Neural tanks-in-series: a physics-guided neural network extension of the tanks-in-series model for enhanced flow reactor and reaction modelling

Abstract

This paper introduces the neural tanks-in-series (NTiS) model, an extension of the traditional tanks-in-series (TiS) model using physics-guided neural networks (PGNNs). The NTiS model integrates physical principles with data-driven approaches to improve the accuracy and reliability of flow reactor modeling. The NTiS can optimize physical parameters and learn unmodeled dynamics while ensuring physically feasible predictions, even for out-of-domain predictions. The approach is validated using simulations and experimental data from a Paal–Knorr pyrrole reaction, demonstrating its capability to model flow reactor systems under varying conditions. The NTiS framework offers a new, robust, and flexible tool for advancing chemical flow reactor modeling.