Biomass gasification modelling based on physics-informed neural network with hard constraints
, , ,
1Key Laboratory of Energy Thermal Conversion and Control of Ministry of Education, School of Energy and Environment, Southeast University, Nanjing, PR China
In recent years, predicting biomass gasification syngas components using machine learning (ML) methods has received significant attention and interest. However, insufficient experimental data can hardly guide ML models in capturing complete knowledge of the gasification process, which can be reflected explicitly in the mismatch between ML model predictions and physical knowledge. Therefore, this study proposes a novel hard-constrained physics-informed neural network (HC-PINN) to predict biomass gasification syngas components by integrating experimental data with prior monotonicity relationships. Firstly, the monotonicity constraints between syngas components and key operating parameters are derived as inequality constraints involving solely the network parameters. Since the inequality constraints bypass the presence of model input parameters, the HC-PINN model can ensure that the model predictions in the entire input space satisfy the monotonicity relationships while avoiding the hard-to-trade-off synthetic samples. Secondly, the constrained particle swarm optimization algorithm is used to optimize the network parameters between regression losses and inequality constraints. The additional physical loss terms introduced in the soft constraints are substituted by searching for the feasible regions that satisfy the inequality constraints, avoiding the complexity associated with weight trade-offs across multiple loss terms. Finally, the results show that the HC-PINN model can strictly adhere to the monotonicity relationships even in the no-training-sample region, reflecting better interpretability and generalization than the three state-of-the-art traditional ML methods.
Ajorloo, M. et al. (2022) ‘Recent advances in thermodynamic analysis of biomass gasification: A review on numerical modelling and simulation’, Journal of the Energy Institute. Elsevier Ltd, 102(January), pp. 395–419.
Aslam Khan, M. N. et al. (2024) ‘Prediction of hydrogen yield from supercritical gasification process of sewage sludge using machine learning and particle swarm hybrid strategy’, International Journal of Hydrogen Energy. Hydrogen Energy Publications LLC, 54, pp. 512–525.
Bradley, W. et al. (2022) ‘Perspectives on the integration between first-principles and data-driven modeling’, Computers and Chemical Engineering. Elsevier Ltd, 166(May), p. 107898.
Devasahayam, S. and Albijanic, B. (2024) ‘Predicting hydrogen production from co-gasification of biomass and plastics using tree based machine learning algorithms’, Renewable Energy. Elsevier Ltd, 222(March 2023), p. 119883.
Kim, J. Y. et al. (2023) ‘Predicting and optimizing syngas production from fluidized bed biomass gasifiers: A machine learning approach’, Energy. Elsevier Ltd, 263(PC), p. 125900.
Di Natale, L. et al. (2023) ‘Towards scalable physically consistent neural networks: An application to data-driven multi-zone thermal building models’, Applied Energy. Elsevier Ltd, 340(March), p. 121071.
Perez, R. E. and Behdinan, K. (2007) ‘Particle swarm approach for structural design optimization’, Computers and Structures, 85(19–20), pp. 1579–1588.
Ren, S. et al. (2024) ‘Disentangled Representation Aided Physics-Informed Neural Network for Predicting Syngas Compositions of Biomass Gasification’, Energy and Fuels, 38(3), pp. 2033–2045.
Ren, S., Wu, S. and Weng, Q. (2023) ‘Physics-informed machine learning methods for biomass gasification modeling by considering monotonic relationships’, Bioresource Technology. Elsevier Ltd, 369(November 2022), p. 128472.
Shin, H. and Choi, M. (2023) ‘Physics-informed variational inference for uncertainty quantification of stochastic differential equations’, Journal of Computational Physics. Elsevier Inc., 487, p. 112183.
Silva, I. P. et al. (2019) ‘Thermodynamic equilibrium model based on stoichiometric method for biomass gasification: A review of model modifications’, Renewable and Sustainable Energy Reviews. Elsevier Ltd, 114(August), p. 109305.
Umenweke, G. C. et al. (2022) ‘Machine learning methods for modeling conventional and hydrothermal gasification of waste biomass: A review’, Bioresource Technology Reports. Elsevier Ltd, 17(January), p. 100976.
Wang, Z. et al. (2022) ‘The role of machine learning to boost the bioenergy and biofuels conversion’, Bioresource Technology. Elsevier Ltd, 343(August 2021), p. 126099.
Xie, K. et al. (2021) ‘Physics-guided deep learning for rainfall-runoff modeling by considering extreme events and monotonic relationships’, Journal of Hydrology. Elsevier B.V., 603(PC), p. 127043.
How to cite this article
Weng, Q., Jun, R., Zhu, B. and Si, F. (2024). 'Biomass gasification modelling based on physics-informed neural network with hard constraints'. SET Proceedings, 21(3), pp. 199-208. https://doi.org/10.65582/set.2024.125