Chapter 2: Data-driven Model Construction

The amount of data generated by modern reaction engineering systems has increased exponentially. A natural progression is to apply the data to gain information about the process, be it for scale-up, control, or optimisation. Data-driven models enable this acquisition of knowledge by transforming data into predictions. These predictions are increasingly important as systems become complex beyond human comprehension and logical reasoning fails. Data-driven models come in two main forms: parametric models and non-parametric models. Parametric models contain parameters similar to kinetic models. While a modeller carefully considers the placement of each parameter and its meaning in their kinetic model, a parametric data-driven model often bears no relation to the underlying system. In deliberately defining such a general parameterisation, the model gives data a chance to ‘speak for itself’ without the inductive bias of a human modeller. Non-parametric models directly apply the data without using parameters to make predictions. In this chapter, key parametric and non-parametric data-driven models for reaction engineering will be introduced. Examples will be given of many popular use cases, and the benefits of each method will be described.