal Solid Waste (MSW) is addressed which is shown in Fig. 1. Regarding the PINN, the physics knowledge integrated into neural network aims to enhance the prediction accuracy where a physics loss based on thermodynamic monotonic trends are taken into account. Through Aspen Plus platform with automated Python for real-time data updates, the predictions of outlet compositions such as temperature, moisture content, equivalence ratio, and syngas compositions are successfully validated according to experimental data. To optimize the PINN, hyperparameters are well-tuned by Bayesian optimization. Finally, the proposed hybrid data-driven PINN has been benchmarked against pure data-based ML methods such as Support Vector Machine (SVM) and Extreme Gradient Boosting (XGB) due to larger Coefficient of Determination (R2) and smaller root-mean-square error (RMSE) values.
