(173z) Automated Kinetic Rate Equation Discovery – a Methodological Framework

Exponential population growth and global warming push for the optimal design of novel catalysts. However, catalyst’s compositional and structural effects on reaction kinetic parameters and rates are not well understood. No automated framework has been developed to determine the effects of a catalyst’s composition and structure on the rate of a chemical reaction. Nevertheless, before tackling this challenge, there is a critical problem that must be solved first: the automated discovery of true kinetic rate models.

In our work, we propose a novel methodological framework for the automated discovery of true kinetic rate equations. Our methodology is composed of four major steps. Firstly, symbolic regression is used, with limited but important prior knowledge, to fit the kinetic data available. The function that fits the data is selected via a rigorous model selection method paired with model-based design of experiments (MBDoE). Secondly, the derivatives of the selected function are computed analytically or numerically. Thirdly, using symbolic regression again, the computed derivatives are fitted. The symbolic regression algorithm is once again guided by minimal prior knowledge, and the function is rigorously identified. Lastly, the final model is numerically integrated to compare the results with the kinetic data. If the model does not provide a satisfactory fit, design of experiments is used to optimally generate more data, and the methodology is repeated.

Our methodological framework was benchmarked against simulated kinetic case studies, outperforming artificial neural networks and state of the art hybrid models based on Gaussian Processes, whilst recovering the system’s true underlying dynamics. Our method also managed to propose a new kinetic rate equation that, compared to other equations found in the literature, more accurately describes the hydrogenation of acetylene over a supported palladium catalyst. In summary, our proposed methodological framework translates kinetic data to an accurate, predictive, and interpretable dynamic model.