Comparison of Neural Network Architectures in the Thermal Explosion Approximation Problem

ICLR 2026 Conference Submission17115 Authors

Keywords: Neural networks, Neural network architecture, Hydrogen detonation, Thermal explosion, Deep learning in chemical kinetics

Abstract:

This study investigates the effect of neural network architecture on the accuracy of data-driven modeling of thermal explosions in a hydrogen–oxygen–air mixture. Using a reduced kinetic mechanism for 11 reagents, the thermal explosion process is simulated under specified initial pressure and temperature conditions, generating time-resolved data. We compare three architectures: a standard multilayer perceptron (MLP), a DeepONet–inspired model, and our U-Net–style residual network, evaluating their ability to capture transient dynamics and key reaction regimes.

Our results demonstrate that network architecture has a decisive impact on predictive performance. The U-Net architecture consistently outperformed the other models, achieving a mean squared error (MSE) of 0.0013 with a standard deviation (STD) of 0.0218, demonstrating high fidelity in capturing both rapid transients and slower reaction dynamics. In contrast, the DeepONet-inspired model and the MLP achieved MSEs of 0.0181 (STD 0.0581) and 0.0202 (STD 0.0682), respectively, indicating reduced accuracy and greater variability in predictions. The large spread in error is due to the fact that neural networks are not always able to accurately approximate the various modes of the combustion process. Despite testing various architectures and using a fairly large dataset, the problem remains unresolved.

These findings highlight the importance of selecting appropriate network architectures for combining deep learning with chemically detailed kinetic simulations. Such careful selection paves the way for more reliable and interpretable predictive models in combustion and reactive-flow applications.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 17115